Europe’s Sovereignty Stack: CADA, Compute, and the Limits of Autarky

Executive Summary

This issue brief examines the European Commission’s proposed Cloud and AI Development Act (CADA). The Act pursues a legitimate goal: reducing Europe’s dependence on foreign cloud and artificial-intelligence infrastructure. The proposal also has important strengths. It is formally risk-based, focused on public-sector procurement, and paired with a permitting fast-track for data-center construction.

But CADA’s highest assurance levels risk confusing sovereignty with insulation from non-EU law, ownership, and supply chains. That insulation is not realistically available for the cloud and AI services Europe most needs. Advanced chips, frontier AI models, critical software, financing, and energy infrastructure remain deeply embedded in allied and global supply chains.

CADA’s top tiers should therefore require stronger security, not a narrower focus on nationality. Ownership and citizenship tests should give way to technical and organizational safeguards: customer-held encryption keys, zero-retention architectures, data-in-use protections, personnel screening and clearance, migration plans, anti-tamper controls, and multi-cloud portability.

The proposal should also distinguish adversaries from allies. The relevant question is not whether a provider is non-EU, but whether its home state is likely to weaponize it against Europe and whether the provider can meaningfully resist. Article 18 should be retargeted accordingly.

Finally, Europe cannot procure or certify its way to frontier capability. The durable path to sovereignty is to build leverage: more compute capacity located in Europe, governed by European law, and deployed at sufficient scale that dependence with allies becomes mutual. That requires grid-connection reform, deeper capital markets, a workable pan-EU corporate form, and better tax treatment for employee equity.

I.   Introduction

The European Commission has proposed the Cloud and AI Development Act (CADA), the legislative centerpiece of its European Technological Sovereignty Package, alongside a Chips Act 2.0. CADA seeks to create demand for EU-native cloud services, while Chips Act 2.0 aims to increase Europe’s supply of domestically produced semiconductors.

Together, these measures seek to reduce Europe’s dependence on foreign technology. That is a legitimate and important objective. But CADA’s highest assurance levels—its most demanding tiers for cloud services used in sensitive public-sector contexts—risk confusing sovereignty with insulation from the outside world. For the cloud and artificial-intelligence (AI) services that matter most, such insulation is not realistically available.

CADA begins from a risk-based premise. Its requirements apply formally to the public sector, and its Buy European procurement preference is ancillary by design. But the act also creates incentives to push more workloads toward the highest assurance levels. The Commission estimates that only about 10% of public-sector use cases would require those levels. In practice, that share may grow, and the pressure may spill over into private markets.

The stakes are significant because, at the highest levels, CADA treats sovereignty as freedom from third-country control, foreign legal compulsion, and operational dependence. In effect, it asks cloud providers to demonstrate that non-EU governments and non-EU supply chains cannot affect their services. No provider of major cloud or AI workloads can credibly make that promise.

The problem is not limited to where a provider is incorporated or who owns it. For many high-value workloads, the relevant risks turn on leverage: who controls essential chips, software, AI models, facilities, power, financing, and legal chokepoints. Even a fully EU-based company may face foreign legal process, as OVH did in Canada, or may be exposed to foreign mandates because it depends on foreign hardware, software, or other critical inputs.

Those dependencies are especially important for AI clouds—cloud infrastructure used to train, host, or run AI models. The EU package partly acknowledges the semiconductor problem, but its proposed solution would arrive only in the 2030s. It does far less to confront the gap between U.S.-based frontier AI laboratories and everyone else. Frontier AI models are the most capable systems at the leading edge of development. Today, the leading labs are mostly American, and they invest at a scale and speed that European competitors have not matched.

The EU strategy therefore appears to rest on a contested premise: that building more compute capacity in Europe will allow EU-developed AI models to reach the frontier. Compute capacity—the data centers, advanced chips, and electricity needed to train and run AI systems—is necessary. But it is not sufficient. The package offers little reason to believe Europe will match future frontier AI capabilities, rather than continue to lag behind a frontier that is moving quickly and backed by much larger investment elsewhere.

The responsible course is not technological autarky. The Commission itself acknowledges that Europe cannot isolate itself from global technology markets. Dependence is inevitable. The policy objective should be to make that dependence more secure, more diversified, and more mutual, especially with allied democracies.

The risks became concrete on June 12, when a U.S. export-control directive forced Anthropic, a leading American AI lab, to cut every non-U.S. user—including allied users—off from its two state-of-the-art models overnight.[1] No ownership rule, Level 4 certificate, or EU-citizen staffing requirement would have preserved that access. The decisive layer was not cloud procurement. It was control over the best AI models, a layer upstream of CADA’s main regulatory tools.

The lesson is not that Europe can build its own frontier models on any timeline relevant to current policy. It likely cannot. Europe’s more durable source of leverage lies in building the infrastructure on which frontier AI depends: data centers and power, located in Europe, governed by European law, and deployed at sufficient scale that foreign partners would incur real costs by cutting Europe off. In other words, Europe should seek to make technological dependence mutual rather than mistake ownership labels for sovereignty.

Country-of-origin rules do have a legitimate role, but the relevant distinction is between adversaries and allies, not between EU and non-EU providers as such. The key question is whether a provider’s home government is likely to weaponize that provider against Europe, and whether the provider would meaningfully resist.

That distinction matters. A European or American firm ordered to act against an ally may be expected to resist, litigate, appeal, and delay. A firm answerable to a nondemocratic adversary government is less likely to have those options. CADA should therefore sort providers along the ally/adversary line, not the EU/non-EU line.

Once a non-EU jurisdiction is recognized as an ally, CADA should focus on technical and organizational safeguards that reduce concrete risks to confidentiality and availability. These include encryption, customer-held keys, access controls, personnel screening, service-continuity planning, portability, and other measures that make data harder to access and services harder to disrupt. Those safeguards should apply equally to EU-native firms and allied foreign providers.

The package’s “build” agenda also remains incomplete. Its headline 12-month permitting clock excludes the steps that most often constrain data-center construction: environmental review and power-grid connection. The package also does little to address capital, corporate law, and taxation, even though those are among the largest obstacles to scaling European firms, particularly startups.

This issue brief develops those points in seven steps. Section II explains CADA’s risk-based structure and the upward pressure built into its assurance levels. Section III shows why the top tiers promise more immunity from foreign control than cloud and AI providers can deliver. Section IV argues that CADA should distinguish allies from adversaries rather than EU from non-EU providers. Section V explains why demand-side procurement preferences cannot substitute for real capacity. Section VI turns to the build agenda, including permitting, grid connection, capital markets, and compute infrastructure. Section VII offers concrete recommendations. The conclusion returns to the central claim: European sovereignty in cloud and AI will come less from excluding allied providers than from building capacity, hardening systems, and making dependence mutual.

II.   CADA’s Risk-Based Promise—and Its Sovereignty Trap

CADA begins from a sound premise: not every public-sector cloud use is equally sensitive, and not every risk requires an EU-only response. Section II.A explains that basic risk-based structure. Section II.B shows why that structure may be unstable in practice, as political incentives, incumbent pressure, and CADA’s own procedures may push more workloads toward the higher assurance levels. Section II.C examines the legal consequences of that shift, showing how the top tiers move from managing concrete risks to policing ownership, corporate control, software dependence, and personnel nationality. Section II.D then asks what an EU-only circle of trust can realistically achieve for the narrow set of government functions where such an approach may be coherent—and what costs it would impose.

A.             CADA Starts with a Sensible Risk-Based Frame

On June 3, the European Commission proposed the Cloud and AI Development Act (CADA), the legislative centerpiece of its European Technological Sovereignty Package. The package also includes Chips Act 2.0, an EU Open Source Strategy, and an energy-digitalization roadmap.[2] In a recent ICLE issue brief,[3] I argued that the European Union should approach AI-cloud sovereignty on a risk-based basis, rather than through categorical exclusions, and that a more ambitious conception of sovereignty—“immunity from non-EU law”—would impose costs on European AI users without delivering the immunity it promises.

CADA deserves credit for its basic structure. At least on paper, it is not a bloc-wide “European-only” rule. The regime is risk-based in principle, and it does not directly bind private commercial cloud purchases. As discussed below, however, that limitation may be less significant in practice than it first appears.

The act establishes “a Union cloud computing sovereignty framework comprising four Union assurance levels” under Article 16, with the substantive criteria set out in Annex II. These assurance levels are tiers of required safeguards: the higher the level, the more demanding the sovereignty, security, and control requirements. The levels are not supposed to apply uniformly. Member states and EU institutions must conduct periodic risk assessments under Article 29 to identify which public-sector activities affect “the preservation of public order” and determine which assurance level is appropriate.

The lowest tier is the default. Public bodies whose activities are not identified as relevant to public order “shall use cloud computing services that have been recognised … as having a Union assurance level 1” under Article 30(2). Higher tiers apply only to more sensitive domains: sectors covered by the EU’s Network and Information Security Directive, known as NIS2, as well as national and internal security, border management, defense, justice, and law enforcement. NIS2 imposes baseline cybersecurity duties on operators in critical sectors, including energy, transportation, banking, health, and digital infrastructure. For those sensitive activities, public bodies “shall only procure” services at assurance levels 2, 3, or 4 under Article 30(3).

Even there, CADA includes proportionality safeguards. Article 30(4) allows public bodies to avoid the higher-tier obligation where no adequate service exists or where the cost would be disproportionate. That is an important limitation. A sovereignty requirement that mandates unavailable or unreasonably costly services would not strengthen public-sector resilience; it would merely constrain procurement.

Recital 52 states the underlying logic: “Most public services would not require the highest levels of assurance.” Only “specific cases” may require levels 3 or 4, and the required risk assessment is meant to ensure proportionality and subsidiarity. Subsidiarity is the EU-law principle that the Union should act only where member states cannot adequately address the issue themselves.

The Commission describes the design as “risk-based” in Recital 62. That framing matters because it rejects, at least formally, the categorical “effective control” rule some had urged.[4] The Commission estimates that levels 3 and 4 will apply to only about 10% of public-sector use cases.[5]

The same restraint appears in CADA’s procurement preference. Article 32 requires contracting authorities to score a bidder’s contribution “to the development of a European cloud and AI ecosystem.” But those criteria must remain “ancillary and not decisive in the award of the contract” under Article 32(2)(d). Recital 67 suggests capping them at 15 out of 120 evaluation points, “subordinate to the core contract award criteria.” A bounded Buy European preference in public tenders can be a defensible industrial-policy instrument. It is materially different from a categorical exclusion.

The main difficulty arises higher up the assurance ladder.

B.             Risk-Based on Paper, Upward-Ratcheting in Practice

A risk-based design is only as sound as the risk assessments that implement it, and CADA’s incentives may push toward the higher tiers. European incumbents gain a more protected market when the highest assurance levels apply, giving them a strong incentive to argue that more workloads belong there. Cloud Infrastructure Services Providers in Europe (CISPE) has already described U.S.-linked “wrapper” arrangements—European-branded offerings that still rely on U.S. cloud providers—as “sovereignty washing.”[6] The practical implication is clear: if such structures count as sovereign, the protected market for EU-native providers becomes smaller.

CADA’s own procedures also tilt upward. The Commission may override a member state’s chosen assurance level if it considers that level inadequate under Article 29(5). A delegated-act power could turn today’s voluntary private-sector assessments into mandatory requirements for highly critical sectors under Article 31(3). A delegated act is a legally binding measure adopted by the Commission to supplement or amend nonessential elements of EU legislation. In addition, the Commission expressly anticipates that public-sector requirements may “be mirrored by private-sector entities … with subsequent spillover effects,” as Recital 66 puts it. That is how a formally public-sector rule can influence the private market.[7]

The political signal points in the same direction. Executive Vice President Henna Virkkunen, who leads the Commission’s tech-sovereignty portfolio, told reporters that the goal is to ensure no provider of critical workloads holds a “kill switch,” that “we want to make sure that our most critical sensitive data is stored in Europe,” and that U.S. companies would “struggle to reach the highest sovereignty tier.”[8]

Thus, “most public services stay at Level 1” is not a guarantee. With several incentives pushing in the other direction, the top tiers may become more important than the proportionality language suggests. Their actual requirements therefore warrant close scrutiny.

C.             The Top Tiers Turn Risk Management into Control Tests

At Level 1, CADA allows third-country control, subject to limited safeguards. At Level 2, a provider controlled from outside the European Union may still qualify, but only through an EU-localized service structure. The provider must also show that it has adopted the “necessary legal, technical and organisational measures” to prevent three risks: third-country access to customer data, service disruption or degradation, and coerced compliance with third-country sanctions or embargoes under Annex II 2.1(g).

Above Level 2, CADA shifts from managing risk to policing control.

At Level 3, the provider and relevant subcontractors must be “not subject to the control of a third country or a legal entity established in a third-country,” unless the Commission has recognized that country under the associated-third-country mechanism and the provider satisfies the same separation and prevention requirements under Annex II 3.1(g) and Article 18. In this context, “control” refers not merely to ownership, but to the ability to exercise decisive influence over the provider’s operations or strategic decisions.

Level 4 makes the control ban absolute. There is no associated-third-country exception. It also adds an “effective control” test for software: a third country must not be able to “materially influence the technical evolution, maintenance priorities, security remediation, and long-term continuity” of relevant software components. Union-citizen personnel requirements appear at Level 3 and continue at Level 4 under Annex II 3.1(d), 4.1(d), 4.1(g), and 4.1(i)(ii).

The result is a significant turn from technical assurance toward legal and corporate insulation. The higher the tier, the less CADA asks whether a provider can manage specific risks, and the more it asks whether the provider can demonstrate separation from the non-EU world.

D.            What EU-Only Sovereignty Can—and Cannot—Buy

There is one category of use cases for which a categorical EU-only approach is at least coherent: a narrow set of core governmental functions where member states may want to keep the circle of trust inside the European Union and are willing to bear the cost of doing so. The cost is not only financial. It may also mean accepting weaker capabilities, including the inability to use state-of-the-art AI.

That tradeoff should be evaluated clearly. The question is not whether the highest assurance level is reassuring in the abstract. It is what that level can actually deliver, and at what cost.

Consider the strongest possible case. Assume a cloud provider is as secure as current technology and operations allow. Assume also that it satisfies, as far as possible, CADA’s Level 4 requirements. That allows the analysis to isolate the maximum security benefit an EU-exclusive policy could plausibly provide.

Even under those assumptions, real independence from foreign pressure requires far more than EU ownership and organizational separation from non-EU subsidiaries, assets, or business relationships. Any asset exposed to foreign legal or extralegal pressure can become a vulnerability. Employee travel can also create risk; a third country may detain personnel and thereby exert pressure on the employer. Even a provider with no foreign customers or subsidiaries may remain vulnerable to sanctions, secondary sanctions, export controls, and other forms of economic pressure.

True independence would therefore require resilience across the whole business: hardware, software, financing, banking relationships, borrowing, and even the personal financial exposure of key employees and shareholders. That is a demanding standard. It may be easier for some state-owned enterprises than for private firms, but it would be costly and limiting in any event. It is doubtful that many proponents of EU-exclusive sovereign clouds could satisfy it in practice.

Some cloud services can be delivered under such constraints if buyers are willing to accept the costs. Others cannot. The limiting factor is often the hardware and software the provider can obtain without depending on foreign-controlled supply chains. Those limits are likely to matter most for defense and intelligence—the very domains where the highest assurance levels may appear most attractive.

The most important limitation concerns frontier AI. The frontier of commercially available AI—the most capable models at the leading edge—belongs to a small group of mostly U.S. labs, with OpenAI and Anthropic at the front and Google close behind. Unless AI-development techniques change substantially, those frontrunners are likely to remain ahead while the rest of the world trails for the foreseeable future.

Europe can access and build AI that is sufficient for many uses. EU-based labs are producing capable models, including open-weight models, meaning models whose parameters are made available for others to run or adapt. But AI that is sufficient for many uses is not the same as frontier AI.[9] That gap matters most for users in defense, intelligence, and perhaps law enforcement, where second-best capabilities may be inadequate.

Even non-frontier AI faces a second constraint: compute. Compute refers to the processing power—chips, data centers, and electricity—needed to train and run AI systems. Europe does not have enough compute capacity to meet all projected demand. The Commission’s own impact assessment projects a large and widening data-center-capacity gap, a problem discussed below. And those projections count all data-center capacity. The share that is genuinely independent of third-country leverage is far smaller, and likely close to nonexistent outside government facilities.

CADA partially recognizes this limitation. Article 30(4) lets a contracting authority depart from the assurance tiers when “no adequate or reasonable alternative … service exists.” That is precisely the frontier-AI problem. A procurement rule cannot create a missing model or supply the compute needed to run it.

There is also a problem from the opposite direction. The EU-only line assumes that all member states belong within the same circle of trust for the most sensitive workloads. That assumption merits scrutiny. Member states differ in their geopolitical orientations, including in their relationships with third countries that other member states may regard as hostile.

That matters for national security. Some member states may serve, intentionally or not, as gateways for adversaries: allowing front companies to establish themselves inside the European Union, facilitating movement of personnel, or creating opportunities for intelligence assets to operate under an EU label. If so, drawing the trust boundary at the EU border becomes less defensible. For some sensitive uses, trusting certain non-EU allies may be more rational than trusting every EU member state. I return to that point below.

III.   CADA Promises Immunity, but Delivers Fragility

Beyond a narrow core of highly sensitive government functions, CADA’s top tiers promise more than they can deliver. Stripped of procedural detail, they rest on a central premise: that keeping ownership, control, and personnel inside the European Union can shield cloud services from foreign legal and economic pressure. The act targets two real risks: that a foreign authority could gain access to European data, and that a foreign authority could disrupt a critical service. It also offers one formal exception from the Level 3 ownership and control ban: Commission recognition of a provider’s home country under Article 18.

The subsections that follow test those premises. On confidentiality, ownership does not stop a production order. Legal and economic leverage—not nationality—drives disclosure risk, and data in use raises technical problems that ownership rules cannot solve. On availability, the relevant pressure points—sanctions, export controls, supply-chain chokepoints, model access, and upstream software dependencies—often reach past an EU ownership label. Article 18, the one exit from the Level 3 control ban, also appears likely to exclude the very allied providers on whose technology Europe continues to depend.

A.             Foreign Leverage Does Not Stop at the EU Border

King v. OVH illustrates why EU ownership cannot do the work CADA assigns it. In that case, an Ontario court ordered OVH—a leading European sovereign-cloud provider—to produce subscriber data held on servers in France and other countries outside Canada. The court reasoned that the French company’s “virtual presence” in Canada brought it within Canadian legal reach, and it treated France’s data-blocking statute as a low-risk obstacle.[10] The result is significant for CADA’s sovereignty theory: a Canadian order reached EU-located data held by an EU-headquartered provider.

The point is not that Canada is uniquely aggressive. It is that foreign leverage can take many forms. Production orders are one example. A foreign government may also rely on sanctions, export controls, bank accounts, subsidiaries, employees, suppliers, or other assets within its jurisdiction. To be fully insulated from all such pressure, a provider would need no meaningful foreign nexus and no critical foreign inputs.

No major AI-cloud provider is likely to satisfy that standard. AI clouds depend on non-EU sourced hardware and software, including advanced chips, accelerators, networking equipment, and critical software layers. Many of the most important inputs remain substantially shaped by U.S. technology and U.S. export-control law. It is therefore reasonable to assume that virtually all existing EU providers would fail the strongest version of CADA’s “not subject to the control of a third country” test if that test were applied as a genuine inquiry into foreign leverage.

The Commission’s impact assessment appears to recognize this problem, at least in part. It notes that properly designed technical measures—such as “customer-controlled encryption, strict role segregation or data access minimisation”—“can technically constrain the practical ability of a third-country authority to access the data, regardless of the provider’s jurisdiction.”[11] In other words, system architecture can often mitigate disclosure risk more effectively than nationality-based controls.

The same document also acknowledges that ownership cannot make legal risk disappear. “[E]ven when a service is exceptionally well protected,” a legally binding third-country request may still compel a provider to grant access to data. “This risk exists independently of the technical robustness of the service.” The Commission therefore recommends “cannot comply” architectures, in which “compliance with external access requests [is] rendered technically, operationally or legally impossible.” It cites “the OVH case in Canada” as the relevant cautionary example.

That is the appropriate lesson, but only up to a point. The answer to foreign legal pressure is not to assume that EU ownership creates immunity. It is to design systems so that no provider, whether European or foreign, can disclose data it does not control.

B.             Data Security Depends on What the Data Is Doing

Technical protection depends on the state of the data. CADA blurs an important distinction between data at rest, meaning stored data, and data in use, meaning data being actively processed. The distinction matters because tools that protect stored data do not necessarily protect live computation.

Start with historical data and a production order that arrives after the fact. In this setting, genuine “inability to comply” can be technically achievable. If the provider retains no data, there is nothing to produce. If the provider stores only ciphertext—encrypted data that cannot be read without a key—under client-side encryption, and the customer alone holds the keys, the provider stores data it cannot read.

That appears to be what CADA’s audit standard seeks to capture when it demands inability to access “customer data, including encrypted data” under Annex III. The phrase addresses the common case in which cloud data is encrypted, but the provider manages the encryption keys. A provider that can decrypt data on demand is not truly “unable” to comply with a disclosure order.[12] Read this way, the standard is meaningful and achievable. As the Commission’s own analysis recognizes, it can work regardless of the provider’s nationality.

But CADA outsources this technical layer to a certification scheme that does not yet exist. At Levels 2 through 4, the audited service must hold a European cybersecurity certificate under a cloud-certification scheme “to be established” under the Cybersecurity Act. Levels 2 and 3 require the “substantial” assurance level; Level 4 requires “high.” That scheme is the European Union Cybersecurity Certification Scheme (EUCS), which the proposal’s explanatory memorandum acknowledges “has not yet been adopted” and on which work “will resume.”

This is a significant dependency. EUCS has been stalled since December 2020, largely because of the same sovereignty requirements that CADA now revives through procurement rules. Its adoption has been postponed without a new deadline.[13]

Until EUCS is adopted, Annex II falls back on national certification schemes “where they exist.” That approach re-fragments cloud-security assurance along national lines, even though CADA’s premise is a single Union framework. Where no national scheme exists, providers may self-demonstrate compliance with “the highest cybersecurity standards under applicable Union law,” a phrase that does not identify a clear or uniform standard.

The unadopted EUCS candidate scheme, however, already contains the control that does the most important confidentiality work for stored data. At the “high” level, the draft requires encryption keys to be “known exclusively by the cloud customer and without exceptions.” At the “substantial” level, the same requirement allows exceptions. CADA requires only the exception-permitting certificate at Levels 2 and 3, and reserves the customer-exclusive-keys certificate for Level 4. That allocation is difficult to justify.

If EU ownership delivered immunity, requiring customer-exclusive keys in addition to Level 4’s absolute control ban would add comparatively little. If, as King v. OVH shows, ownership does not stop a production order, then key custody is the control that matters—and CADA withholds it from the lower tiers where it would often be most useful. If the Commission wants providers to be unable to comply with foreign production orders, the relevant lever already exists: adopt EUCS and require customer-exclusive key custody wherever confidentiality is critical, regardless of who owns the provider.

Even customer-held keys leave residual trust questions. Are the keys truly controlled only by the customer? Are they copied, escrowed, or exposed through support processes? These questions show where the ally/adversary distinction legitimately enters the analysis. I return to that issue below.

The problem changes when data is being processed after an order is served. This issue is especially important for AI. To train or run an AI model on data, the system must decrypt that data into memory. Encryption at rest does not solve that problem. What matters is the security of the live processing environment.

The main technical candidate for protecting data in use is confidential computing in trusted execution environments. Confidential computing refers to technologies that seek to protect data while it is being processed. A trusted execution environment is a protected area inside a processor whose memory is designed to remain unreadable even to the machine’s operator. In theory, this allows a customer to process sensitive data on another party’s cloud without exposing that data to the cloud provider. In practice, CADA has two gaps.

The first gap is that CADA does not require this protection. CADA’s sovereignty-assurance criteria and audit evidence do not require confidential computing, trusted execution, or other data-in-use protections. Nor does the underlying certification stack fill the gap. The EUCS candidate scheme’s cryptography requirements cover data in transit and data at rest, but not data in use. The NIS2 implementing rules for cloud providers likewise stop at “data at rest and data in transit.”[14]

The second gap is that even trusted execution environments may not deliver the absolute “inability to comply” CADA’s top tiers imply against an adversary with physical control of the machine. Security research has repeatedly shown that commercial enclaves can be vulnerable to physical attacks by someone who controls the hardware.[15] Vendors patch specific techniques, but physical access remains a live threat category. And the party with physical control of the server is the host—the same party CADA fears may be coerced by a non-EU government.

This creates a further difficulty for CADA’s sovereignty framing. The enclave silicon and its root of trust are themselves largely controlled by U.S. firms: Intel, AMD, and Nvidia. Thus, the leading technical response to non-EU jurisdiction often depends on U.S.-controlled chips.

Taken literally, the “unable to comply … including encrypted data” standard is not achievable with the measures CADA actually requires, at least for data in use. Read loosely, it collapses into an ownership test: a formal label that does little to prevent intrusion and cannot eliminate legal compulsion. The likely result is a protectionist proxy presented as technical assurance.

C.             Resilience Beats the Kill-Switch Fantasy

The discussion so far has focused on one risk: a foreign power gaining access to European data. The Commission’s headline concern is different: that a provider could switch off, degrade, or deny access to a critical service. This is the “kill switch” problem. The practical response is not to assume the risk can be eliminated, but to ensure that users can exit, switch providers, and run critical workloads elsewhere.

CADA’s own risk taxonomy recognizes several forms of this risk: “embargos or sanctions,” “technology lock-ins,” and “sabotage.” EU ownership helps against only one narrow scenario: a foreign parent company ordering its European subsidiary to discontinue or degrade service. That is the strongest case for sovereign control in the act. It is also only a limited part of the overall risk.

Sanctions and embargoes do not require a parent company. A foreign state can sanction a European firm directly, cut it off from currency clearing or correspondent banking, restrict access to technology, or pressure its suppliers and customers through secondary sanctions. EU ownership prevents none of these measures.

For frontier AI, the most important dependency sits even further upstream: chips, software, and state-of-the-art AI models that Europe does not currently produce at comparable levels. CADA implicitly recognizes this point. To address kill-switch scenarios, the top tiers do not rely only on ownership. They require a “documented migration plan in the event that the vendor fails or a third country imposes restrictions,” controls to “block any remote features that could materially tamper with or disrupt” the system, and a “switchover plan” to “minimal viable functionality” under Annex II and Annex III.

Those requirements address resilience, not immunity. Resilience—the ability to maintain or restore service despite disruption—does not depend on the provider’s nationality.

A categorical EU-only route may even reduce resilience. Concentrating critical workloads among a small number of EU-native providers could recreate the “concentration risk” CADA’s recitals warn against. Nationality-blind multi-cloud portability—designing systems so they can run across more than one provider—would do more to reduce that risk than a formal EU ownership requirement.[16]

Higher assurance should therefore remain risk-based and grounded in technical and organizational measures. The relevant question should be whether the user can continue operating if one provider, jurisdiction, or supply chain fails. It should not be whether the provider has the preferred nationality.

The June 12 cutoff of Anthropic’s Fable 5 and Mythos 5 illustrates the limits of CADA’s approach. The cutoff occurred above the layer CADA regulates. CADA governs ownership, personnel, data location, and cloud-service structure. It does not determine who receives access to frontier AI models in the first place.

Only one CADA criterion reaches that layer: Level 4’s test of whether a third country can “materially influence … the long-term continuity” of software under Annex II 4.1. The U.S. directive confirms that a frontier American model would fail that test. But CADA’s remedy for such failure is exclusion. In effect, its answer to the frontier-AI kill switch is to forgo the state-of-the-art capability.

Nor is the Anthropic cutoff necessarily an isolated warning. On June 2—10 days earlier, and one day before CADA itself was proposed—the White House issued an executive order creating a voluntary process for developers to provide the federal government up to 30 days’ pre-release access to “covered frontier models.” The same process gives the government a role in selecting the “trusted partners” who receive those models before public release.[17] Washington is positioning itself as a gatekeeper for frontier-model distribution. Nothing guarantees Europe a place on that list.

This reality should clarify the policy choice. Europe cannot solve the kill-switch problem by assuming that ownership rules control every upstream dependency. It can reduce the risk only by building resilience: portable workloads, diversified supply chains, stronger exit rights, more compute on European soil, and enough mutual dependence that cutting Europe off would impose meaningful costs on the countries doing the cutting.

D.            CADA’s Ally Problem

For Level 3, CADA offers only one exit from the control ban: Commission recognition of the provider’s home country under Article 18. Level 4 offers no such path. That structure matters because the Article 18 criteria are cumulative, and the United States could be argued to fail several of them.

To qualify, a third country must first benefit from an adequacy decision under the General Data Protection Regulation (GDPR)—that is, a Commission finding that the country’s data-protection regime is essentially equivalent to the European Union’s. It must also have no measures that allow it to control a provider in ways that conflict with the EU Data Act’s limits on third-country access to nonpersonal data. That criterion points directly to the United States’ CLOUD Act and Foreign Intelligence Surveillance Act (FISA). The country must have no measures that could compel degradation or disruption of the service, and no sanctions leverage over the provider. That criterion points to the U.S. sanctions regime. Finally, it must have “no measures … to impede the provision of state-of-the-art technologies.” That criterion points to U.S. export controls on advanced AI chips.

At the top of the ladder, CADA may therefore operate as a categorical exclusion of U.S.-controlled providers. Read that way, the design also risks colliding with trade law. U.S. and EU cloud services are “like services,” and an ownership-or-headquarters test resembles “less favorable treatment” under the World Trade Organization’s General Agreement on Trade in Services (GATS). ICLE has made this discrimination point before in the EUCS context.[18] If CADA repeats the move, the Commission would need to show that its public-order defense justifies the restriction.

The contradiction extends beyond trade law. The Commission says Europe should remain inside the Western architecture for controlling sensitive technologies. Yet CADA’s criterion barring any country with measures that “impede the provision of state-of-the-art technologies” would disqualify a partner precisely because it operates the export controls that help constitute that architecture.[19]

This is not a hypothetical tension. On June 3—the same day it proposed CADA—the European Union also moved to join Pax Silica, the U.S.-led initiative launched in December 2025 to secure semiconductor, AI, and critical-minerals supply chains against China. Under the parallel EU-U.S. trade understanding, the bloc is expected to buy at least $40 billion in American AI chips.[20] That alliance rests on the same coordinated export controls CADA may treat as disqualifying.

The result is an internal inconsistency. The Commission seeks to bind Europe more closely to the American technology stack while using CADA’s top tiers to exclude that same ally’s providers from the most sensitive workloads.

Tethering high-tier eligibility to a GDPR adequacy decision adds another source of fragility. A Schrems-style lapse could disqualify a major provider overnight. That is precisely the kind of access shock a sovereignty framework should seek to avoid.[21]

IV.   Draw the Line at Adversaries, Not Allies

None of this means CADA should avoid third-country distinctions altogether. Some distinctions are necessary. But they should track the relevant risk. A provider controlled from China or Russia presents a categorically different risk from a provider controlled from an allied democracy. The reason is not, as Article 18 appears to suggest, that the provider’s home state possesses intelligence-collection capabilities. All capable states possess such authorities, including allies. A test built on capabilities alone therefore captures the United States and proves too much.

The more relevant question is motivation and constraint: whether the home state is likely to weaponize a provider against Europe, and whether the provider could meaningfully resist. Disqualifying an ally because it has surveillance authorities is not a sound risk proxy. It penalizes partners Europe depends on while doing too little to distinguish the adversaries it should treat with greater caution.

A Western provider served with an order targeting an ally may have courts, legal standing, commercial incentives, and diplomatic channels through which to resist. OVH, for example, is litigating the Canadian production order on appeal. A provider answerable to a nondemocratic adversary’s security apparatus is far less likely to have comparable ability to refuse, delay, litigate, or disclose the pressure it faces.

The EU border is also an imperfect trust boundary in the opposite direction. Treating all member states as a single circle of trust assumes a degree of geopolitical homogeneity the European Union does not have. A provider’s EU status says little about whether its owners, personnel, or facilities are located in a member state that an adversary could use as a gateway. Front companies, permissive local networks, and intelligence assets do not become harmless because they operate through an EU member state.

This is not an argument against drawing lines. It is an argument for drawing them around allegiance, conduct, and verifiable controls. A test built on those factors can assess risk both inside and outside the European Union. A test built only on the EU border cannot. It may include some providers that present unacceptable residual risk while excluding allied providers that can satisfy meaningful technical and organizational safeguards.

This is where the limits of audit-based regimes become important.[22] Assurance is partly a technical exercise and partly a trust regime. Auditors can verify whether controls exist, or at least whether they exist at the time of assessment. But they may struggle—or lack sufficient incentive—to pierce corporate structures when hostile ownership is involved. They also assess snapshots, not continuous practice.

Technical measures reduce the trust a customer must extend, but they do not eliminate it. Client-side encryption can be misconfigured, including by design. A zero-day vulnerability—that is, a software flaw unknown to the vendor and therefore not yet patched—can be targeted at a specific customer. Trusted execution environments, the protected processor areas discussed above, have also been shown to be vulnerable to some attacks by actors with physical access to the machine. These residual channels are precisely the channels a motivated adversary would seek to exploit while maintaining formal compliance.

Allegiance is therefore not a substitute for technical controls. It is a judgment about whose residual trust Europe can reasonably extend after technical and organizational controls have reduced risk as much as practicable. The system should be designed for the worst-case provider, just as a data-sharing regime must be designed for the worst-case recipient. But the worst case should be identified correctly.

Technical measures are also not the whole toolkit. The residual channels just described—the misconfigured key store, the targeted update, and the hands-on attack against a protected processor—are ultimately insider-risk problems. The relevant controls are organizational: personnel screening, security clearances, separation of duties, two-person control for sensitive operations, and monitored and logged administrative access.

These controls are not novel. NIS2 lists “human resources security” among its required risk-management measures. The European cybersecurity certificates that CADA’s own tiers presuppose also include personnel-security controls.

Two points follow. First, organizational controls, like technical controls, are not inherently nationality-dependent. Any provider can screen, clear, and constrain its staff. These controls retain their value among allies and among EU member states, where ownership tests often do little work. Second, to the extent organizational controls necessarily rest on trust, that trust is relational and verifiable among allies. Security-of-information agreements routinely provide for mutual recognition of clearances and procedures for handling classified information. No comparable assurance is available from an adversary on meaningful terms.

CADA partly recognizes this problem but relies on the wrong proxy. At Level 2, a public body may demand “additional personnel screening and Union citizenship requirements.” At Levels 3 and 4, all personnel must be Union citizens, with a national-security clearance required only “when handling classified information.” For classified material, the citizenship rule may track existing EU rules on classified information. But when extended to all personnel at the top tiers, the rule mistakes the proxy for the protection. The relevant protection comes from reliable screening and clearance, conducted by the provider’s own state or by an allied state—not from citizenship alone.

A vetted and cleared national of an allied country is a stronger answer to insider risk than an unvetted EU citizen. EU citizenship should not be treated as a security guarantee. The Commission has opened infringement proceedings against member states for investor-citizenship schemes, sometimes described as “golden passport” programs. In 2025, the Court of Justice held that Malta’s scheme—naturalization “in exchange for predetermined payments or investments”—was incompatible with EU law.[23] CADA’s personnel criteria should therefore require vetting to a defined standard, with allied clearances recognized under security-of-information agreements, rather than relying on citizenship as such.

The harder question is why Europe should extend trust to the United States, an ally with expansive disclosure laws and recent evidence of coercive leverage. Three answers are relevant.

First, for state-of-the-art AI, there is no realistic substitute. The frontier AI stack is overwhelmingly American, and the Commission’s own documents recognize that technological autarky is not available. The package itself states that Europe should remain open to like-minded partners. Alignment with allies is therefore not a concession; it is a practical necessity, notwithstanding real political frictions.

Second, an ally’s worst case differs in kind from an adversary’s worst case. The June 2026 Anthropic cutoff was abrupt and consequential. But allied coercion is more likely to be overt, reversible, and contestable through law, politics, diplomacy, and commercial pressure. Adversary coercion is less likely to provide those avenues.

Third, trust among allies is supported by institutional mechanisms, including adequacy decisions, mutual legal assistance, reciprocity, and shared security commitments. Against an adversary, verification stands largely alone, and verification alone can be defeated.

There are important limits to this argument. U.S. surveillance practices and laws, including FISA Section 702, led the Court of Justice of the European Union to invalidate the legal basis for personal-data transfers to the United States twice. More recently, the oversight board charged with reviewing the current framework’s safeguards lost its quorum after dismissals in January 2025, a gap the Commission’s own adequacy review identified. The General Court upheld the Data Privacy Framework in September 2025, but an appeal is pending before the Court of Justice. Section 702 itself has also been operating under short-term congressional extensions while renewal remains contested.[24]

But a Section 702 directive operates through a provider’s technical ability to access customer data. The appropriate response is to remove that ability where confidentiality is critical: customer-held keys, zero retention, and data-in-use protections—the “cannot comply” architecture the Commission’s own impact assessment recommends. Ownership does less because it does not address how the legal obligation operates.

Section 702 compels “electronic communication service providers,” a category defined by access rather than nationality. It covers providers “who [have] access to wire or electronic communications,” was expanded in 2024 to reach “any other service provider who has access to equipment that is being or may be used to transmit or store” such communications, and extends to any “officer, employee, custodian, or agent” of such a provider.[25] An EU-headquartered provider with a U.S. office, subsidiary, or employee may fall within that framework. What remains outside it is a service in which no entity or person within U.S. jurisdiction has access to plaintext or encryption keys. At that point, a directive can compel only ciphertext.

The recommendation follows from the diagnosis: CADA should retain a control inquiry, but retarget it from any third-country control to adversary control. Article 18 should become an openly political determination of allegiance and trust. For allied providers, CADA should rely on the technical and organizational safeguards its high tiers already identify: customer-held keys, vetted and cleared personnel, migration plans, anti-tamper controls, source-code audits, switchover plans, and portability. Those safeguards should apply on their own merits, not as accessories to an EU passport.

V.   Demand-Side Industrial Policy Cannot Build What Europe Lacks

Before turning to what the package should have built, one additional defense of CADA’s top tiers should be addressed. As claims of genuine immunity from foreign legal and economic pressure become harder to sustain, this defense is likely to become more prominent.

The argument is that Levels 3 and 4 are not primarily about immunity. They are instead a form of industrial policy. The state, acting as an anchor customer, reserves a portion of public procurement for EU-native providers. That predictable demand is then expected to give those providers the revenue base needed to invest in European capacity.

Taken on its own terms, this is the strongest argument for the top tiers. It is still insufficient.

First, the restriction is unlikely to remain narrow. As explained above, CADA’s institutional machinery operates as a ratchet, pushing more workloads toward the higher assurance levels over time.

Second, anchor tenancy works best when the missing input is a creditworthy customer. Europe’s cloud and AI gap is not primarily a demand problem. The binding constraints are advanced chips that Europe does not make, grid connections that can take up to a decade to build, capital markets too shallow to finance late-stage scale, and a frontier-model gap that no procurement schedule can close. Reserved demand solves none of these problems. Even the Commission’s own impact assessment expects reliance on non-European providers to remain “inevitable” across its planning horizon.

When supply cannot meet the reservation, demand-side policy does not create the missing capability. It produces supply shaped to the reservation. Engineering talent and capital are diverted toward satisfying sovereignty criteria rather than closing the capability gap. The public sector, as anchor customer, may then accept inferior technology as the price of certification.

That failure mode has already been modeled in detail, and not by Europe’s critics. The Europe 2031 scenario, written by European researchers who want Europe to compete, describes a CADA-like instrument closely.[26] Sovereignty becomes “the new buzzword in European capitals.” Announced funds turn out to be “largely a repackaging of existing funds.” A fictional Digital Sovereignty Regulation eventually requires critical public workloads to run entirely on European software.

When the capacity to meet such a mandate does not materialize, the mandate still binds. In the scenario’s wave of AI-enabled cyberattacks, the public bodies most committed to buying European are “the ones paying the ransoms,” because peers using stronger foreign models are better able to defend themselves.

VI.   Build the Capacity That Creates Leverage

The durable way to reduce dependence is not to close the market. It is to make dependence more mutual by building additional cloud and AI capacity that is located, operated, and financed in Europe. That requires greater attention to the practical constraints that determine where compute infrastructure is built: permitting, power availability, grid connection, capital formation, corporate law, tax treatment, access to advanced chips, and access to frontier models. If Europe wants leverage in the frontier-AI economy, it needs more than certificates. It needs data-center capacity, financing, and reliable energy at scale.

A.             The Permit Clock Stops Before the Hard Part

CADA’s Title III is the package’s most concrete build component, and some of its streamlining measures are meaningful. Member states must designate data-center acceleration zones under Article 10, create single information points under Article 12, combine and front-load environmental procedures at the zone level, and ensure that the project permit “shall not exceed 12 months” under Article 13(5). Strategic projects also receive an aggregated baseline permit and priority status under Article 14.

These measures are useful, but they do not reach the most important constraints.

The 12-month deadline begins only “from the moment a comprehensive application has been submitted.” The zone’s single procedure covers permits “commonly required” within the area, but expressly excludes grid-connection permits. Nor does CADA waive environmental assessment. Member states must still carry out “all necessary … environmental assessments” under Article 13(3). Unlike renewable-energy projects under the EU’s third Renewable Energy Directive, data centers do not receive “overriding public interest” status, a designation that can make it easier to justify derogations from otherwise applicable environmental rules. Part of the promised acceleration also depends on a separate, not-yet-adopted regulation on environmental assessments.

The two steps CADA brackets out may be the steps that matter most: environmental review and power-grid connection. On power, CADA largely refers to other instruments. It asks member states to analyze each zone’s future energy needs “to feed into grid planning.” It says they “should facilitate” connection procedures and “should promote” power-purchase agreements. But actual connection acceleration is left to the separate European Grids Package, the Electricity Directive, and “an upcoming legal proposal.”

That gap matters because a data center can receive its 12-month permit and still wait years to be energized. For compute projects, the central question is not only whether the project may be built, but when it can connect to sufficient electricity. The time required to energize a site often determines where compute infrastructure is built.

This is the Commission’s own diagnosis. Its impact assessment finds that connecting a data center to the grid “can take between 3 and 10 years”—seven to 10 years in established hubs, and up to 13 years where congestion is worst. It also finds that grid limits have already produced “moratoria on new DC connections,” while EU industrial-power prices are two to three times U.S. and Chinese levels. The same document acknowledges that the remedy sits outside CADA: the “necessary improvements to grid availability will be driven by the Grids package.”[27]

CADA identifies the bottleneck but leaves the principal remedy to another instrument.

B.             Europe Cannot Certify Its Way to Scale

The package largely omits the capital and corporate-law reforms needed to support European scale. That omission is revealing. The accompanying Commission communication repeats the basic diagnosis: Europe accounts for roughly 5% of global venture capital, compared with 52% for the United States.[28] But the proposed response is limited. The Commission proposes a consultation with the European Investment Bank on “European equity capacity at scale,” without specifying a funding amount, a binding legal instrument, or a concrete mechanism for building the financial infrastructure Europe’s technology ambitions require.[29]

The dependence the package seeks to reduce is large, and the Commission’s own documents acknowledge it. The European Union spends roughly €264 billion a year, much of it on proprietary U.S. information technology. The impact assessment finds that 58% of global data-center investment over the past five years went to the United States, while Europe’s financing landscape “lacks the depth of an integrated capital market.”[30] That is a capital-market problem, not a certification problem.

The flagship build program is also struggling. The €20 billion InvestAI envelope for AI gigafactories has reportedly seen bidding slip to July. Only two of five planned sites appear fundable before 2028, and the field of serious bidders has narrowed from roughly 70 to about 10.[31] These developments suggest that project financing and execution remain significant constraints.

The structural solution is integration: a real Capital Markets Union that mobilizes Europe’s pension and insurance pools for late-stage equity, and a pan-EU corporate form that gives founders the legal certainty they often seek abroad. The Capital Markets Union is the long-running EU project to integrate national capital markets so firms can raise financing more easily across member states.

The Commission did introduce a separate “EU Inc.” proposal in March 2026: a 28th-regime corporate form that would operate alongside national corporate laws. A “28th regime” is an optional EU legal framework that firms could choose instead of relying exclusively on one member state’s law. But the proposal arrived in a diluted form that commentators have judged insufficient to meet the challenge. In any event, it sits outside this package.[32] The pattern is familiar: the European Union acts more decisively where the lever is a rule or certificate than where the task is to build scalable capacity.[33]

Tax is another missing lever. Whether employee stock options are taxed at exercise or sale—and at labor-income or capital-gains rates—often determines whether a European startup can use equity compensation effectively. Equity compensation is especially important for startups because it allows firms with limited cash to compete for talent by offering employees a share in future upside. The spread across member states is substantial. In the most burdensome regimes, the standard startup compensation model becomes exceedingly difficult to use.[34]

Sweden’s experience illustrates how much fiscal design matters on the investor side as well. Its 2003 rule allowing tax-deferred rollover of gains from unlisted shares into new unlisted ventures is credited with helping create the angel-investor cycle behind Sweden’s unicorn record.[35] The 28th-regime proposal gestures at an EU-wide stock-option standard, but in the diluted, defer-to-national-law form its critics fault.[36]

CADA’s own impact assessment effectively brackets these questions. It states that taxation and depreciation “are dictated by national fiscal policies,” and that the cost of capital requires “macrofinancial instruments rather than operational regulations.” That is a candid acknowledgment that the binding constraints on building European cloud and AI capacity sit outside the package.[37] Assurance levels can regulate eligibility for procurement. They cannot by themselves deepen capital markets or produce the advanced chips needed for AI infrastructure.

C.             Build Compute, Not Autarky

The sovereignty debate often conflates three different questions. They should be kept distinct.

The first is whether European cloud providers can operate high-quality infrastructure. On that measure, Europe is competitive. Several European providers already rank in the upper tiers of independent operator assessments. Where they trail the leading providers, the gaps tend to be operational and remediable: reliability, networking, support, and monitoring.[38] Europe’s problem is not a basic inability to operate first-rate cloud infrastructure.

The second question is capacity: how much frontier-grade compute is physically located inside the European Union. Here, Europe falls short, and it cannot catch up quickly. The Commission’s own impact assessment puts the EU data-center-capacity gap at almost 3 gigawatts today, widening to roughly 19 gigawatts by 2036 under its central scenario.[39] The inputs that would close the gap fastest—advanced accelerators and the software around them—are not inputs the European Union can produce quickly by regulatory mandate. Accelerators are specialized chips, such as graphics processing units (GPUs), designed to perform the large-scale calculations required by AI systems.

The impact assessment does not avoid the implication. It expects reliance on non-European providers to remain “inevitable,” European providers to keep “struggling to work at the frontier,” and even “highly critical use cases” to continue depending on hyperscalers.[40] Hyperscalers are very large cloud providers that operate global data-center networks at massive scale.

The third question is frontier-model capability: who builds the most capable AI models themselves, as distinct from the data centers that run them. This is the gap the package addresses only indirectly. The frontier belongs to a small group of U.S. labs. Europe’s best models are useful and increasingly open, but they are not in the same class. Additional EU compute would help, but it would not by itself close the gap. Capacity can be built, albeit slowly. The model frontier continues to move, driven by larger and faster investment elsewhere.

Capacity and model capability are the real constraints, and both point toward alignment, not autarky. The advanced silicon and software Europe does not produce are the same supply-chain dependencies that ownership rules cannot solve. The Chips Act 2.0 bets on a domestic frontier fabrication facility that reaches only pilot production in 2030 to 2033. CADA’s headline goal of tripling data-center capacity adds megawatts, not models.[41]

Europe will build frontier compute, if and when it does, on imported accelerators and a largely non-European software stack. The responsible course is to secure that supply by remaining inside the Western technology-control architecture, rather than seeking self-sufficiency the Union cannot reach on a policy-relevant timeline.[42]

The point became concrete on June 12, when a U.S. export-control directive forced Anthropic to cut every non-U.S. user, including allied users, off from its Fable 5 and Mythos 5 models overnight. The lesson is not that Europe should pursue a sovereign frontier model on an implausible timeline. It is that Europe should build what it can realistically build: compute capacity on European soil, in tens of gigawatts, under European law. The relevant question is not who owns every accelerator or whose models run on them, but whether the infrastructure on which frontier AI depends is located in Europe at sufficient scale.

That capacity would create leverage. Demand for compute will likely exceed even American supply. If Europe can offer large, modern, quickly available capacity for U.S. firms’ workloads, it can create mutual dependence that makes a future cutoff more costly. As Europe 2031 argued, what hyperscalers and AI labs most need from Europe is not another subsidy, but speed: permitting and grid connection fast enough that building in the European Union is more attractive than waiting elsewhere.[43]

The same leverage could help place model weights—the files containing a model’s learned parameters—on EU-located infrastructure under contractual and physical safeguards. That would not eliminate cutoff risk. Chips, software updates, and operators would remain exposed. But it would provide partial protection, and whether Washington would allow it depends on the leverage Europe brings to the negotiation. That is another reason to build the capacity that creates leverage.

Europe does hold one imperfect chokepoint: ASML Holding N.V., the sole maker of the extreme-ultraviolet lithography machines used to manufacture the most advanced chips. Extreme-ultraviolet lithography is the process used to pattern the smallest features on advanced semiconductors. But ASML also illustrates why chokepoints should not be overstated. The company is European-led and based in the Netherlands, but its two largest shareholders are U.S. asset managers. Its light sources are built by Cymer in San Diego, and key optics work takes place in Wilton, Connecticut. U.S. Foreign Direct Product Rule jurisdiction already reaches its sales, and Washington has used that leverage to shape Dutch export policy.[44] The Foreign Direct Product Rule allows the United States to regulate some foreign-made products when they are produced using specified U.S. technology or software.

ASML is strategically important, but it does not give Europe comprehensive autonomy.

More durable leverage would come from EU-based compute, built with allies that hold other essential inputs: chips, energy, talent, and software. The objective should not be to build only within the EU-27. It should be to ensure that, when an ally rations a strategic asset, Europe is part of the negotiation rather than merely subject to the result.

VII.   Recommendations

The central claim of this issue brief is that CADA’s top tiers should require stronger security and rely less on nationality. The recommendations below follow from that premise. Technical and organizational safeguards should replace—not merely supplement—ownership and citizenship tests. Where a requirement is technical, the control details and audit evidence should come from the certification layer CADA already presupposes, extended where necessary, rather than from a parallel CADA-specific catalog.

1. Retarget the control test. Replace “not subject to the control of a third country” at Levels 3 and 4 under Annex II 3.1(g) and 4.1(g) with control by a non-allied country. Convert Article 18 from a cumulative capabilities test that no major ally can pass into an openly political determination of allegiance, institutional reliability, and trust. This would also reduce CADA’s exposure under GATS.
2. Adopt the certification scheme CADA presupposes. Every tier’s cybersecurity baseline depends on a certificate under a cloud scheme “to be established”—the EUCS certification scheme, which has been stalled since 2020 and remains unadopted. Now that CADA has moved the sovereignty dispute into procurement law, the technical certification scheme should be adopted on its own merits. Until then, the fallback to national schemes will re-fragment the assurance framework the act is intended to unify.
3. Make technical and organizational assurance do the work, without nationality distinctions. For confidentiality-critical workloads at Levels 2 and 3, require customer-exclusive key custody or zero retention for EU and non-EU providers alike. The candidate EUCS scheme already defines customer-exclusive key custody at the “high” level, but CADA’s “substantial” certificates for Levels 2 and 3 leave that safeguard optional. The safeguard should be required.
4. Close the data-in-use gap. Nothing in the current stack—CADA, the candidate EUCS scheme, or the NIS2 implementing rules—protects data in use. That is where many AI workloads present their most important risks. Annex II should state the relevant outcome: state-of-the-art protections, such as confidential computing, for high-assurance AI workloads. The European Union Agency for Cybersecurity should then develop the corresponding certification profile, so the controls remain in one framework and are audited once.
5. Make the audit standard meetable. Replace “unable to comply … including encrypted data” in Annex III with concrete definitions matched to the state of the data, paired with appropriate organizational controls. Client-side encryption and processor enclaves depend on configuration and on the insiders who administer them. For data at rest, the provider should be unable to decrypt because the customer alone holds the keys, or because nothing is retained. For data in use, where genuine inability to comply cannot be promised honestly, the standard should require attested state-of-the-art protections such as confidential computing. A precise standard will be less likely to collapse into an ownership proxy; an impossible standard almost certainly will.
6. Fix the personnel criteria. Replace the blanket Union-citizenship requirement in Annex II 3.1(d) and 4.1(d) with screening and clearance to a defined standard. Recognize allied clearances under security-of-information agreements. Retain citizenship requirements only where EU classified-information rules already require them, while still requiring screening.
7. Treat resilience as the answer to the kill switch. Apply migration plans, switchover plans, anti-tamper controls, and portability requirements at the high tiers on their own merits, regardless of provider nationality. Address concentration risk through multi-cloud portability, rather than by moving workloads onto a small number of EU incumbents.
8. De-fragilize Article 18. Decouple ongoing high-tier eligibility from the continued validity of a GDPR adequacy decision, or at least provide transition periods. A Schrems-style lapse should not become the access shock the framework is intended to prevent.
9. Harden the risk-based architecture against upward drift. Require the Commission to publish reasoned justifications when it overrides member-state risk assessments under Article 29(5). Strike or sunset the Article 31(3) delegated power to mandate private-sector assurance levels. Require periodic public reporting on the actual distribution of workloads across assurance levels, so any drift from the Commission’s own 10% estimate becomes visible.
10. Extend the data-center buildout fast-track to the real constraints. Bring grid-connection permits within the 12-month clock, or create a parallel binding deadline. Give data centers in acceleration zones overriding-public-interest status for environmental derogations, using the third Renewable Energy Directive as the model.
11. Build compute as leverage, and use it to negotiate frontier-AI access. Treat European compute capacity as a strategic asset, not merely a utility. Build it in tens of gigawatts through partnerships with U.S. firms and by hosting their workloads. Compete on permitting and grid-connection speed rather than subsidy. The goal is mutual dependence: enough EU-based capacity that cutting Europe off would impose meaningful costs on the countries doing the cutting. Use that leverage to secure assured access to frontier models and, where feasible, to host model weights on EU soil under enforceable safeguards.
12. Pair CADA with the missing instruments. CADA cannot do the build job alone. It should be paired with late-stage capital reforms, including a Capital Markets Union that unlocks pension and insurance pools; an undiluted 28th-regime corporate form; and harmonized tax treatment of employee equity.

One final drafting point: the proposal’s cross-references to the recognition mechanism appear inconsistent. Article 19 appears where Article 18 seems to be intended.

VIII.   Conclusion

The Commission made several important choices worth preserving. CADA is risk-based by default, formally limited to the public sector, and paired with a meaningful permitting fast-track. But the package also retains a categorical approach at the top tiers, where sovereignty becomes a proxy for ownership, citizenship, and distance from non-EU law.

The decisive choices now move to the Council, Parliament, and implementing acts. Those choices will determine where the assurance threshold for sensitive public-sector data settles: at Level 2, where European-operated “wrapper” ventures may remain viable, or at Levels 3 and 4, where they likely cannot. They will also determine whether Article 31(3) converts voluntary private-sector assessments into mandates, what evidence auditors will accept that foreign access “is prevented,” and whether the build half of the package—especially grid connection and capital-market reform—is completed.

CADA’s top tiers are where the package departs most clearly from its stated premise.[45] An ownership cordon that excludes Europe’s closest technology partners risks producing the isolation the Commission’s broader communication disavows, but under the label of assurance. Europe does need sovereignty in cloud and AI. But sovereignty will not come from assuming that passports secure data, certificates create chips, or ownership labels build frontier models. It will come from building capacity, hardening systems, extending trust to allies where trust is justified, and drawing the relevant line against adversaries rather than against partners Europe cannot afford to exclude.

[1] Anthropic, Statement on the U.S. Government Directive to Suspend Access to Fable 5 and Mythos 5 (June 12, 2026), https://www.anthropic.com/news/fable-mythos-access (complying with a U.S. export-control directive issued on national-security grounds following an alleged jailbreak. The directive barred access by all foreign nationals, which Anthropic implemented by disabling both models for all users. Anthropic’s other models remained available).

[2] Eur. Comm’n, Commission Proposes Tech Sovereignty Package to Strengthen Europe’s Digital Autonomy and Resilience, Press Release IP/26/1187 (June 3, 2026), https://ec.europa.eu/commission/presscorner/detail/en/ip_26_1187; Eur. Comm’n, Proposal for a Regulation of the European Parliament and of the Council Laying Down Measures to Strengthen Cloud and Artificial Intelligence Capacities in the Union (Cloud and AI Development Act), COM (2026) 502 final (June 3, 2026), https://digital-strategy.ec.europa.eu/en/library/proposal-cloud-and-ai-development-act-cada. Unless otherwise noted, references in the text to articles and recitals refer to COM (2026) 502 final, while references to Annex II and Annex III refer to its annexes. The press release states that the proposal aims to triple Europe’s data-center capacity within five to seven years.

[3] Mikołaj Barczentewicz, Build AI, Don’t Block Access: The European Union’s Digital-Sovereignty Trap, Int’l Ctr. for L. & Econ. (2026), https://laweconcenter.org/resources/build-ai-dont-block-access-the-european-unions-digital-sovereignty-trap. See also Mikołaj Barczentewicz, Build, Don’t Block—What the Compute Market Says About EU AI Sovereignty, EUTechReg (May 18, 2026), https://eutechreg.com/p/build-dont-block-what-the-compute.

[4] Cloud Infrastructure Servs. Providers in Eur. (CISPE), Put Cloud Sovereignty, Resilience, and Fair Competition at the Heart of CADA (Mar. 17, 2026), https://www.cispe.cloud/cloud-sovereignty-letter-march-2026; Dan Robinson, Don’t Let Hyperscalers Hijack Digital Sovereignty, EC Told, The Register (Mar. 18, 2026), https://www.theregister.com/off-prem/2026/03/18/dont-let-hyperscalers-hijack-digital-sovereignty-ec-told/5222651 (reporting CISPE’s “sovereignty-washing” critique and Microsoft’s concession, in France, that it could not guarantee data against a binding order).

[5] Eur. Comm’n, Commission Staff Working Document, Impact Assessment Report Accompanying the Cloud and AI Development Act, SWD (2026) 502 final (June 3, 2026) [hereinafter Impact Assessment], pt. 1, at 46, 71 (estimating, based on France’s SecNumCloud framework, that the “most sensitive use cases”—those falling within “sovereignty level 3” and “sovereignty level 4”—account for roughly 10% of public-sector use cases).

[6] CISPE, supra note 4.

[7] Comput. & Commc’ns Indus. Ass’n (CCIA), Discriminatory EU Cloud and AI Development Act Risks Severe Market Fragmentation (June 3, 2026), https://ccianet.org/news/2026/06/discriminatory-eu-cloud-and-ai-development-act-risks-severe-market-fragmentation.

[8] Kai Nicol-Schwarz, Europe Unveils Tech Sovereignty Package Amid Growing Concerns Over Reliance on U.S. Tech: ‘We Want to Be Sure Nobody Has a Kill Switch’, CNBC (June 3, 2026), https://www.cnbc.com/2026/06/03/europe-tech-sovereignty-us-tech-reliance.html (quoting Henna Virkkunen’s statement that providers of critical cloud workloads should not have a “kill switch,” that “our most critical sensitive data” should remain in Europe, and that U.S. firms would “struggle to reach the highest sovereignty tier” because of obligations under the CLOUD Act).

[9] Eur. Comm’n, Communication on European Tech Sovereignty Accompanied by an EU Open Source Strategy, COM (2026) 503 final (June 3, 2026) [hereinafter Tech Sovereignty Communication], https://digital-strategy.ec.europa.eu/en/library/communication-european-tech-sovereignty-accompanied-eu-open-source-strategy (describing Mistral AI as developing “high-performance open-weight large language models” that serve as a “sovereign alternative to closed-source systems”; characterizing the openEuroLLM project—co-funded with €20 million of its €38 million budget—as a “flagship initiative” to “secure European leadership in generative AI” through “high performing, ‘truly open’ European foundation models”; and noting that the GenAI4EU call allocates €50 million to “advance open AI models”).

[10] See Shanzay Pervaiz, What Canada’s King v. OVH Case Reveals—and Affirms—About Cross-Border Data Access, Priv. Across Borders (Feb. 3, 2026), https://privacyacrossborders.org/2026/02/03/what-canadas-king-vs-ovh-case-reveals-and-affirms-about-cross-border-data-access; Richard Speed, Canadian Data Order Risks Blowing a Hole in EU Sovereignty, The Register (Nov. 27, 2025), https://www.theregister.com/off-prem/2025/11/27/canadian-data-order-risks-blowing-a-hole-in-eu-sovereignty/2615140. The Ontario Court of Justice upheld a Royal Canadian Mounted Police production order against OVH’s French parent company for subscriber data stored on servers in France, the United Kingdom, and Australia. The court relied on the parent company’s “virtual presence” in Canada and treated France’s blocking statute as posing little risk of enforcement. The decision is currently on appeal.

[11] Eur. Comm’n, Impact Assessment, supra note 5, pt. 2, at 88–89 (quoting the “cannot comply” paradigm and the “OVH case in Canada”; noting that customer-controlled encryption, role segregation, and data-access minimization “can technically constrain the practical ability of a third-country authority to access the data, regardless of the provider’s jurisdiction”; acknowledging that legal compulsion “exists independently of the technical robustness of the service”; and describing architectures that “ensur[e] that the encryption keys are not accessible to the provider or are held exclusively by the customer”).

[12] Eur. Comm’n, Impact Assessment, supra note 5, pt. 2, at 88–89.

[13] Eur. Union Agency for Cybersecurity (ENISA), EUCS – Cloud Services Scheme (candidate scheme, Dec. 2020 draft), https://www.enisa.europa.eu/publications/eucs-cloud-service-scheme, Annex A, CKM-03.2, CKM-03.4 (providing that, at the “substantial” assurance level, encryption keys for customer data at rest “shall be known only to the cloud customer … with the possibility of exceptions,” while at the “high” assurance level they “shall be known exclusively by the cloud customer and without exceptions”). The December 2020 text remains the only publicly available candidate scheme. ENISA did not publish later revision drafts, and the scheme has not been adopted. Reports indicate that contested sovereignty requirements were removed from a 2024 draft and that adoption was postponed indefinitely. See Ctr. for Eur. Pol’y, EU Cloud Certification at an Impasse, https://www.cep.eu/eu-topics/details/eu-cloud-certification-at-an-impasse.html. The explanatory memorandum to COM (2026) 502, supra note 2, likewise acknowledges that the scheme “has not yet been adopted” and that work on it “will resume.”

[14] Commission Implementing Regulation (EU) 2024/2690 of 17 Oct. 2024, Annex, point 9.2(a) (requiring policies governing the cryptographic measures used to protect assets, “including data at rest and data in transit”; the provision does not address data in use).

[15] See Zitai Chen et al., VoltPillager: Hardware-Based Fault Injection Attacks Against Intel SGX Enclaves Using the SVID Voltage Scaling Interface, in Proc. 30th USENIX Sec. Symp. 699 (2021) (describing a voltage-glitching attack that defeats SGX on fully patched systems and arguing that the results “may require a rethink of the SGX adversarial model where a cloud provider is untrusted and has physical access to the hardware”); Robert Buhren et al., One Glitch to Rule Them All: Fault Injection Attacks Against AMD’s Secure Encrypted Virtualization, in Proc. 2021 ACM SIGSAC Conf. on Comput. & Commc’ns Sec. (2021) (demonstrating voltage-glitching attacks against the AMD Secure Processor and extending them to SEV-SNP attestation); Jesse De Meulemeester et al., BadRAM: Practical Memory Aliasing Attacks on Trusted Execution Environments, in Proc. 2025 IEEE Symp. on Sec. & Priv. (2025) (describing a sub-$10 DRAM-aliasing attack that defeats AMD SEV-SNP integrity protections and identifying CVE-2024-21944), together with its runtime successor, Battering RAM (2025), which defeats the boot-time aliasing checks Intel (TDX) and AMD adopted in response.

[16] Satya Marar & Jack Trotter, OECD Cloud-Computing Competition Study Offers Solutions in Search of a Problem, Truth on the Mkt. (July 21, 2025), https://truthonthemarket.com/2025/07/21/oecd-cloud-computing-competition-study-offers-solutions-in-search-of-a-problem.

[17] Exec. Order No. 14,409, Promoting Advanced Artificial Intelligence Innovation and Security, 91 Fed. Reg. 34,565 (June 5, 2026), https://www.whitehouse.gov/presidential-actions/2026/06/promoting-advanced-artificial-intelligence-innovation-and-security; see also WilmerHale, New Executive Order Addressing Early Government Access to Frontier AI Models (June 2, 2026), https://www.wilmerhale.com/en/insights/client-alerts/20260602-new-executive-order-addressing-early-government-access-to-frontier-ai-models. The order establishes a voluntary framework under which developers may provide the federal government with up to 30 days’ pre-release access to “covered frontier models,” a category defined through a classified national-security benchmarking process, before release to government-selected “trusted partners” and the public. The order neither requires licensing nor grants the government authority to block deployment.

[18] Eric Fruits, Lazar Radic, Mario A. Zúñiga, Mikołaj Barczentewicz & Ben Sperry, ICLE Comments to the USTR on Significant Foreign Trade Barriers, Int’l Ctr. for L. & Econ. (Oct. 30, 2025), https://laweconcenter.org/resources/icle-comments-to-the-ustr-on-significant-foreign-trade-barriers (arguing that the EUCS “immunity from non-EU law” standard is undermined by King v. OVH and that EU ownership and headquarters requirements accord U.S. “like services” “less favorable treatment,” creating potential tension with the General Agreement on Trade in Services (GATS)); see also Dirk Auer, Testimony on ‘Anti-American Antitrust: How Foreign Governments Target U.S. Businesses’, Int’l Ctr. for L. & Econ. (Dec. 16, 2025), https://laweconcenter.org/resources/testimony-on-anti-american-antitrust-how-foreign-governments-target-u-s-businesses.

[19] Mikołaj Barczentewicz, US Export Controls on AI and Semiconductors: Two Divergent Visions, Int’l Ctr. for L. & Econ. (Mar. 24, 2025), https://laweconcenter.org/resources/us-export-controls-on-ai-and-semiconductors-two-divergent-visions; see also Mikołaj Barczentewicz, US Export Controls on AI and Semiconductors, Int’l Ctr. for L. & Econ. (Mar. 25, 2025), https://laweconcenter.org/resources/us-export-controls-on-ai-and-semiconductors.

[20] See Emanuele Rossi, Why Europe Is Joining Pax Silica, Decode39 (June 2026), https://decode39.com/15107/why-europe-is-joining-pax-silica; Luca Bertuzzi, The EU Is Set to Join US-Led Chip Alliance ‘Pax Silica’ to Counter China’s AI Race, Euronews (June 1, 2026), https://www.euronews.com/my-europe/2026/06/01/the-eu-is-set-to-join-us-led-chip-alliance-pax-silica-to-counter-chinas-ai-race; Agence Europe, Member States Approve EU Participation in US ‘Pax Silica’ Initiative, Europe Daily Bulletin No. 13,880 (June 2026); U.S. Dep’t of State, Pax Silica, https://www.state.gov/pax-silica. Member-state representatives authorized the European Commission to formalize EU participation on June 3, 2026, with formal Council adoption expected on June 8. The initiative originated as a nonbinding political declaration adopted at a December 2025 summit. Reports on the parallel EU-U.S. trade understanding indicate that it includes commitments to purchase at least $40 billion in U.S. AI chips.

[21] Mikołaj Barczentewicz, Schrems III: Gauging the Validity of the GDPR Adequacy Decision for the United States, Int’l Ctr. for L. & Econ. (Sept. 25, 2023), https://laweconcenter.org/resources/schrems-iii-gauging-the-validity-of-the-gdpr-adequacy-decision-for-the-united-states.

[22] Mikołaj Barczentewicz, The European Commission’s Search-Data Trust Fall, Truth on the Mkt. (May 27, 2026), https://truthonthemarket.com/2026/05/27/the-european-commissions-search-data-trust-fall.

[23] Eur. Comm’n, Investor Citizenship Schemes: European Commission Opens Infringements Against Cyprus and Malta for “Selling” EU Citizenship, Press Release IP/20/1925 (Oct. 20, 2020), https://ec.europa.eu/commission/presscorner/detail/en/ip_20_1925 (announcing infringement proceedings against Cyprus and Malta for granting nationality—and thus EU citizenship—in exchange for “a pre-determined payment or investment” without a “genuine link” to the member state, and raising similar concerns about Bulgaria’s scheme); Case C-181/23, Comm’n v. Malta (Grand Chamber Apr. 29, 2025), https://eur-lex.europa.eu/legal-content/EN/TXT/?uri=CELEX:62023CJ0181 (holding that “transactional naturalisation … in exchange for predetermined payments or investments” violates Article 20 TFEU and the duty of sincere cooperation under Article 4(3) TEU); see also Simon Cox, The EU Free Market Does Not Extend to Citizenship, Verfassungsblog (Apr. 30, 2025), https://verfassungsblog.de/the-eu-free-market-does-not-extend-to-citizenship.

[24] Case C-362/14, Schrems v. Data Prot. Comm’r, ECLI:EU:C:2015:650 (Oct. 6, 2015) (invalidating the Safe Harbour adequacy decision); Case C-311/18, Data Prot. Comm’r v. Facebook Ir. Ltd., ECLI:EU:C:2020:559 (July 16, 2020) (invalidating the Privacy Shield adequacy decision based on surveillance authorities under Section 702 of the Foreign Intelligence Surveillance Act and Executive Order 12,333, as well as the lack of effective judicial redress). See also Silvia Lorenzo Perez, What the PCLOB Firings Mean for the EU-US Data Privacy Framework, Ctr. for Democracy & Tech. (2025), https://cdt.org/insights/what-the-pclob-firings-mean-for-the-eu-us-data-privacy-framework (discussing the January 2025 dismissals that deprived the Privacy and Civil Liberties Oversight Board of a quorum). The General Court upheld the third adequacy decision in Case T-553/23, Latombe v. Commission (Sept. 3, 2025); the appeal is pending as Case C-703/25 P. See also Justin Papp, FISA Section 702: Congress Passes Short-Term Surveillance Program Extension Just Before Deadline, CNBC (Apr. 30, 2026), https://www.cnbc.com/2026/04/30/fisa-section-702-congress-extension.html (reporting on Congress’ short-term extensions of Section 702 in spring 2026).

[25] 50 U.S.C. § 1881(b)(4) (defining “electronic communication service provider”); Reforming Intelligence and Securing America Act, Pub. L. No. 118-49 (2024) (adding § 1881(b)(4)(E), which extends coverage to “any other service provider who has access to equipment that is being or may be used to transmit or store wire or electronic communications,” subject to exclusions for accommodation facilities, dwellings, community facilities, and food-service establishments). Congress enacted the provision in response to a Foreign Intelligence Surveillance Court of Review decision rejecting the government’s effort to classify an entity—reportedly a data-center operator—as a covered provider. An amicus curiae before the Foreign Intelligence Surveillance Court warned that the new language could reach landlords and virtually any business with communications equipment on its premises. The government subsequently narrowed the provision’s scope, but the details of that interpretation remain classified. See Andreas Kuersten, FISA Section 702 and the 2024 Reforming Intelligence and Securing America Act, Cong. Rsch. Serv., R48592 (July 8, 2025), https://www.congress.gov/crs-product/R48592; Restore the Fourth, Scope of FISA Sec. 702 ECSP Provision Narrowed but Remains Classified (June 4, 2024), https://restorethe4th.com/scope-of-fisa-sec-702-ecsp-provision-narrowed-but-remains-classified.

[26] Daan Juijn, Stan van Baarsen, Judith Dada, Lily Stelling, Philip Fox, Alex Petropoulos & Michiel Bakker, Europe 2031, ARQ Found. (June 2026), https://europe2031.ai. The scenario posits a “Digital Sovereignty Regulation requiring critical public sector workloads to run on 100% European cloud and AI software by 2032” and explores the consequences after frontier cyber-offensive capabilities proliferate. In its fictional 2026–2029 chronology, “[t]hose agencies most committed to the ‘Buy European’ agenda—organisations which have procured exclusively from European providers—are now the ones paying the ransoms,” while “[o]rganisations that kept an American contract on the side are doing better.”

[27] Eur. Comm’n, Impact Assessment, supra note 5, pt. 1 (finding that connecting a data center to the electricity grid “can take between 3 and 10 years”—3 to 5 years in emerging markets, 7 to 10 years in established hubs, and up to 13 years in the most congested locations; noting that grid constraints have led to “moratoria on new DC connections”; and reporting that industrial electricity prices in the European Union are two to three times higher than in the United States and China, while accounting for 40% to 50% of data-center operating costs). The report further notes, in footnote 90, that “the necessary improvements to grid availability will be driven by the Grids package.”

[28] See Mario A. Zúñiga, Draghi Report Highlights Why to Be Wary of the ‘Brussels Effect’, Truth on the Mkt. (Sept. 26, 2024), https://truthonthemarket.com/2024/09/26/draghi-report-highlights-why-to-be-wary-of-the-brussels-effect (discussing the European Union’s aspirational Capital Markets Union, venture-capital investment that remains a fraction of U.S. levels, and the tendency of European “unicorns” to relocate abroad).

[29] Eur. Comm’n, Tech Sovereignty Communication, supra note 9 (reporting EU and U.S. shares of global venture-capital investment and announcing the European Investment Bank’s “European equity capacity at scale” consultation, as well as the €120 billion and €200 billion investment figures).

[30] Eur. Comm’n, Impact Assessment, supra note 5, pt. 1 (finding that 58% of global data-center investment over the previous five years occurred in the United States; that AI-computing infrastructure can be “ten to thirty-times more expensive” than general-purpose data centers; and that the European Union’s “fragmented financing landscape lacks the depth of an integrated capital market”); see also Eur. Comm’n, Tech Sovereignty Communication, supra note 9 (reporting that the European Union spends roughly €264 billion annually, much of it on proprietary U.S. information-technology products and services).

[31] Ana-Maria Stanciuc, EU AI Gigafactory Plan Stumbles as Delays Alienate Partners, The Next Web (June 2026), https://thenextweb.com/news/eu-ai-data-centre-gigafactory-delays-funding-stumble; Christina Kyriasoglou, Gian Volpicelli & Paula Doenecke, EU’s AI Data Center Plans Stumble Due to Delays, Funding Issues, Bloomberg (June 2, 2026), https://www.bloomberg.com/news/articles/2026-06-02/eu-s-ai-data-center-plans-stumble-due-to-delays-funding-issues. Reports indicate that the €20 billion InvestAI envelope—combining EU and member-state subsidies with private capital—would support up to five AI gigafactories. The bidding process slipped to July 2026, only two of the five proposed sites appear financeable before 2028, and the number of interested consortia reportedly fell from roughly 70 to about 10.

[32] Luca Enriques, Casimiro Nigro & Tobias Tröger, Why the 28th Regime Proposal Falls Short of Europe’s Challenge, Oxford Bus. L. Blog (Mar. 19, 2026), https://blogs.law.ox.ac.uk/oblb/blog-post/2026/03/why-28th-regime-proposal-falls-short-europes-challenge; see also Eur. Comm’n, Proposal for a Regulation Establishing a 28th Regime (“EU Inc.”), COM (2026) 321 final (Mar. 18, 2026).

[33] Mikołaj Barczentewicz & Kristian Stout, How Not to Use Industrial Policy to Promote Europe’s Digital Sovereignty, Truth on the Mkt. (Oct. 5, 2022), https://truthonthemarket.com/2022/10/05/how-not-to-use-industrial-policy-to-promote-europes-digital-sovereignty; see also Mikołaj Barczentewicz & Kristian Stout, The EU’s Cybersecurity Draft Shifts Toward Hard Protectionism, Truth on the Mkt. (Nov. 14, 2023), https://truthonthemarket.com/2023/11/14/eus-cybersecurity-draft-shifts-toward-hard-protectionism (criticizing the EUCS “immunity from non-EU law” approach from which CADA’s highest sovereignty tiers descend).

[34] Index Ventures, Rewarding Talent: A Guide to Stock Options for European Entrepreneurs, https://www.indexventures.com/rewarding-talent (comparing national stock-option tax regimes and finding that the Baltic states rank among Europe’s most favorable jurisdictions, while Belgium, Germany, and Spain rank among the most burdensome).

[35] Luis Garicano & Per Strömberg, Why Sweden Has So Many Unicorns, Silicon Continent (Feb. 16, 2026), https://www.siliconcontinent.com/p/why-sweden-has-so-many-unicorns (arguing that Sweden’s 2003 reform—which defers taxation of gains from unlisted shares when investors reinvest the proceeds in other unlisted companies—helped create the country’s angel-investor ecosystem).

[36] Luis Garicano & Ulrike Malmendier, Calling Something the 28th Regime Does Not Make It One, Silicon Continent (Mar. 17, 2026), https://www.siliconcontinent.com/p/calling-something-the-28th-regime (noting the proposal’s EU-wide stock-option regime while criticizing its reliance on national law and arguing for a regime focused on young, cross-border-scalable firms with genuinely harmonized equity-compensation and tax rules for key employees).

[37] Eur. Comm’n, Impact Assessment, supra note 5, pt. 2 (reporting sensitivity analysis of the data-center financial model and finding that the weighted average cost of capital is “predominantly affected by financial market conditions and sovereign risk, necessitating macrofinancial instruments rather than operational regulations to influence it,” and that “[t]axation and depreciation are dictated by national fiscal policies”).

[38] Barczentewicz, supra note 3.

[39] Eur. Comm’n, Impact Assessment, supra note 5, pt. 1 (projecting an EU data-center-capacity gap of “almost 3 GW” in 2025, widening under the central scenario to roughly 19 GW by 2036. The assessment projects supply reaching about 42 GW by 2036, but demand growing even faster).

[40] Eur. Comm’n, Impact Assessment, supra note 5, pt. 1 (finding, under the baseline scenario, that the limited scale and scope of EU providers “makes reliance on non-European providers inevitable”; that “European service providers [are] struggling to work at the frontier”; and that dependence on hyperscalers, “particularly for highly critical use cases,” will persist).

[41] Eur. Comm’n, Proposal for the Chips Act 2.0 (June 3, 2026), https://digital-strategy.ec.europa.eu/en/library/proposal-chips-act-20. The proposal envisions a first-of-its-kind European foundry combining leading-edge manufacturing at the 3 nm node and below with chiplet integration and advanced 3D packaging. It targets pilot production between 2030 and 2033 and allocates roughly €30 billion to the foundry as part of an estimated €120 billion semiconductor-investment requirement through 2035. The proposal also replaces the original Chips Act’s goal of a 20% global market share by 2030 with the less specific objective of securing a “growing global share.” The text contains no reference to high-bandwidth memory or frontier logic chips.

[42] Mikołaj Barczentewicz, You Can’t Regulate a GPU Into Existence, Truth on the Mkt. (May 18, 2026), https://truthonthemarket.com/2026/05/18/you-cant-regulate-a-gpu-into-existence.

[43] Juijn et al., supra note 26. The Europe 2031 scenario’s closing retrospective frames compute capacity as the critical opportunity Europe failed to seize. A fictional European Commission trade official argues that Europe needed “the metal on our soil”—data centers, chips, and power supplies “anchored under European law, in jurisdictions Washington could not commandeer at six hours’ notice”—and that capacity “had to be in the tens of gigawatts.” The retrospective contends that Europe “could have got to fifteen, maybe twenty per cent [of global compute] in five years,” that the prospect of losing access to that scale of compute would have constrained U.S. action, and that speed was decisive because “bringing a single data centre online a month faster was worth billions.” It concludes that the priority should have been ensuring that the data centers “that are built are bolted to our floor.”

[44] ASML Holding N.V., Shares, https://www.asml.com/en/investors/shares. ASML is headquartered in the Netherlands and remains European-led, but its two largest shareholders are U.S. asset managers. According to the company’s disclosures as of Feb. 18, 2026, BlackRock held 6.83% of its ordinary shares and Capital Research and Management Company (Capital Group) held 5.09%. ASML’s EUV light sources are developed by Cymer in San Diego, key optics work is performed in Wilton, Connecticut, and U.S. export-control jurisdiction extends to its equipment through the Foreign Direct Product Rule. See Barczentewicz, supra note 19.

[45] Eur. Comm’n, Tech Sovereignty Communication, supra note 9 (stating that European technological sovereignty rests on “openness, partnership and fair competition”; emphasizing that it “does not mean isolation, protectionism, or tech decoupling”; and committing the European Union to remain “open to the world” while pursuing “mutually beneficial technology partnerships with countries that share our vision”).

Executive Summary

The real technological race is not over who invents the next breakthrough application. It is over who builds the infrastructure that allows innovation ecosystems to form, scale, and persist. Communications and computing infrastructure increasingly function as general-purpose economic inputs, much like electricity or transportation networks. Their value depends not only on deployment, but on complementary assets such as skilled labor, capital, energy availability, and regulatory environments that support experimentation and growth.

This issue brief argues that next-generation technological leadership will depend on a converging infrastructure chain that includes wireless spectrum, terrestrial broadband, fiber backhaul, satellite systems, edge computing, hyperscale data centers, and AI accelerators. These layers are interdependent: spectrum without backhaul creates stranded capacity; data centers without fiber become isolated facilities; edge computing without wireless access cannot effectively serve users.

Innovation also remains geographically concentrated. Breakthrough inventions, venture capital, technical talent, and specialized infrastructure continue to cluster in jurisdictions that reduce deployment friction and support complementary ecosystems. Artificial intelligence (AI), augmented reality (AR), virtual reality (VR), and immersive computing intensify these dynamics because they require low latency, high-capacity connectivity, large-scale compute resources, and reliable power infrastructure.

At the same time, permitting delays, fragmented rights-of-way rules, environmental-review requirements, pole-attachment disputes, and energy constraints can significantly slow infrastructure deployment. As networks densify toward 5G, 6G, and edge-computing architectures, these transaction costs become more consequential. Indeed, reducing deployment friction may produce larger long-run innovation gains than targeted subsidies for favored technologies. Performance-based, technology-neutral policies are generally better suited to this challenge.

The central policy lesson is that infrastructure investment is the most effective form of horizontal industrial policy. Jurisdictions that make it easiest to build, connect, compute, and experiment—and that cultivate the complementary ecosystem of skills, capital, and institutions—will attract the firms and innovation clusters that define the next digital economy.

I.                    Infrastructure and the New Geography of Innovation

Public debate frequently warns that the United States risks “losing the race” to China in technologies such as 5G, artificial intelligence (AI), augmented reality (AR), and next-generation wireless systems. That framing often implies that a rival nation can “capture” a technology market, as though 5G or AI were finite prizes one jurisdiction wins and another loses. But these technologies are not zero-sum product markets. They are general-purpose infrastructure layers whose value depends on deployment breadth and ecosystem depth.

The “race” metaphor nevertheless points to a real concern. Jurisdictions that fail to deploy the underlying infrastructure risk losing the developers, investors, and firms that build on top of it. The danger is not merely losing market share. It is losing the locus of innovation—the places where the next generation of products and services is conceived, tested, and scaled. Technological leadership depends not only on who invents a technology, but also on where the infrastructure exists to commercialize it.

That dynamic has become more important as the computing frontier converges around common infrastructure requirements. Generative AI, agentic systems, and immersive augmented- and virtual-reality applications increasingly depend on ultra-low latency, distributed computing capacity, and high-capacity wireless networks. Where infrastructure gets built—and how quickly—now helps determine who participates in the next wave of innovation.

This is not a new insight. Despite decades of falling communication costs, technological innovation remains geographically concentrated. Roughly 56% of the most consequential new technologies emerge from just two U.S. corridors: Silicon Valley and the Northeast. The geographic diffusion of related employment then takes roughly 50 years.[1] Far from dissipating, this concentration has intensified. Tech clusters capture an increasing share of U.S. patenting activity—not because knowledge itself is rivalrous, but because the complementary inputs required to produce innovation, including talent, capital, and specialized infrastructure, remain scarce and geographically concentrated.[2]

The persistence of place-based clustering despite dramatic improvements in connectivity suggests that infrastructure alone cannot explain innovation geography. The interaction among infrastructure availability, deployment regulation, and the self-reinforcing dynamics of agglomeration also matters.[3] Innovation ecosystems cluster in jurisdictions that make it easier to deploy communications and computing infrastructure, experiment with new architectures, and scale digital platforms.

Knowledge spillovers—the external benefits that arise when ideas and know-how spread among nearby firms, workers, and institutions—also remain geographically localized. They fade slowly over time and remain bounded not only by physical distance, but also by jurisdictional lines.[4] Differences in permitting requirements, energy regulation, spectrum allocation, and procurement practices create policy discontinuities that can segment innovation networks even among nearby regions. As deployment frictions rise, investment and experimentation migrate toward jurisdictions that impose fewer barriers to value creation.

Next-generation computing architectures—including AI inference, immersive computing, and edge computing—depend on dense infrastructure capable of supporting low-latency connectivity and massive computational throughput. But the relevant infrastructure extends well beyond the endpoint connection. Communications networks, including fiber backhaul, satellite systems, and wireless access, increasingly converge with computing infrastructure, including data centers, edge nodes, and accelerator clusters. Technological leadership therefore depends on the full infrastructure stack, not merely one layer.

The familiar “race to China” framing from industrial-policy debates is therefore not entirely wrong. But infrastructure’s economic value does not come from its mere existence. It comes from the local complements—human capital, organizational capacity, specialized services, and supporting institutions—that turn capacity into productivity. A jurisdiction that deploys fiber, spectrum, and edge computing capacity, but fails to create the conditions for those complements to emerge, will not capture the innovation gains that infrastructure enables.

In that sense, losing a technological race is not principally a question of whether a country builds infrastructure. It is a question of whether it builds in ways that allow local ecosystems to scale, coordinate, and reinforce one another.

This issue brief maps how these infrastructure layers interact and identifies the regulatory and economic dynamics that will determine whether the United States remains at the center of the next generation of technology ecosystems. It is the first in a series examining the infrastructure domains—including spectrum, deployment barriers, satellite connectivity, and broader policy design—that will shape the emerging competitive landscape.

II.                    Infrastructure as a General-Purpose Technology

Communications and computing infrastructure increasingly function as general-purpose economic inputs, analogous to electricity or transportation networks.[5] The general-purpose technology (GPT) framework helps explain both the breadth of infrastructure’s economic effects and the long lag between deployment and measurable returns. Communications networks exhibit all three canonical GPT characteristics: they are pervasive across sectors, capable of sustained technical improvement, and highly complementary with downstream applications. The internet’s status as a GPT is now well-established.

What distinguishes the current transition from earlier rounds of edge-network evolution—such as content-delivery networks, flatter peering arrangements, or Differentiated Services-era traffic management—is its scale and integration. Computation and intelligence are migrating outward from centralized facilities toward the network edge. AI inference at cell sites, programmable network slicing, and compute-aware routing increasingly distribute computational capacity throughout the network itself. Communications infrastructure is evolving from a passive conduit for transmitting data into an active computing platform.

Whether this converged “network-as-compute-platform” model has fully achieved GPT status remains debatable. What is clear is that the infrastructure chain’s geographic dependence is intensifying. Each layer increasingly relies on physical proximity to the others to satisfy latency, bandwidth, and computational demands.[6] These characteristics create increasing returns and a coordination problem: the network’s value depends on the density of applications built on top of it, while developers’ incentives depend on the network’s quality, scale, and reach.

New infrastructure therefore rarely generates immediate productivity gains. Firms and workers must reorganize production, retrain employees, and redesign business processes around the new technology. Those complementary investments take years to complete and even longer to appear in measured output.

Economists describe this pattern as the “productivity J-curve,” which helps explain the “Solow Paradox” identified by Robert Solow in 1987: one could “see the computer age everywhere but in the productivity statistics.”[7] The paradox becomes less puzzling once the lag between deployment and measurable productivity gains is understood as a predictable feature of GPT adoption, rather than evidence of technological failure.[8]

A complementary explanation may also exist. Organizations often appear to absorb efficiency gains into coordination, task-switching, and workplace slack, rather than converting them directly into measured output. Researchers have documented this pattern across multiple waves of automation technology.[9]

Historical experience supports both explanations. The productivity gains from electrification emerged only decades after initial deployment, and jurisdictions differed substantially in how quickly they captured those gains. Organizational restructuring, competitive conditions, and regulatory environments all shaped adoption.[10] The same pattern appeared with steam power, electricity, and information and communications technology (ICT): each produced substantial economic gains, but only after long adjustment periods, with innovation activity concentrating in early-deploying jurisdictions.[11]

These complementarity effects are also deeply geographic. The inputs necessary to exploit a GPT—including skilled labor, organizational capital, specialized suppliers, developer communities, and venture financing—are themselves geographically concentrated. The returns from infrastructure deployment therefore depend partly on the local density of these complementary assets.

Jurisdictions that deployed earlier generations of communications infrastructure enter each successive technology wave with compounding advantages.[12] Industry-level evidence confirms that ICT-using industries—not merely ICT-producing industries—drove the productivity acceleration after 1995, and that total factor productivity growth correlated with ICT investment during the preceding decade.[13] Cross-country evidence on intangible investment further suggests that jurisdictions with greater organizational-capital formation extract disproportionately greater value from equivalent levels of infrastructure deployment.[14]

A.             Agglomeration and Innovation Geography

Digital technology reduces five categories of economic friction: search, replication, transportation, tracking, and verification.[15] But not all frictions matter equally for innovation geography. Agglomeration economies—the benefits firms derive from geographic concentration—operate through three primary mechanisms: sharing, matching, and learning.[16] Each has distinct empirical implications.

Of these mechanisms, learning appears most resistant to digital substitution. Frontier innovation depends heavily on tacit knowledge transfer, informal exchange, and iterative feedback loops that rely on trust and high-fidelity interaction.[17] Infrastructure that lowers latency and enables real-time collaboration may therefore have different geographic effects than infrastructure that merely increases bandwidth.

The empirical evidence on broadband’s economic effects—including employment growth, productivity, and innovation—is similarly nuanced. Broadband consistently supports marginal product and process innovation through technology adoption and diffusion, even if it does not directly generate breakthrough invention. That distinction matters. Breakthrough technologies may emerge disproportionately from dense innovation clusters, while the marginal gains from deploying communications infrastructure may be greatest in areas where connectivity was previously weak or unavailable.

Broadband’s economic effects also vary substantially by region. Employment gains tend to be strongest in lower-density areas, rather than major urban markets. Productivity effects, meanwhile, depend heavily on pre-existing human capital and technical expertise.[18] Broadband investment generally produces stronger productivity gains in regions with deeper knowledge stocks and more highly skilled workforces.

Communications infrastructure is often evaluated primarily through a bandwidth lens—how much data can move through a network per second. That focus risks obscuring other network characteristics that shape innovation geography. Latency, reliability, edge-computing capacity, and network density can matter just as much for the formation of innovation clusters.

Broadband deployment can increase innovation activity even without contemporaneous productivity gains.[19] That pattern is consistent with the J-curve dynamics associated with GPTs, in which complementary investments and organizational adaptation precede measurable productivity improvements. Innovation clusters may therefore begin forming before conventional economic metrics register meaningful gains.

Historical evidence points in the same direction. A 10% increase in mobile penetration has been associated with a 0.59% to 0.76% increase in GDP per capita, with larger effects in countries with stronger human-capital bases.[20] Venture-capital geography likewise illustrates how infrastructure and complementary ecosystem inputs reinforce spatial concentration.

Venture-capital investment remains highly clustered because investors face substantial information asymmetries and benefit from proximity-based monitoring. Reduced travel time between venture-capital firms and portfolio companies can improve innovation output and increase exit probabilities.[21] Conversely, when venture-capital supply contracts in a region, startups often relocate elsewhere, suggesting that capital availability remains a binding constraint on ecosystem formation.[22]

The COVID-19 pandemic briefly appeared to weaken these geographic constraints by enabling more remote investment. Yet the resulting “death of distance” proved incomplete. Remote deal-making still depends heavily on high-quality digital infrastructure, while physical proximity remains especially important for large transactions and ventures marked by severe information asymmetries.[23]

Computing infrastructure shows similar concentration dynamics. Third-party data centers tend to cluster near urban demand centers, while hyperscale cloud providers often locate facilities in lower-density regions optimized for land, electricity, and cooling costs.[24] At the regional edge layer, peering locations and Internet Exchange Points carry substantial economic consequences. Sparse interconnection points or restrictive peering policies can degrade content quality and increase costs for platform providers.[25]

Specialized human capital further reinforces these dynamics. Broadband investments correlate with faster productivity growth primarily in labor markets with complementary technical skills.[26] In some cases, faster connectivity can even reduce productivity if firms and workers lack the expertise needed to integrate new technologies into existing workflows.[27] Infrastructure investment without complementary ecosystem inputs may therefore generate expectations that outpace economic results.

Jurisdictions that combine communications infrastructure with strong talent pipelines, skilled labor markets, and complementary institutions are far more likely to capture the productive and innovative gains these technologies enable.[28]

The broader pattern is clear: innovation clusters emerge where three conditions align. First, jurisdictions must minimize deployment frictions for communications and computing infrastructure. Second, they must support dense connectivity networks that enable experimentation and collaboration. Third, they must sustain capital markets capable of scaling successful technologies.

Clusters thus operate as bundles of complements, including specialized suppliers, deep labor pools, supporting institutions, and infrastructure networks. Infrastructure alone is rarely sufficient to create a new innovation center. Industry-life-cycle research consistently finds that firms cluster geographically near early industry leaders, from automobiles to semiconductors.[29] Historical examples include the development of Wi-Fi and the U.S. unlicensed-spectrum ecosystem, broadband infrastructure’s role in cloud-computing growth, and divergent deployment environments across the United States, Europe, and China.[30]

B.             The Converging Infrastructure Chain

The geographic concentration of technological innovation is becoming more pronounced as next-generation applications demand increasingly sophisticated infrastructure. These applications do not depend on any single network layer. They rely instead on a converging “infrastructure chain”—the layered set of physical and logical systems that support modern digital services.

Historically, communications networks operated in discrete domains: cable networks delivered television, fiber served enterprise connectivity, wireless networks enabled mobility, and satellites provided remote coverage. Those boundaries are rapidly dissolving. Nearly all networks now support Internet Protocol (IP)-based connectivity and cloud-computing services.[31] At the same time, the distinction between communications and computing infrastructure is fading. Data centers, edge nodes, and AI accelerators increasingly function as core infrastructure components, no less essential than the networks connecting them.[32]

This converging infrastructure chain consists of five interdependent layers. The first is the local-access layer, which includes Wi-Fi, 5G, ethernet, and emerging 6G standards. It enables devices to communicate with nearby computing resources. The second is the terrestrial broadband layer, including fiber and cable networks that transport large volumes of data between local access networks and regional or hyperscale data centers.

The third layer consists of backhaul and edge infrastructure, which enables low-latency processing closer to end users. The fourth is satellite connectivity, particularly low-Earth-orbit (LEO) constellations. Satellite systems cut across the infrastructure chain rather than occupying a single layer: the same network can provide consumer broadband in underserved regions while supplying backhaul capacity for terrestrial networks.

Satellite infrastructure nevertheless raises distinct policy concerns. Spectrum allocation, orbital coordination through the International Telecommunication Union, and the economics of globally shared infrastructure create regulatory issues with no direct terrestrial analogue.

The fifth layer is compute itself: data centers, graphics-processing-unit (GPU) clusters, and edge servers that provide the processing power required for AI systems, immersive applications, and cloud services.

The economic stakes are substantial. One recent industry study estimated the value of Wi-Fi in the United States at roughly $1.3 trillion in 2023, with projections reaching $2.4 trillion by 2027. The study attributed much of that growth to enterprise-productivity gains, expanded unlicensed use of the 6 GHz spectrum band, and increasing augmented- and virtual-reality adoption.[33]

Licensed mobile services generate comparable effects. The U.S. wireless industry contributes more than $825 billion annually to gross domestic product, while 5G deployment is projected to generate approximately $1.5 trillion in economic growth and 4.5 million jobs over the course of this decade. Globally, the GSM Association estimates that mid-band 5G alone will contribute $610 billion to global GDP by 2030, accounting for nearly 65% of the total socioeconomic value generated by 5G.[34] Realizing those gains depends heavily on adequate spectrum allocation and timely infrastructure deployment.

Mobile edge computing further raises the geographic stakes by moving computational workloads from centralized data centers closer to end users. That architecture enables latency-sensitive and computation-intensive applications on devices with limited local processing capacity.[35]

The transition toward 6G intensifies these constraints. Proposed 6G standards target sub-millisecond latency and terabit-per-second throughput.[36] At those speeds, physics becomes economically consequential. Because light travels through fiber-optic cables at roughly 200,000 kilometers per second,[37] applications requiring round-trip latencies below one millisecond cannot tolerate compute resources located much farther than roughly 100 kilometers away. Jurisdictions unable to support dense, low-latency edge deployments will therefore face structural disadvantages in hosting the most demanding next-generation applications.

The convergence of wireless infrastructure and AI computing is already underway. NVIDIA’s AI-RAN architecture deploys GPU-accelerated platforms at cell sites capable of simultaneously supporting wireless-network and AI workloads on shared infrastructure. Under this model, base stations evolve from single-purpose radio-access points into multi-purpose edge-computing platforms capable of handling voice, video, data, and AI inference simultaneously.

In October 2025, NVIDIA and a coalition of U.S. firms—including T-Mobile, Cisco, Nokia, and Booz Allen—announced an “all-American” AI-RAN stack built on NVIDIA’s AI Aerial platform. The initiative aims to accelerate the transition toward 6G through AI-driven spectrum management and integrated sensing-and-communications systems relevant to public safety and national-security applications. The U.S. Defense Department (DOD), for example, is exploring integrated sensing-and-communications technologies for drone detection and tracking.[38]

In March 2026, T-Mobile and NVIDIA announced the deployment of “physical AI” applications—including computer vision, video search, and autonomous-system coordination—on AI-RAN-ready infrastructure.[39] That same month, a broader international coalition—including BT Group, Deutsche Telekom, Ericsson, Nokia, SK Telecom, SoftBank, and T-Mobile—committed to building 6G systems on open, secure, AI-native platforms. NVIDIA Chief Executive Officer Jensen Huang described the effort as transforming “the world’s telecom networks into AI infrastructure everywhere.”[40]

Beyond AI-enabled cell sites, firms are also developing “AI Grid” architectures that distribute computing workloads across geographically dispersed regional hubs, central offices, metro nodes, and edge data centers. These systems do not necessarily require direct co-location with wireless infrastructure. NVIDIA’s AI Grid reference design allows telecom providers and distributed cloud operators to route AI-inference workloads dynamically to the location offering the best combination of latency, cost, and available capacity.

Early benchmarks suggest these distributed architectures can reduce cost per token by as much as 76% relative to centralized deployments. Major operators—including AT&T, Comcast, and Indosat—have begun deploying AI-grid systems, leveraging a global network of roughly 100,000 distributed network data centers that may eventually support more than 100 gigawatts of additional AI-computing capacity.

III.                    The Infrastructure Demands of Next-Generation Computing

Infrastructure’s economic value depends largely on the applications built on top of it. Over the past two decades, smartphones, app-based platforms, and gig-economy services built on fourth-generation (4G) wireless networks and high-speed broadband helped drive U.S. leadership in the digital economy. Those applications turned communications infrastructure into a foundation for new markets, business models, and forms of economic coordination.

The next generation of computing technologies will impose even greater infrastructure demands. AI, AR, virtual reality (VR), and related immersive-computing systems require more than bandwidth. They also depend on ultra-low latency, high-capacity wireless connectivity, edge computing, reliable fiber backhaul, and access to large-scale compute resources.

These technologies differ from earlier internet applications in important ways. Many rely on continuous real-time inference, bidirectional data flows, deterministic network performance, and geographically distributed compute architectures. Their workloads place simultaneous pressure on communications networks, data centers, power systems, and spectrum resources.

As the previous section explained, innovation ecosystems form where complementary infrastructure layers and supporting institutions co-locate. The same dynamic applies here. AI systems, immersive applications, and edge-computing architectures increasingly depend on dense clusters of compute, connectivity, power, and specialized human capital operating as an integrated infrastructure stack.

This section examines two dimensions of that transformation: how AI and hyperscale computing are reshaping the geography and economics of data-center infrastructure, and how AR, VR, and immersive computing are changing the technical requirements of communications networks.

A.             Compute Infrastructure and the Geography of AI

The computing demands of the digital economy are compounding rapidly. For the past two decades, centralized cloud services—storage, web hosting, and software-as-a-service (SaaS) platforms—have run on general-purpose processors housed in a relatively small number of hyperscale data centers. That model persists and continues to expand. Large-scale AI, however, has introduced a qualitatively different level of resource demand.

Training frontier AI models requires thousands of specialized accelerators operating in parallel for weeks or months. Inference—the deployment of trained models to generate real-time outputs—creates continuous, geographically distributed demand for computing resources that differs substantially from traditional cloud workloads. AI systems are evolving from isolated cloud-based tools into persistent, multimodal agents operating across phones, vehicles, wearables, and AR and VR platforms. Those applications generate continuous, latency-sensitive, and increasingly upload-heavy traffic flows.[41]

In this sense, data centers have become the factories of the digital economy. Their geographic distribution shapes innovation much as the location of steel mills shaped industrial development a century ago. Specialized hardware requirements—including GPUs, tensor-processing units, and custom AI accelerators—have created supply-chain bottlenecks and concentrated compute capacity among the small number of firms capable of deploying infrastructure at scale.

The location of that compute capacity is not economically neutral. Firms developing AI applications need low-latency access to inference infrastructure and reliable high-bandwidth connectivity to deliver outputs to users. Training workloads are more geographically flexible, tending to cluster in regions with inexpensive land, abundant electricity, and sufficient water for cooling. Inference workloads, by contrast, depend heavily on proximity because they must respond to user requests in real time.

The emergence of frontier AI models outside the traditional U.S. hyperscaler ecosystem—most notably DeepSeek—does not undermine this geographic-concentration thesis. Those models still depend heavily on hardware designed and manufactured through U.S.-led supply chains. China’s aggressive investment in AI infrastructure likewise reinforces the central point: compute geography matters.

Data-center infrastructure exhibits three economic characteristics that shape these geographic dynamics.

First, massive capital requirements create substantial barriers to entry. A single hyperscale data center can cost between $500 million and $1 billion to construct, while large campuses often require multibillion-dollar investments.[42] Those costs create strong economies of scale that favor concentration among relatively few operators. The global data-center industry is now valued at roughly $250 billion, with approximately 500 major facilities concentrated in the United States, Europe, and China.[43]

Second, energy consumption has become a binding growth constraint. Data centers already account for a substantial and growing share of electricity demand, and AI workloads consume significantly more energy than traditional cloud computing. Some projections suggest global data-center electricity consumption could double by 2030 under current growth trajectories.[44] Access to affordable and reliable electricity increasingly determines where new computing infrastructure can be deployed. Energy policy and grid capacity have therefore become core infrastructure-policy concerns.

Third, computing infrastructure is consolidating around hyperscale firms that combine physical infrastructure, software platforms, and AI models into vertically integrated stacks. This integration reflects the economics of frontier AI. Training runs that cost hundreds of millions of dollars naturally favor firms with the balance sheets to finance them. Access to high-performance computing is becoming a critical input to competitiveness: the availability, cost, and proximity of compute resources increasingly shape which firms, industries, and jurisdictions can participate in AI-driven innovation.

Newer techniques such as model distillation and mixture-of-experts architectures—including DeepSeek’s R1 model—show that capable systems can be derived at far lower cost than frontier training runs. But distillation still depends on frontier models from which knowledge can be extracted. These techniques reduce the cost of deploying advanced capabilities; they do not eliminate the concentrated compute requirements necessary to advance the frontier itself. The asymmetry in high-performance compute therefore persists at the layer most relevant to technological leadership.

Data-center economics also reflect a persistent tension between concentration and distribution. Hyperscale cloud providers cluster facilities in locations optimized for land, power, and cooling efficiency.[45] That concentration improves scale economies, but it also creates resilience risks and geographic disparities in access. Regions far from hyperscale clusters often face higher latency, reduced redundancy, and less competitive cloud-service pricing.

Edge computing provides a partial counterweight. By distributing computational capacity closer to end users, edge infrastructure can support latency-sensitive applications that centralized cloud architectures cannot efficiently serve. But edge deployment introduces tradeoffs. Edge nodes sacrifice some efficiency advantages of large-scale data centers in exchange for physical proximity to users and applications. Their viability depends on sufficient local demand and supporting infrastructure, including fiber backhaul and reliable power.

AI infrastructure also depends on the communications networks connecting data centers to users and one another. Fiber backhaul functions as the circulatory system of the AI economy. Training data must move from collection points to training clusters, trained models must be distributed to inference endpoints, and inference traffic must travel between users and compute resources with minimal latency. The scale of this traffic is increasing rapidly as AI-enabled applications spread across sectors.

AI-driven network demand differs from traditional internet traffic in several respects. Training workloads require massive, sustained data transfers capable of saturating even high-capacity links. Inference traffic, while modest per query, creates continuous demand across distributed endpoints. Many emerging AI applications—particularly those involving real-time sensor data, video analysis, and multimodal agents—also produce symmetrical or upload-heavy traffic flows that invert the traditional download-heavy structure of consumer internet use.

Many AI applications also require guaranteed quality of service: predictable latency and bandwidth rather than the best-effort delivery model that historically characterized internet connectivity.[46]

B.             Immersive Computing and Network Architecture

AR, VR, and mixed-reality applications place unusually demanding requirements on communications and computing infrastructure, even relative to AI. The primary constraint is the human perceptual system. Visual-vestibular conflicts caused by latency above roughly 20 milliseconds can induce motion sickness, creating a hard engineering threshold that network architecture must satisfy.

Latency is only part of the challenge. Immersive applications also require sustained high-bandwidth throughput for high-resolution stereoscopic rendering, deterministic performance rather than best-effort delivery, and continuous bidirectional data flows as devices constantly upload sensor data and receive rendered outputs.

Legacy communications networks were not designed for these traffic patterns. Consumer internet usage historically followed heavily asymmetrical download-to-upload ratios, often approximating 90/10. Streaming video, web browsing, and file downloads dominated demand, while uploads remained comparatively limited. Cable and digital-subscriber-line (DSL) architectures were optimized accordingly, allocating most network capacity to downstream traffic.

Immersive computing fundamentally alters that model. AR and VR devices continuously transmit sensor data, camera feeds, positional-tracking information, and environmental maps to edge or cloud-computing systems while receiving rendered video frames, spatial audio, and haptic feedback. The resulting traffic patterns are often symmetrical or upload-heavy—precisely the type of demand for which many legacy access technologies are poorly suited.

This is not a marginal engineering issue, but a structural limitation of existing network architecture. Fiber-optic networks, with their inherently symmetrical capacity, and advanced wireless systems such as 5G and Wi-Fi 6E/7 are better suited to support immersive-computing workloads at scale.[47]

Meeting these demands requires coordination across multiple layers of the converging infrastructure chain. 5G wireless ultra-reliable low-latency communications capabilities provide the wireless performance guarantees needed for untethered AR and VR devices. Wi-Fi 6E and Wi-Fi 7, operating in the 6 GHz band, support high-density indoor deployments with the bandwidth and latency characteristics needed for immersive applications in enterprise, education, and health care.

Edge computing provides another critical layer by offloading computationally intensive tasks from wearable devices to nearby infrastructure. Network slicing—the creation of virtualized dedicated networks within shared physical infrastructure—allows providers to guarantee the deterministic performance immersive applications require.

The Federal Communications Commission’s (FCC) 2023 authorization of Very Low Power (VLP) devices in the 6 GHz band may prove especially significant. VLP operations allow untethered AR and VR headsets to achieve sub-millisecond latency over 160 MHz channels without requiring incumbent-protection mechanisms. Economic spillovers associated with AR and VR applications enabled by expanded Wi-Fi capacity are projected to reach $47.8 billion by 2027.[48]

The offload architecture itself is equally important. Next-generation wearable devices face unavoidable constraints involving heat, battery capacity, weight, and device size. Performing all computation locally is often impractical. The emerging solution distributes intensive tasks—including rendering, AI inference, and spatial mapping—to nearby edge-computing resources.

These architectures require the simultaneous availability of high-bandwidth wireless connectivity, low-latency edge compute, and reliable fiber backhaul—in other words, the full converging infrastructure chain described above.

AR and VR technologies function simultaneously as infrastructure drivers and beneficiaries. As applications, they generate demand for network upgrades and edge-computing deployment that benefit the broader digital economy. As platforms, they create new markets in enterprise collaboration, surgical training, industrial design, remote education, and immersive entertainment.

The economic potential is substantial. Realizing it, however, depends on infrastructure readiness. Immersive applications cannot achieve widespread adoption if the underlying networks cannot support them.

IV.                    Infrastructure Policy and Deployment Frictions

Getting infrastructure policy right today will shape innovation geography for decades. Infrastructure investments are long-lived, their returns emerge slowly, and jurisdictions that attract early deployment often capture cumulative advantages that become hard to dislodge. Private capital does not automatically flow where investment would produce the greatest social returns. It responds to regulatory environments, permitting timelines, and subsidy structures that can accelerate, distort, or foreclose deployment.

Infrastructure deployment is constrained not only by engineering costs and market demand, but also by the regulatory transaction costs between investment decisions and operational infrastructure. Because these projects are capital-intensive and largely irreversible, time matters. Permitting delays impose carrying costs, foreclose alternative investments, and increase the risk that market conditions shift before deployment is complete. In some cases, delay can determine whether a project proceeds at all.

These problems are especially acute for network infrastructure, which often requires sequential approvals from multiple governmental bodies with different authorities and timelines. A single fiber route may require federal environmental review, state highway-right-of-way permits, county or municipal road-crossing approvals, and negotiations with private property owners. Each process applies different standards, follows its own timeline, and is administered by officials with varying technical familiarity with telecommunications infrastructure.

The United States’ layered system of government compounds these coordination problems. Federal lands may involve overlapping jurisdiction from agencies such as the Bureau of Land Management, the U.S. Forest Service, and the DOD, each with separate environmental-review and permitting regimes. State highway rights-of-way are governed by 50 different regulatory systems with varying fee structures, technical requirements, and processing timelines. Local zoning and permitting add thousands of jurisdiction-specific rules. Infrastructure permitted by right in one municipality may require a conditional-use hearing in the next.

These frictions become more consequential as networks evolve toward dense 5G and 6G wireless deployments. Traditional macro-cell networks required relatively few tower permits per jurisdiction. Small-cell architectures require orders of magnitude more attachment points on utility poles, buildings, and street furniture. Cost-and-coverage modeling suggests that reaching roughly 90% population coverage is generally achievable under baseline assumptions, while the final 10% becomes exponentially more expensive and highly sensitive to regulatory design.[49] Pole-attachment disputes alone may account for as much as 25% of rural broadband deployment costs.[50]

Federal environmental review under the National Environmental Policy Act (NEPA) and historic-preservation review under Section 106 of the National Historic Preservation Act can serve legitimate public interests. But they can also impose substantial friction on next-generation infrastructure deployment.[51] The policy challenge is to calibrate review to the actual environmental or historical impact of the activity. Routine deployments—such as attaching equipment to existing structures, installing fiber within existing rights-of-way, or collocating antennas on existing towers—usually present minimal environmental or preservation concerns, yet they often trigger review frameworks designed for major construction projects.[52] Categorical exclusions and streamlined review pathways for routine deployments can reduce timelines without materially undermining environmental or preservation goals.

The FCC’s wireless-siting rules address part of this coordination problem through “shot clock” requirements that obligate local governments to act on siting applications within specified timeframes.[53] These rules create a federal backstop against indefinite local delay. Enforcement, however, remains difficult. Some local governments have circumvented shot-clock requirements by declaring applications incomplete on technical grounds, imposing temporary moratoria, or requiring supplemental studies that restart the clock.[54] Similar timeframe mechanisms for wireline deployment—including fiber permitting, pole attachments, and rights-of-way access—could reduce another major source of friction, though reforms should balance speed against meaningful procedural review.

Regulatory frictions are not the only way public policy shapes deployment. Subsidy programs also play a central role. Subsidies carry familiar economic risks: they can distort investment incentives, invite rent-seeking, and miss their intended beneficiaries. They nevertheless remain a persistent feature of infrastructure policy, both because some market failures may justify public support and because political economy makes subsidy programs hard to avoid. Given that subsidies will continue, program design matters enormously. Poorly designed subsidies can lock in inferior technologies, suppress competition, and create dependencies that outlive their purpose.

A recurring design question is whether subsidy programs should mandate particular technologies or allow providers to select deployment approaches based on local conditions. The case for technology neutrality rests on a straightforward insight: regulators are poorly positioned to predict which technologies will prove most cost-effective across the country’s varied geography and demand conditions. Fiber, fixed wireless, and LEO satellite systems each have comparative advantages in different environments. Mandating one technology risks foreclosing better solutions.[55]

Recent federal programs illustrate these tradeoffs. The Rural Digital Opportunity Fund (RDOF) revealed the risks of inadequate performance verification when some winning bidders proved unable to meet deployment commitments.[56] The Broadband Equity, Access, and Deployment (BEAD) program likewise encountered implementation challenges after initially prioritizing fiber in high-cost regions where projected deployment costs exceeded available funding.

In June 2025, the National Telecommunications and Information Administration (NTIA) issued a restructuring policy notice—described as the “Benefit of the Bargain” reform—that eliminated the prior administration’s fiber preference, removed rate-regulation mandates, and required states to conduct at least one additional competitive subgrantee-selection round open to all technologies meeting statutory performance benchmarks of 100/20 megabits per second and 100 milliseconds latency.[57] The reform required states and territories to reopen competition on a technology-neutral basis, allowing fixed wireless, cable, and LEO satellite providers to compete alongside fiber providers on equal terms. NTIA estimated the restructuring would reduce taxpayer costs by at least $21 billion through greater competition and increased private-sector matching contributions.

The BEAD restructuring illustrates a broader principle: subsidy programs that mandate particular technologies risk inflating costs and foreclosing better deployment options. Performance-based standards—which set minimum service thresholds while allowing providers to choose deployment methods—better fit the geographic diversity of the United States, where mountainous terrain, remote areas, and widely varying population densities make no single technology universally cost-effective.

V.                    Conclusion

Four conclusions emerge from the preceding analysis.

First, infrastructure policy likely matters more for innovation than direct industrial targeting. Telecommunications infrastructure has a positive causal relationship with economic growth. But its effects are not uniform. Infrastructure investment can produce both concentration and dispersion, depending on the type of connectivity and the local stock of complementary assets, including skilled labor, capital, and institutions. Even technologies that sharply reduce distance frictions can reinforce existing economic centers. Submarine fiber-optic cables, for example, reduced certain spatial frictions by as much as 80%, yet increased London’s share of global trading activity by roughly one-third.[58]

Second, the converging infrastructure chain requires integrated policymaking. Spectrum policy, broadband deployment, data-center siting, edge computing, and satellite connectivity are typically governed by separate agencies under separate regulatory frameworks. Yet each layer depends on the others. Spectrum without backhaul creates stranded capacity; data centers without fiber connectivity become isolated facilities; edge computing without wireless access networks cannot serve users effectively.

Demand for one layer also increases the value of adjacent layers. The relationship between licensed and unlicensed spectrum illustrates the point. Wi-Fi offloads an estimated $33 billion annually in cellular-network capital expenditures, while expanded unlicensed access to the 6 GHz band generated an estimated $870 billion in incremental economic value during its first two years.[59] The growth of advanced wireless services outside the home also increases demand for high-speed residential fiber and in-home wireless networking. Policymaking that treats these technologies in isolation risks creating coordination failures that market participants cannot resolve on their own.

Third, reducing deployment friction across the connectivity and compute stack may generate greater long-run innovation benefits than targeted subsidies for favored technologies. The productivity J-curve associated with general-purpose technologies implies that jurisdictions must sustain investment through extended periods of apparently low returns. Policies that reduce deployment costs—including permitting reform, streamlined environmental review, technology-neutral subsidy design, and energy-policy modernization—may therefore produce larger gains than efforts to select technological winners directly.

Fourth, recent federal actions suggest an emerging bipartisan consensus that next-generation infrastructure requires coordinated federal strategy rather than piecemeal regulatory adjustment. Executive Order 14179 and the subsequent AI Action Plan identified maintaining global AI leadership as an explicit federal policy objective and directed agencies to reduce barriers to AI deployment.[60] The December 2025 executive order establishing a national AI policy framework went further, directing the U.S. Justice Department (DOJ) to challenge state AI laws deemed inconsistent with federal policy and conditioning certain BEAD funds on states’ AI regulatory posture.[61]

Whether one views these measures as necessary coordination or federal overreach, they reflect the jurisdictional-competition dynamics described throughout this brief. The federal government is trying to prevent cross-jurisdictional regulatory fragmentation that economic-geography research associates with slower infrastructure deployment and weaker innovation clustering. The central policy question is not whether coordination is necessary, but how best to design it—balancing regulatory uniformity against state experimentation and the risk that federal preemption may suppress legitimate local objectives.

Public debate over whether the United States risks “losing the race” in technologies such as 5G, AI, and AR often focuses too narrowly on the technologies themselves. Technological leadership depends more fundamentally on the environment in which new technologies can be deployed, scaled, and commercialized.

Next-generation computing architectures depend on a converging infrastructure chain that includes wireless spectrum, terrestrial broadband, edge and hyperscale computing, and satellite systems. The infrastructure challenge no longer concerns connectivity alone. Data centers, AI accelerators, and edge-computing nodes now join fiber networks, towers, and spectrum as core inputs to the innovation economy. Their deployment depends on regulatory environments, subsidy structures, energy policy, and workforce availability—policy domains historically treated separately, but which increasingly function as parts of a single system.

Technological leadership ultimately follows infrastructure. Infrastructure investment is therefore best understood as horizontal industrial policy: it creates broad enabling conditions rather than selecting specific firms or technologies for support.[62] Vertical interventions require policymakers to predict future winners, a task regulators have historically performed poorly.[63] Horizontal infrastructure investment expands the complementarity base for innovation. It lowers entry barriers across sectors, enables applications policymakers cannot anticipate, and allows competitive markets rather than political allocation to determine which technologies succeed.

Jurisdictions that make it easiest to deploy and experiment with next-generation connectivity and compute—and that cultivate the complementary ecosystem of skills, capital, and institutions—will attract the developers, investors, and firms that define the future digital economy.

The technological race is real. Winning it requires treating infrastructure deployment not as a discrete capital expenditure, but as the foundation on which the complementary inputs that generate innovation can emerge and scale. The papers that follow in this series examine these infrastructure domains in greater detail: spectrum policy, the transaction costs of fiber and wireless deployment, satellite connectivity, and policy frameworks for efficient infrastructure expansion.

[1] Aakash Kalyani et al., The Diffusion of New Technologies, 140 Q.J. Econ. 1299 (2025), https://doi.org/10.1093/qje/qjaf002.

[2] Brad Chattergoon & William R. Kerr, Winner Takes All? Tech Clusters, Population Centers, and the Spatial Transformation of U.S. Invention, 51 Research Pol’y 104418 (2021), https://doi.org/10.1016/j.respol.2021.104418. Patenting activity, however, does not necessarily correspond to greater innovation. See, e.g., Andrew W. Torrance, Lisa C. Friedman & Tanya Singh, Is China’s Patent Boom a Bust?, 63 Hous. L. Rev. 453 (2025), https://houstonlawreview.org/article/154481-is-china-s-patent-boom-a-bust. Patent counts are an imperfect proxy for innovation and historically have correlated only weakly—or even inversely—with total factor productivity growth. They nevertheless remain useful for tracking the geographic distribution of formal intellectual-property activity.

[3] See, e.g., Chris Forman, Avi Goldfarb & Shane Greenstein, The Internet and Local Wages: A Puzzle, 102 Am. Econ. Rev. 556 (2012), https://doi.org/10.1257/aer.102.1.556 (finding that advanced internet adoption correlated with wage growth only in the 6% of U.S. counties that were already wealthy, educated, and IT-intensive, suggesting that improved connectivity reinforced rather than dispersed existing geographic concentration).

[4] Adam B. Jaffe, Manuel Trajtenberg & Rebecca Henderson, Geographic Localization of Knowledge Spillovers as Evidenced by Patent Citations, 108 Q.J. Econ. 577 (1993), https://doi.org/10.2307/2118401. Knowledge spillovers are not the only force driving agglomeration. Asset complementarities—including shared labor pools, specialized suppliers, common infrastructure, and accumulated industry experience—also exhibit strong geographic concentration. See, e.g., Alfred Marshall, Principles of Economics bk. IV, ch. X (8th ed. 1920) (identifying localized skilled-labor pools, specialized input suppliers, and knowledge spillovers as the three classical sources of agglomeration); Michael E. Porter, The Competitive Advantage of Nations 148–54 (1990) (describing how factor conditions, demand conditions, related industries, and firm rivalry cluster geographically); Maryann P. Feldman & David B. Audretsch, Location, Location, Location: The Geography of Innovation and Knowledge Spillovers, WZB Discussion Paper No. FS IV 96-28 (1996) (discussing how complementary innovation inputs co-locate and reinforce one another). The relevant point here is not that knowledge spillovers alone drive geographic concentration, but that they are the externality most directly tied to the policy question of where innovation emerges and why it persists there.

[5] Describing communications and computing infrastructure as a “general-purpose input” is an economic observation about its role in enabling a broad range of downstream activities, not a legal claim about the proper regulatory framework. General-purpose inputs vary substantially in both market structure and regulation. Electricity and freight rail, for example, are subject to common-carrier or public-utility obligations, while commercial real estate and cloud computing are not. Whether a particular layer of digital infrastructure warrants common-carriage regulation depends on market-specific conditions—including contestability, switching costs, and the availability of substitutes—not merely on the input’s generality.

[6] See, e.g., Yong Mao, Changshuo You, Jun Zhang, Kaibin Huang & Khaled B. Letaief, A Survey on Mobile Edge Computing: The Communication Perspective, 19 IEEE Commc’ns Surveys & Tutorials 2322 (2017), https://doi.org/10.1109/COMST.2017.2745201 (explaining that edge computing’s latency constraints require physical proximity between computing and radio resources).

[7] See Erik Brynjolfsson, Daniel Rock & Chad Syverson, The Productivity J-Curve: How Intangibles Complement General Purpose Technologies, Nat’l Bureau of Econ. Rsch., Working Paper No. 25148 (2018), https://www.nber.org/papers/w25148.

[8] See Daron Acemoglu, David Autor, David Dorn, Gordon H. Hanson & Brendan Price, Return of the Solow Paradox? IT, Productivity, and Employment in U.S. Manufacturing, 104 Am. Econ. Rev. 394 (2014), https://doi.org/10.1257/aer.104.5.394.

[9] See Kristian Stout & Ian Adams, The Hype Cycle Meets Malpractice Law: Why the Jobs Persist, Truth on the Mkt. (Apr. 2, 2026), https://truthonthemarket.com/2026/04/02/the-hype-cycle-meets-malpractice-law-why-the-jobs-persist; Daron Acemoglu & Pascual Restrepo, Automation and New Tasks: How Technology Displaces and Reinstates Labor, 33 J. Econ. Persps. 3 (2019), https://doi.org/10.1257/jep.33.2.3; Brynjolfsson et al., supra note 7.

[10] Paul David & Gavin Wright, General Purpose Technologies and Surges in Productivity: Historical Reflections on the Future of the ICT Revolution, in The Economic Future in Historical Perspective (2006), https://doi.org/10.5871/bacad/9780197263471.003.0005.

[11] Nicholas Crafts, Artificial Intelligence as a General-Purpose Technology: An Historical Perspective, 37 Oxford Rev. Econ. Pol’y 521 (2021), https://doi.org/10.1093/oxrep/grab012.

[12] See Steven Klepper, Industry Life Cycles, 6 Indus. & Corp. Change 145 (1997), https://www.jstor.org/stable/2696374; Susanto Basu & John G. Fernald, Information and Communications Technology as a General-Purpose Technology: Evidence from US Industry Data, 8 Ger. Econ. Rev. 146 (2007), https://doi.org/10.1111/j.1468-0475.2007.00402.x (finding that total factor productivity growth during the post-1995 acceleration correlated with prior-decade ICT investment intensity, suggesting that earlier infrastructure deployment predicted later productivity gains).

[13] Basu & Fernald, supra note 12.

[14] Carol Corrado et al., Innovation and Intangible Investment in Europe, Japan, and the United States, 29 Oxford Rev. Econ. Pol’y 261 (2013), https://doi.org/10.1093/oxrep/grt017.

[15] Avi Goldfarb & Catherine E. Tucker, Digital Economics, 57 J. Econ. Lit. 3 (2019), https://doi.org/10.1257/jel.20171452.

[16] Gilles Duranton & Diego Puga, Micro-Foundations of Urban Agglomeration Economies, in 4 Handbook of Regional and Urban Economics 2063 (J. Vernon Henderson & Jacques-François Thisse eds., 2004), https://doi.org/10.1016/S1574-0080(04)80005-1.

[17] Taiwan’s semiconductor cluster illustrates the point. The tacit process-engineering knowledge concentrated in Hsinchu Science Park—built through decades of real-time collaboration among fab engineers, equipment vendors, and design houses—reflects precisely the sort of learning externality that resists replication through remote coordination, as ongoing U.S. reshoring efforts have demonstrated. See Sujai Shivakumar & Charles Wessner, The Role of Industrial Clusters in Reshoring Semiconductor Manufacturing, Ctr. for Strategic & Int’l Studs. (Oct. 2024), https://www.csis.org/analysis/role-industrial-clusters-reshoring-semiconductor-manufacturing.

[18] See Irene Bertschek, Daniel Cerquera & Gordon J. Klein, More Bits—More Bucks? Measuring the Impact of Broadband Internet on Firm Performance, 25 Info. Econ. & Pol’y 190 (2013), https://papers.ssrn.com/sol3/papers.cfm?abstract_id=1852365 (finding that broadband adoption significantly improved firms’ capacity for process and product innovation, but had no statistically significant direct effect on labor productivity); Jed Kolko, Broadband and Local Growth, 71 J. Urb. Econ. 100 (2012), https://www.sciencedirect.com/science/article/abs/pii/S0094119011000490 (finding a positive relationship between broadband expansion and local employment growth, particularly in lower-density areas); Stefanie A. Haller & Seán Lyons, Broadband Adoption and Firm Productivity: Evidence from Irish Manufacturing Firms, 39 Telecomm. Pol’y 1 (2015), https://www.sciencedirect.com/science/article/abs/pii/S0308596114001554 (finding no statistically significant effect of broadband adoption on firms’ overall productivity or productivity growth); Elizabeth A. Mack & Alessandra Faggian, Productivity and Broadband: The Human Factor, 36 Int’l Reg’l Sci. Rev. 392 (2013), https://journals.sagepub.com/doi/abs/10.1177/0160017612471191 (concluding that broadband improved regional productivity only in areas with substantial human capital and highly skilled workforces).

[19] See Bertschek, Cerquera & Klein, supra note 18.

[20] Kalvin Bahia, Pau Castells & Xavier Pedrós, The Impact of Mobile Technology on Economic Growth: Global Insights from 2000–2017 Developments, Int’l Telecomms. Soc’y Conf. Paper (2019), https://www.econstor.eu/bitstream/10419/205164/1/Bahia-et-al.pdf.

[21] Shai Bernstein, Xavier Giroud & Richard R. Townsend, The Impact of Venture Capital Monitoring, 71 J. Fin. 1591 (2016), https://doi.org/10.1111/jofi.12370.

[22] Jun Chen & Michael Ewens, Venture Capital and Startup Agglomeration, 80 J. Fin. 2343 (2025), https://doi.org/10.1111/jofi.13451.

[23] Pengfei Han, Chunrui Liu, Xuan Tian & Kexin Wang, Invest Local or Remote? The Effects of COVID-19 Lockdowns on Venture Capital Investment Around the World, 70 Mgmt. Sci. 3640 (2025), https://doi.org/10.1287/mnsc.2024.04374.

[24] Tommy Pan Fang & Shane Greenstein, Where the Cloud Rests: The Economic Geography of Data Centers, Harv. Bus. Sch. Working Paper No. 21-042 (2025), https://www.hbs.edu/ris/Publication%20Files/21-042_47feacb8-215d-408f-81e2-8b101b9a8f0a.pdf.

[25] Stefan Bauer, William H. Lehr, Georgios Smaragdakis & Volker Stocker, The Growing Complexity of Content Delivery Networks: Challenges and Implications for the Internet Ecosystem, 41 Telecomm. Pol’y 1003 (2017), https://doi.org/10.1016/j.telpol.2017.08.008.

[26] Mack & Faggian, supra note 18.

[27] Id.

[28] Id.

[29] Michael E. Porter, Clusters and the New Economics of Competition, 76 Harv. Bus. Rev. 77 (1998), https://hbr.org/1998/11/clusters-and-the-new-economics-of-competition.

[30] See, e.g., Giuseppe Colangelo, Regulatory Myopia and the Fair Share of Network Costs: Learning from Net Neutrality’s Mistakes, Int’l Ctr. for L. & Econ. (2023), https://laweconcenter.org/resources/regulatory-myopia-and-the-fair-share-of-network-costs-learning-from-net-neutralitys-mistakes.

[31] See, e.g., Eric Fruits, Ben Sperry & Kristian Stout, The Competitive Effects of the Proposed Charter/Cox Transaction, Int’l Ctr. for L. & Econ. (2025), https://laweconcenter.org/resources/the-competitive-effects-of-the-proposed-charter-cox-transaction (describing intermodal competition in broadband connectivity); see also Kristian Stout & Geoffrey A. Manne, Promoting Competition and Innovation in the Evolving Video Sector, Int’l Ctr. for L. & Econ. (2025), https://laweconcenter.org/resources/promoting-competition-and-innovation-in-the-evolving-video-sector (discussing the convergence of video communications networks).

[32] This process has been underway since the introduction of computer equipment in the 20th century. The 1956 consent decree in United States v. W. Elec. Co. barred AT&T from entering any business other than common-carrier communications services, effectively preventing the Bell System from manufacturing or selling computer equipment. That artificial separation between communications and computing persisted for more than two decades, until the 1982 Modified Final Judgment dissolved it. See United States v. W. Elec. Co., No. 17-49 (D.N.J. 1956) (final judgment); see also United States v. Am. Tel. & Tel. Co., 552 F. Supp. 131 (D.D.C. 1982) (Modified Final Judgment).

[33] Raul Katz, Juan Jung, Fernando Callorda & Ramiro Valencia, Assessing the Economic Value of Wi-Fi in the United States, Telecom Advisory Servs. (2024), https://wififorward.org/wp-content/uploads/2024/09/Assessing-the-Economic-Value-of-Wi-Fi.pdf.

[34] GSM Ass’n, Press Release, GSMA Research Finds Mid-Band 5G Could Deliver $610B GDP Growth by 2030 (Oct. 13, 2022), https://www.gsma.com/newsroom/press-release/gsma-research-finds-mid-band-5g-could-deliver-610b-gdp-growth-by-2030.

[35] Yong Mao, Changshuo You, Jun Zhang, Kaibin Huang & Khaled B. Letaief, A Survey on Mobile Edge Computing: The Communication Perspective, 19 IEEE Commc’ns Surveys & Tutorials 2322 (2017), https://doi.org/10.1109/COMST.2017.2745201.

[36] Walid Saad, Mehdi Bennis & Mingzhe Chen, A Vision of 6G Wireless Systems: Applications, Trends, Technologies, and Open Research Problems, 34 IEEE Network 134 (2020), https://doi.org/10.1109/MNET.001.1900287.

[37] Klint Finley, Brits Approach (True) Speed of Light Over Fiber Cable, Wired (Mar. 20, 2013), https://www.wired.com/2013/03/internet-at-the-speed-of-light.

[38] NVIDIA, Press Release, NVIDIA and US Telecom Leaders Unveil the All-American AI-RAN Stack to Accelerate the Path to 6G (Oct. 28, 2025), https://nvidianews.nvidia.com/news/nvidia-us-telecom-ai-ran-6g.

[39] NVIDIA, Press Release, T-Mobile and Partners Integrate Physical AI Applications on AI-RAN-Ready Infrastructure (Mar. 16, 2026), https://nvidianews.nvidia.com/news/t-mobile-nvidia-ai-ran-physical-ai.

[40] NVIDIA, Press Release, NVIDIA and Global Telecom Leaders Commit to Build 6G on Open and Secure AI-Native Platforms (Mar. 1, 2026), https://nvidianews.nvidia.com/news/nvidia-and-global-telecom-leaders-commit-to-build-6g-on-open-and-secure-ai-native-platforms.

[41] Shashank Modi, Simonas Matulionis, Michael Kleeman, Ernesto Wandeler, John Fleury, Jaison Justin, Heinz T. Bernold & Braden Holstege, 6G: The Network for the Future of AI and Immersive Connectivity, Bos. Consulting Grp. & Qualcomm (Feb. 26, 2026), https://www.bcg.com/publications/how-6g-networks-will-shape-the-next-era-of-ai.

[42] Ari Natter & Will Wade, SoftBank Plans Giant Ohio AI Data Center Powered by Gas Plants, Bloomberg (Mar. 20, 2026), https://www.bloomberg.com/news/articles/2026-03-20/softbank-plans-giant-ohio-ai-data-center-powered-by-gas-plants; see also Nicol Turner Lee & Darrell M. West, The Future of Data Centers, Brookings Inst. (Nov. 5, 2025), https://www.brookings.edu/articles/the-future-of-data-centers.

[43] Konstantin Pilz & Lennart Heim, Compute at Scale: A Broad Investigation into the Data Center Industry, arXiv:2311.02651 (Nov. 22, 2023), https://arxiv.org/abs/2311.02651.

[44] Int’l Energy Agency, World Energy Outlook 2025 (2025), https://www.iea.org/reports/world-energy-outlook-2025.

[45] Jingxi Liang, Dong Chen & Shiyu Xu, Energy-Constrained Optimization of Data Center Layouts: An Integer Linear Programming Approach, 18 Energies 5040 (2025), https://doi.org/10.3390/en18185040.

[46] Khaled B. Letaief, Yuanming Shi, Jianmin Lu & Junshan Lu, Edge Artificial Intelligence for 6G: Vision, Enabling Technologies, and Applications, 40 IEEE J. on Selected Areas in Commc’ns 5 (2022), https://doi.org/10.1109/JSAC.2021.3126076.

[47] Modi et al., supra note 41.

[48] Raul Katz, Juan Jung, Fernando Callorda & Ramiro Valencia, Assessing the Economic Value of Wi-Fi in the United States 71–78, Telecom Advisory Servs. (Sept. 2024), https://wififorward.org/wp-content/uploads/2024/09/Assessing-the-Economic-Value-of-Wi-Fi.pdf.

[49] Edward Oughton & Zoraida Frias, The Cost, Coverage and Rollout Implications of 5G Infrastructure in Britain, 42 Telecomm. Pol’y 636 (2018), https://doi.org/10.1016/j.telpol.2017.07.009.

[50] Kristian Stout & Spencer Kahn, Comments of ICLE In the Matter of Accelerating Wireline Broadband Deployment by Removing Barriers to Infrastructure Investment, WC Docket No. 17-84, Fed. Commc’ns Comm’n (2022), https://laweconcenter.org/resources/comments-of-the-international-center-for-law-economics-in-the-matter-of-accelerating-wireline-broadband-deployment-by-removing-barriers-to-infrastructure-investment.

[51] See, e.g., Eric Fruits & Kristian Stout, ICLE Comments to FCC on CTIA Petition for Rulemaking, CTIA Petition for Rulemaking, RM-12003, Fed. Commc’ns Comm’n (2025), https://laweconcenter.org/wp-content/uploads/2025/04/2025-CTIA-NEPA-Comments.pdf.

[52] Id.; see also Jeffrey Westling & Kristian Stout, Reply Comments of the International Center for Law & Economics, WT Docket No. 25-276, Fed. Commc’ns Comm’n (2026), https://laweconcenter.org/wp-content/uploads/2026/01/Build-America-Wireless-infrastructure-Comments.pdf.

[53] Westling & Stout, supra note 52.

[54] Accelerating Wireless Broadband Deployment by Removing Barriers to Infrastructure Investment, Notice of Proposed Rulemaking, FCC 25-67, WT Docket No. 25-276 (Sept. 30, 2025).

[55] Kristian Stout, Eric Fruits & Ben Sperry, ICLE Comments on the NTIA’s Proposed BEAD Alternative Broadband Technology Guidance, Nat’l Telecomms. & Info. Admin. (2024), https://laweconcenter.org/resources/icle-comments-on-the-ntias-proposed-bead-alternative-broadband-technology-guidance.

[56] Wireline Competition Bureau, Rural Digital Opportunity Fund Bid Defaults Announced, Public Notice, DA 21-910 (rel. July 26, 2021), https://docs.fcc.gov/public/attachments/DA-21-910A1.pdf.

[57] Nat’l Telecomms. & Info. Admin., BEAD Program Restructuring Policy Notice (June 2025), https://broadbandusa.ntia.doc.gov/news/latest-news/bead-restructuring.

[58] Barry Eichengreen, Romain Lafarguette & Arnaud Mehl, Cables, Sharks and Servers: Technology and the Geography of the Foreign Exchange Market, Nat’l Bureau of Econ. Rsch., Working Paper No. 21884 (2016), https://www.nber.org/papers/w21884.

[59] Katz et al., supra note 48.

[60] Exec. Order No. 14,179, Removing Barriers to American Leadership in Artificial Intelligence, 90 Fed. Reg. 8,741 (Jan. 23, 2025), https://www.federalregister.gov/documents/2025/01/31/2025-02172/removing-barriers-to-american-leadership-in-artificial-intelligence.

[61] Exec. Order, Ensuring a National Policy Framework for Artificial Intelligence, White House (Dec. 11, 2025), https://www.whitehouse.gov/presidential-actions/2025/12/eliminating-state-law-obstruction-of-national-artificial-intelligence-policy.

[62] Dani Rodrik, Industrial Policy for the Twenty-First Century 3–11, John F. Kennedy Sch. of Gov’t, Faculty Rsch. Working Paper No. RWP04-047 (2004), https://drodrik.scholar.harvard.edu/publications/industrial-policy-twenty-first-century.

[63] See, e.g., Scott Wallsten, The Effects of Government-Industry R&D Programs on Private R&D: The Case of the Small Business Innovation Research Program, 31 RAND J. Econ. 82 (2000).

Executive Summary

This issue brief examines the Brazilian Administrative Council for Economic Defense’s (CADE) decision to uphold interim measures against Apple’s App Store rules for app distribution and in-app payments on iOS. Although the case settled in December 2025, the May 2025 decision remains CADE’s clearest statement on competition in digital ecosystems.

The brief argues that the Decision rests on fragile foundations. CADE defined the relevant market as “the non-licensable smart mobile operating system iOS,” effectively making Apple dominant within its own platform by construction. That single-brand market definition sidesteps the core antitrust inquiry: whether consumers and developers have meaningful substitutes. They do. Apple competes with Android-based devices across hardware, software, privacy, security, services, artificial intelligence, and overall user experience.

The brief also argues that “ecosystem theory” adds little beyond established doctrines of tying, leveraging, foreclosure, and discrimination. As applied here, it risks treating ordinary platform governance and vertical integration as evidence of dominance and harm, without sufficient proof of market power or consumer injury. The Decision also falls short on evidence. CADE relied heavily on foreign precedent and academic literature, but did little to test its theory against Brazilian market conditions. It also gave insufficient weight to Apple’s procompetitive justifications, including privacy, security, fraud prevention, curation, quality control, and reduced transaction costs.

The remedies underscore the concern. The interim measures and settlement required Apple to open parts of its distribution and payment architecture, while specifying fee structures and implementation terms. That resembles sector-specific regulation more than ordinary antitrust relief.

At the same time, the case undercuts calls for new ex ante digital-market regulation in Brazil. CADE’s existing tools proved powerful enough to address the alleged harms. The Apple case is therefore less a vindication of ecosystem theory than a warning about its limits: antitrust should protect competition and consumers, not redistribute platform rents.

I.                    Introduction

This brief critiques the decision by Brazil’s Administrative Council for Economic Defense (CADE) to uphold interim measures against Apple’s rules for app distribution and in-app payments on iOS. CADE found that Apple’s App Store rules—especially mandatory use of Apple’s in-app purchase (IAP) system, “anti-steering” provisions, and limits on alternative app stores—likely amounted to an abuse of dominance in a market the Tribunal defined as “the non-licensable smart mobile operating system iOS.” The interim measures required Apple to permit third-party app stores, sideloading, alternative in-app payment systems, and external links to other purchasing options. Although the case ended in a December 2025 settlement, the Tribunal’s May 2025 decision remains CADE’s clearest statement on competition in digital ecosystems. Its reasoning may shape future Brazilian enforcement and serve as a reference abroad.

The Tribunal deserves credit for engaging seriously with the literature and case law on digital ecosystems. But the Decision rests on fragile analytical foundations. Its single-brand market definition effectively makes Apple dominant within its own platform by construction. And its ecosystem theories of harm—despite their descriptive sophistication—add little to established doctrines of tying, leveraging, and foreclosure. The Decision also falls short of the evidentiary standard appropriate for abuse-of-dominance findings, even at the interim-measures stage. It leans heavily on foreign precedent while giving insufficient weight to the procompetitive benefits of Apple’s vertically integrated model, particularly for privacy, security, and quality control.

Paradoxically, the case underscores the adequacy of Brazil’s existing competition framework. Law 12,529/2011 and CADE’s current enforcement tools proved sufficient to address the alleged harms. That weakens, rather than strengthens, the case for new ex ante digital-market regulation, including Bill 4,675/2025[1] and the more recent Bill 882/2026, both of which target ecosystem access more directly.[2] The Apple case in Brazil is less a vindication of ecosystem theory than a demonstration of its limits—and of the continuing need for rigorous economic analysis in competition enforcement.

The brief proceeds as follows. Section II provides the case’s factual and procedural background. Section III examines “ecosystem theory” as a competition-law framework, arguing that its core concepts largely repackage established antitrust doctrines without adding analytical value. Section IV turns to CADE’s application of that framework, focusing on its single-brand market definition for iOS app distribution. Section V examines the Decision’s theory of harm, arguing that it falls short of the evidentiary standard required in abuse-of-dominance cases, even for interim measures, and relies heavily on foreign case law while giving limited attention to the challenged conduct’s procompetitive effects. Section VI assesses broader implications for Brazilian competition law and digital regulation. Section VII concludes.

II.                    Background of the Case

In January 2023, Brazil’s Administrative Council for Economic Defense (CADE) opened an administrative inquiry into Apple for alleged abuse of dominance in “the market for the distribution of apps for iOS devices.” The investigation focused on Apple’s rules for app distribution and in-app payments on iOS.[3]

The case arose from a complaint by Mercado Livre—the leading e-commerce platform in Latin America—which alleged that Apple’s App Store policies barred developers from using alternative payment systems and tied access to iOS users to Apple’s proprietary in-app purchase (IAP) system. Apple requires developers that offer in-app purchases to use IAP, for which it charges commissions of up to 30%.

Mercado Livre also challenged Apple’s alleged use of discriminatory contractual terms. It argued that Apple restricted the distribution and promotion of third-party digital services and imposed so-called “anti-steering” provisions that barred developers from telling users about alternative payment options outside the app.

According to Mercado Livre, these practices restricted competition in digital payments, limited developer autonomy, and reinforced Apple’s dominance. At first glance, those allegations may appear to support antitrust intervention. A broader view suggests more skepticism. These disputes often concern how platform revenues are divided,[4] rather than competitive foreclosure that harms consumers and falls within antitrust law’s traditional scope. Like Spotify in the European Union proceedings[5] or Epic Games in U.S. litigation,[6] Mercado Livre appears primarily to challenge Apple’s control over parts of the value chain through which Mercado Livre could further monetize its own complementary services. Whether antitrust law should resolve such disputes is the central question of this brief.

This section provides the case’s factual and procedural background.

In October 2023, Mercado Livre asked CADE to impose interim measures requiring Apple to allow the sale of third-party digital goods and services through iOS or, alternatively, to permit developers to tell users about external offers and link them to alternative purchasing channels.

On Nov. 25, 2024, CADE’s General Superintendence (SG) instituted Administrative Proceeding No. 24/2024[7] and granted interim measures largely reflecting Mercado Livre’s requests. The SG ordered Apple to suspend enforcement of key contractual provisions supporting its anti-steering rules, mandatory use of IAP, and App Store exclusivity. In practical terms, the order required Apple to allow developers to inform users about alternative purchasing options; include external links and call-to-action mechanisms; use competing in-app payment systems; distribute iOS apps outside the App Store, including through sideloading and third-party app stores; and decouple App Store distribution from mandatory use of IAP, even for digital goods.

The SG also barred Apple from adopting functionally equivalent restrictions, required the company to make the necessary technical changes in Brazil, and ordered Apple to notify developers and publicly disclose the decision. To enforce compliance, the SG imposed daily fines of R$250,000, or about US$46,000. Apple had 20 days to implement the required changes in Brazil.

According to the SG, the challenged conduct involved two sets of practices: first, artificial barriers to entry and expansion in app distribution and digital goods within iOS, including mandatory use of Apple’s IAP system; and second, tying app-distribution services to Apple’s payment-processing system. Central to the SG’s analysis was a narrow, ecosystem-specific market definition that closely tracked Mercado Livre’s arguments.

The SG also relied heavily on foreign precedent. It cited the European Union’s Google Android[8] and Apple Music streaming decisions,[9] as well as the U.S. cases Epic Games, Inc. v. Apple Inc.[10] and Epic Games, Inc. v. Google LLC.[11] Apple appealed administratively, arguing that its integrated system protected privacy and security and that neither law nor economics justified forcing it to redesign its platform.

Apple also sought judicial review. A civil court suspended the interim measures, finding them “disproportionate and unnecessary” and imposed “without any demonstration of urgency or concrete risk to the market.”[12] CADE appealed, and an appellate judge partially reinstated the measures pending further review.

In February 2025, CADE held a public hearing on competition issues involving Apple’s iOS and Google’s Android mobile ecosystems.[13] Representatives from industry, civil society, and academia participated.[14]

In May 2025, CADE’s Tribunal upheld the interim measures, concluding that Apple’s conduct could restrict competition in app distribution and digital payments.[15] The proceeding quickly became one of Brazil’s highest-profile digital-platform cases, in part because it mirrored disputes Apple had faced in the United States, Europe, Latin America, and Asia.[16]

The Decision drew attention not only because of its outcome—which effectively opened Apple’s “walled garden”—but also because of its reasoning. The Tribunal described Apple’s business model as a “closed and vertically integrated ecosystem,” emphasizing that developers were “compelled” to rely on the App Store and IAP, while consumers were effectively “locked in” by switching costs. On that basis, CADE defined the relevant market as “the non-licensable smart mobile operating system iOS.”

That definition built on—and extended—the European Commission’s distinction in Google Android between “licensable” and “non-licensable” mobile operating systems.[17] As discussed below, that framing largely ignores the competitive constraints imposed by broader competition in smartphone-device markets.[18] Within the Tribunal’s defined market, Apple was necessarily the sole supplier and therefore dominant by definition.

On the merits, the Tribunal concluded that:

[T]here is strong plausibility in the contention of the unlawfulness of the tying practiced by Apple through the imposition of the IAP. The analysis of economic data and the submissions made by market participants demonstrates that this practice constitutes a sophisticated strategy of intra-product discrimination that disproportionately extracts value from specific developers, notably those of games and streaming applications.[19]

The Decision also marked the first time CADE expressly adopted what it called an “ecosystem theory of harm.”[20] The Tribunal reasoned that Apple’s integration across devices, operating systems, and services could let it foreclose competition in adjacent markets, such as payments. As discussed below, the analytical shortcuts embedded in that framework raise more questions than they answer.

Commissioner Gustavo Augusto Freitas de Lima issued a concurring opinion[21] endorsing the interim measures but rejecting the Tribunal’s market definition. He rejected “from the outset the thesis of the so-called ‘monopoly of oneself,’”[22] arguing that product differentiation and exclusivity alone do not create monopoly power. Freitas de Lima instead proposed a broader market for mobile-app stores, in which Apple and Google operate as a duopoly. Under that framework, Apple’s market power would stem not from exclusivity alone, but from revenue leadership and the purchasing power of its users.

Freitas de Lima also characterized the case as a conventional tying and leveraging dispute, drawing an explicit analogy to CADE’s 1993 Xerox precedent. He further cautioned that the European Union’s Digital Markets Act “does not apply in Brazil and has not been incorporated into the Competition Act.”[23]

In July 2025, the SG formally recommended that CADE convict Apple.[24] Before the Tribunal could issue a final decision, the parties settled. On Dec. 29, 2025, Apple entered into a Termo de Compromisso de Cessação (TCC),[25]  or cease-and-desist agreement. Under the TCC, Apple agreed to allow third-party app stores on iOS devices in Brazil, permit alternative payment processors for in-app purchases, and remove anti-steering restrictions that prevented developers from linking users to external offers.

The settlement may represent one of the broadest outcomes in global app-store antitrust enforcement to date.[26]

III.                    ‘Ecosystem Theory’ and Competition Law

The term “ecosystem” has clear rhetorical appeal. As a metaphor, it captures the interdependence among platform operators, business users, and consumers. It also helps describe how products and services can form broader networks of complementarities and dependencies. But its value in competition law remains doubtful. So far, ecosystem theory has added little to our understanding of market dynamics or to the methodological rigor of antitrust enforcement. In some respects, it may distort the analysis.

As Giuseppe Colangelo argues, the ecosystem concept has succeeded rhetorically without offering much analytical novelty.[27] Despite its frequent use in policy debates, the term lacks a precise legal definition and has not produced a coherent economic framework distinct from established theories of multi-sided markets. What it mostly offers is a convenient label for the scope and integration of large digital firms, not a new way to understand competition. The underlying issues—network effects, data synergies, interoperability, multi-homing, and indirect network effects—were already well understood through platform economics and existing antitrust doctrine.[28]

Colangelo’s review of European case law reinforces the point. He finds that “although a significant number of decisions potentially involve digital ecosystems, the term itself rarely appears and, when it does, it is not used to describe a clear or distinctive scenario.”[29] Even the European Union’s Digital Markets Act (DMA), which targets “contestability” and “fairness” in digital markets, refers to ecosystems only in passing and gives the term no operational meaning.[30]

The problem with ecosystem theory is not merely that it adds little. It can also distort the analysis. As Christopher Yoo observes, “it remains unclear what a concept of ecosystems defined in terms of complementary products adds to the principles governing vertical integration.”[31] The ecosystem framework risks obscuring what decades of economic literature have consistently shown: vertical integration and vertical restraints are often procompetitive and beneficial to consumers.[32]

Yoo further notes that ecosystem theory depends on a platform occupying a “central position” within an ecosystem—a concept related to, but distinct from, dominance.[33] That framing risks turning market success into evidence of liability:

Defining harm in terms of economic significance suffers from significant endogeneity problems. Achievement of a central or important position can be the result of either successful competition on the merits or anticompetitive activity. Using those outcomes as a basis for liability risks treating causes as effects.[34]

The concern is that the ecosystem framework may encourage suspicion toward business practices that are often efficient. Modern antitrust law has largely moved away from presumptions against vertical restraints, recognizing that such arrangements require context-specific analysis. As the Competition Committee of the Organisation for Economic Co-operation and Development’s (OECD) Directorate for Financial and Enterprise Affairs explained:

The ambiguous effect of vertical restraints, and the consequence that one cannot simply outlaw certain vertical restraints and permit others, suggests that a rule of reason (as opposed to per se rule) approach is preferred. Economic analysis in this regard can help to determine which types of agreements are capable of and likely to raise competition concerns, while it can also help to understand what types of justifications could be judged to outweigh identified anti-competitive effects.[35]

Proponents of ecosystem theory do not expressly advocate per se illegality. But when the framework relies on presumptions, circular findings of dominance, or generalized suspicion toward integration, it can produce much the same result.

Applied this way, ecosystem theories of harm may also chill innovation. Integrated design—the seamless interaction among hardware, software, and services—often creates substantial consumer benefits, especially in security, privacy, and quality control.[36] Treating that integration as suspect because it creates “lock-in” could deter firms from pursuing efficiency-enhancing complementarities. The result would revive a structural hostility toward vertical integration that modern antitrust law, after decades of economic learning, had largely abandoned.

IV.                    CADE’s Market Definition and Theory of Harm

As argued above, ecosystem theory—while descriptively useful—risks becoming a doctrinal shortcut. It can encourage excessively narrow market definitions and overstate business users’ “dependencies” on a platform, lowering the evidentiary bar for finding dominance and abuse. This section asks whether that risk materialized in CADE’s treatment of Apple. It appears to have done so. Although the Decision concerned interim measures,[37] which naturally involve a lower evidentiary threshold, CADE appears to have lowered that threshold too far,[38] adopting premises that could distort future antitrust enforcement.

Before addressing the merits, the Decision introduces a “foundational semantic framework regarding the operational structure of mobile digital ecosystems,”[39] describing ecosystems as a “‘new’ explanatory lens for the competitive dynamics of digital markets in which large technology companies exploit business models based on multiple layers of intermediation.”[40]

According to the Tribunal:

The economic logic of ecosystems has relevant implications for antitrust policies. Firstly, ecosystems are characterized by new forms of internal competition, manifested as “vertical competition” between products that are not directly substitutable. This phenomenon calls into question the centrality of the notion of the “market” in antitrust analysis, since the delimitation of spaces of substitutability where competitive forces mutually constrain each other becomes complex, as does the apprehension of anti-competitive effects in interconnected economic activities.

Secondly, there is a strengthening of economic dependencies between the complementors and the central platform orchestrator, the latter being able to unilaterally alter the conditions of competition in the various adjacent markets. While orchestration is crucial for coordination in the ecosystem, design and governance decisions can generate functional and distributive failures, resulting in a loss of value for all participants and hindering innovation. The framing of these decisions under competition laws blurs the boundaries between the exclusionary and exploitative nature of abusive conduct.[41] [Emphasis ours, citation omitted].

The Tribunal’s framing rests on two propositions. First, ecosystems supposedly involve new forms of “vertical competition” that require rethinking market definition in digital-platform cases, effectively narrowing the unit of analysis to the platform itself. Second, “economic dependencies” between complementors and the platform owner allegedly allow the latter to “unilaterally alter the conditions of competition.” Both propositions are problematic.

First, the suggestion that digital ecosystems diminish the relevance of traditional market definition is a non sequitur. A digital ecosystem—whether understood as “an organization that sells interconnected products or services” or as a multi-sided platform[42]—is not inherently more complex than a “traditional” market. Complexity depends on the conduct at issue and the market under examination. Some digital-platform cases are complex; the Google AdTech litigation is an obvious example.[43] But digital markets are not uniquely so. A case involving a social-media platform or e-commerce marketplace is not obviously more complex than one involving health insurance, payment networks, or financial exchanges. Moreover, complexity does not justify analytical shortcuts.

Rigorous competition analysis still requires the same core inquiry: who exercises market power, against whom, and whether that power can be exercised profitably and sustainably. Multi-sided-platform economics provides tools for conducting precisely that analysis. As the U.S. Supreme Court recognized in Ohio v. American Express, platforms that facilitate transactions between different user groups must be analyzed as integrated markets that account for competitive dynamics on all sides.[44] That does not require abandoning traditional market definition. It requires applying it correctly.

The purpose of market definition remains straightforward: identifying the competitive constraints a firm faces through demand substitution, supply responses, and potential entry. That inquiry remains indispensable in digital-platform cases. Without it, there is no principled basis for distinguishing firms with substantial market power from firms that operate successful businesses in competitive markets.

Second, the Tribunal’s emphasis on “economic dependencies” is equally problematic. Dependency is inherently contextual, and treating it as a starting assumption risks creating a self-fulfilling theory of market power.[45] The fact that a platform operator can set the terms for access to its ecosystem does not establish market power or meaningful dependency by business users. In many cases, business users multi-home precisely to reduce dependence on any single platform.[46]

What the Tribunal describes—a platform setting rules for access to its infrastructure—is ordinarily an exercise of property rights and contractual freedom. Every private marketplace, from shopping malls to stock exchanges, sets rules for participation. The relevant antitrust question is not whether a platform controls access to its infrastructure; it inevitably does. The question is whether users had viable alternatives when they chose to participate.

That question matters here because CADE never established the market power needed to support its theory. The Decision correctly notes that Apple users must use the App Store to install apps and Apple’s payment system for digital purchases.[47] Because iOS is a “closed” system,[48] users ostensibly have “no choice” but to use Apple’s infrastructure. Except they do have a choice: consumers can buy different smartphones, and developers can prioritize different platforms.

As Craig Conrath explains:

[I]t is necessary to distinguish between contract power and monopoly power, between post-contractual opportunism and abuse of dominant position. … Therefore, whenever there is a claim of monopolistic practice by a party in an contractual relationship with the person accused of a monopolistic practice, it is important not to ask the question, “Does this party now have any other choices?” This is a question of contract law. Instead, one should ask the relevant antimonopoly law question: When the contract was made, did this party have other choices?[49]

Earlier in the Decision, the Tribunal surveyed foreign cases involving Apple’s App Store practices, purportedly to “establish objective parameters for legal comparability.”[50] To its credit, the Tribunal reviewed precedent thoroughly, including cases that cut against its position. Even so, much of this discussion functions more as rhetoric than analysis.

The Tribunal, for example, relies heavily on Judge Yvonne Gonzalez Rogers’ opinion in Epic Games v. Apple.[51] As the Decision notes, the court rejected Epic’s Sherman Act claims while still finding certain practices problematic under California law.[52] The Tribunal then quotes portions of the opinion suggesting Apple’s conduct could be “exploitative.”

That reliance is difficult to square with the actual holding. The relevant findings arose under California’s Unfair Competition Law,[53] not under the Sherman Act. That statute imposes a substantially less demanding standard than federal antitrust law’s rule-of-reason framework. Its relevance to a Brazilian abuse-of-dominance case is therefore far from obvious.

More importantly, the Tribunal’s reliance on Epic indirectly imports the logic of Eastman Kodak Co. v. Image Technical Services,^[54] where the U.S. Supreme Court accepted that a derivative aftermarket could constitute a distinct market if consumers were “locked in.” Judge Gonzalez Rogers cited Kodak in the passage quoted by CADE:

In the context of technology markets, the open flow of information becomes even more critical. As explained above, information costs can create a lock-in for platforms, since users have no information about the lifetime costs of an ecosystem. Users may also lack the ability to attribute costs to the platform rather than the developer, which further prevents them from making informed choices. In these circumstances, the ability of developers to provide information across platforms is crucial. Although Epic Games has not met its burden of demonstrating actual blocking on this record, the Supreme Court has recognized that such information costs can create the potential for anticompetitive exploitation of consumers. Eastman Kodak, 504 U.S. at 473-75, 112 S. Ct. 2072.^[55]

Subsequent U.S. case law has narrowed the circumstances in which single-brand aftermarket claims can succeed. Judge Gonzalez Rogers herself noted that the 5th U.S. Circuit Court of Appeals held in United Farmers Agents Ass’n v. Farmers Insurance Exchange[56] that such claims fail where customers “would clearly have become aware of [the alleged anticompetitive] policy long before they faced significant switching costs.”[57]

That principle cuts strongly against CADE’s position. Apple’s integrated distribution model—exclusive use of the App Store, mandatory IAP, and restrictions on alternative distribution channels—has been publicly known since the App Store launched in 2008. Developers and consumers therefore entered the ecosystem with advance knowledge of Apple’s rules. Where there is no significant information asymmetry at the time of the initial purchasing decision, the economic logic for recognizing a single-brand aftermarket largely disappears.

Benjamin Klein made a similar point in his critique of Kodak. Echoing Conrath,[58] Klein argues that market power must be assessed before users make platform-specific investments: “the level of competition should be measured before the buyer makes any specific investments. If the market at this time is competitive, then the tie is merely part of the total freely negotiated competitive price.”[59]

At this point, the Decision attempts to distance itself from foreign doctrine, emphasizing that the case rests “exclusively on the application of the Brazilian legal framework,” particularly Article 173, § 4 of the Constitution and Law 12,529/2011.[60] Formally, the alleged infringements involve familiar doctrines: tying through mandatory use of Apple’s IAP system and discriminatory application of the App Store Review Guidelines.[61]

CADE thus presents ecosystem theory not as the theory of harm itself, but as the broader analytical framework within which traditional theories operate. That characterization understates the work ecosystem theory performs in the Decision. The framework drives the single-brand market definition for “distribution of apps for iOS devices”—a market that could not exist without treating iOS as a self-contained competitive space. It also justifies remedies far broader than those adopted elsewhere, including mandatory opening of Apple’s entire distribution and payment architecture.

The relevant question should instead be whether developers and consumers had viable alternatives when they chose to develop for, or purchase, iOS devices. In conventional antitrust terms, the question is whether Apple’s App Store faces meaningful substitutes.

The Tribunal approaches this issue by examining substitutability from the perspective of device manufacturers, developers, and consumers. In each case, it concludes that iOS and Android are not substitutes. The analysis is flawed at every stage.

The Decision first observes that “for mobile device manufacturers, it would not be possible to address substitutability between licensable and non-licensable systems.”[62] That observation is true but analytically irrelevant. Apple’s refusal to license iOS reflects a business strategy built around vertical integration among hardware, software, and services. By choosing that model, Apple gives up licensing revenue to compete through differentiated products and user experience. That does not insulate it from competition.

Apple competes not only with Google’s Android ecosystem, but also with device manufacturers such as Samsung, Xiaomi, OPPO, vivo, and Huawei. As the district court observed in Epic v. Apple, “it is illogical to argue that there is a market for something that is not licensed or sold to anyone. Competition exists for smartphones, which are more than just the operating system.”[63] Consumers buy smartphones, not operating systems.

That point has direct implications the Tribunal largely ignores. Competition at the device level constrains conduct at the operating-system level. If Apple used its App Store rules to extract excessive rents from developers, the resulting deterioration in app quality, availability, or pricing would weaken the iPhone’s competitiveness against Android-based devices.

The Tribunal’s distinction between “licensable” and “non-licensable” operating systems therefore obscures the competitive dynamics at issue. Apple’s integrated model and Google’s licensing-based model are competing strategies for attracting users, developers, and manufacturers to mobile platforms. That rivalry has generated substantial innovation and consumer benefits. As David Evans explains:

Innovations by Apple and Google, in particular, have led to spread of smart mobile phones around the world, enabling billions of people to consumer Internet-based services and millions of businesses to provide mobile-app based services to them. Apple introduced the iPhone, which consisted of a powerful computer, a mobile operating system, and a standard set of applications including a mobile browser in June 2007. Google invested in developing a mobile operating system, Android, which it ran as an open-source project, and developing and organizing an ecosystem of handset makers, mobile network operators, and other technology partners. It introduced the first Android phone in October 2008. Apple and Google also stimulated the production of mobile apps by providing software tools for developing apps for their operating systems, creating a quality certification process for these apps, and creating “app stores” that provided centralized places for developers to distribute apps and for users to download them on their mobile devices.[64]

Competition between Apple and Google extends beyond operating systems. It includes hardware, software integration, AI capabilities, security, privacy, and the overall user experience. As ICLE observed in comments submitted to the UK Competition and Markets Authority:

The fierce competition between Apple and Google is waged through continuous improvements in camera technology, the race for superior processing power (with Apple’s A-series chips and Google’s Tensor chips forming the basis of their respective performance claims), and divergent—yet intensely competitive—approaches to user security and privacy, which have become central pillars of each company’s marketing and value proposition.

Most recently, this rivalry has shifted to the next frontier of innovation: the deep integration of artificial intelligence into the core user experience, with the corollary that both firms have invested massively invested in AI technology (and continue to do so). To observe this landscape of ceaseless innovation and declare it a market with “limited competition” is to misunderstand its fundamental nature.[65]

One recent example is Google’s rollout of Gemini “task automation,” initially available only on recent Pixel and Samsung devices.[66]

By treating these business models as non-substitutable, the Tribunal excludes one of the most important constraints on Apple’s conduct: interbrand competition between mobile ecosystems.

The Tribunal reaches similarly flawed conclusions from the perspective of developers. It argues that iOS and Android are not substitutes because app development differs across systems, migration is difficult, and developers typically operate on both platforms.[67] Ironically, the Tribunal treats multi-homing as evidence of non-substitutability. In reality, multi-homing reduces dependence on Apple. Developers operate on both platforms precisely because they can reach consumers through alternative channels. Android-only development remains commercially viable, especially in Brazil, where Android accounts for more than 80% of smartphones.[68]

From the consumer perspective, the Tribunal concludes that “there would be no substitutability” because operating systems are tied to devices and Android phones generally cost less than iPhones in Brazil.[69]

The Tribunal identifies real differences but draws the wrong conclusion. Consumers choose smartphones based on many characteristics, including operating systems, app ecosystems, cameras, design, connectivity, and AI features. Apple therefore competes not only with Google, but with manufacturers such as Samsung, Xiaomi, Huawei, and Oppo.[70] That fact supports a broader, not narrower, market definition.[71]

The Tribunal also emphasizes iPhones’ higher average prices.[72] Economic theory recognizes that products in different quality or price tiers can constitute separate markets where cross-elasticity of demand is low. But that conclusion requires evidence, not mere price differences.[73]

Android manufacturers increasingly compete directly with Apple in the premium-smartphone segment. Devices priced above US$600 represented 25% of global smartphone unit sales in 2024[74] and more than 60% of industry revenue.[75] Premium-segment sales continued growing through 2025,[76] with Samsung, Huawei, and Google gaining share against Apple.[77] Samsung’s Galaxy S26 Ultra, Google’s Pixel 10 Pro XL, and Apple’s iPhone 17 Pro Max launch at comparable price points,[78] undermining the notion that Apple occupies a uniquely expensive segment insulated from competition.[79]

The Decision also notes that “only a portion of consumers can afford to replace Android with iOS.”[80] But competition occurs at the margins. Effective competitive constraints do not require every Android user to switch to Apple, or vice versa. They require only enough marginal consumers to switch in response to price increases or quality degradation to discipline conduct. As Simon Baker and Laurence Wu explain:

Customers who can easily switch from one product to another and who would do so in response to a price increase are known as marginal customers, whilst those who are unlikely to do so (…) are termed infra-marginal customers. Infra-marginal customers are protected by the willingness of marginal customers to switch so long as there are enough marginal customers and so long as sellers cannot effectively identify the marginal from the infra-marginal customers…

The ability of these customers to switch just a portion of their total purchases from one product to another can be as effective in disciplining pricing as the complete switching of purchases by a few customers.[81]

The Decision also largely ignores substitution from iOS to Android, even though such switching may be easier given Android’s lower average prices. More importantly, CADE never analyzes actual switching rates in Brazil.

Available evidence from other jurisdictions strongly suggests dynamic competition between mobile ecosystems. As ICLE explained:

Consumers can readily find myriad comparisons of Android and iPhone devices online. Moreover, both Apple and Google maintain webpages that offer to help users switch from one platform to the other. The business press has extensively covered the fierce rivalry between the two companies. And numerous academic studies have reached similar conclusions about the nature of their competition. Nicolas Petit refers to Apple and Google as “moligopolists,” while David Evans has described their rivalry as “dynamic competition.” Marshall Van Alstyne and his coauthors have analyzed the strategies that both Google and Apple have deployed to outcompete one another. (Citations omitted).[82]

Dirk Auer made a similar point in criticizing the European Commission’s Google Android decision.[83] He noted that even the Commission’s own data showed substantial switching and contestability:

Take the claim that 82% of Android users stick with Android when they change phones (compared to 78% for Apple), and that 75% of new smartphones are sold to existing users. The Commission asserted, without further evidence, that these numbers prove there is little competition between Android and iOS.

But is this really so? In almost all markets consumers likely exhibit at least some loyalty to their preferred brand. At what point does this become an obstacle to interbrand competition? The Commission offered no benchmark mark against which to assess its claims.

And although inter-industry comparisons of churn rates should be taken with a pinch of salt, it is worth noting that the Commission’s implied 18% churn rate for Android is nothing out of the ordinary, including for industries that could not remotely be called anticompetitive.

To make matters worse, the Commission’s own claimed figures suggest that a large share of sales remained contestable (roughly 39%). Imagine that, every year, 100 devices are sold in Europe (75 to existing users and 25 to new users, according to the Commission’s figures). Imagine further that the installed base of users is split 76–24 in favor of Android. Under the figures cited by the Commission, it follows that at least 39% of these sales are contestable. [Emphasis added].

Likewise, a survey commissioned by the UK Competition and Markets Authority found that 8% of iOS users had switched from Android, 5% of Android users had switched from iOS, and additional users on both sides considered switching. Those are meaningful levels of substitution.[84] Yet the Decision analyzes no comparable data for Brazil.

This omission matters because operating systems are multi-sided platforms. As David Evans and Michael D. Noel explain, prices and competitive effects on one side of a platform cannot be analyzed independently from the other.[85] Even if developers face switching costs, Apple’s market power remains constrained so long as consumers can switch between ecosystems. Conversely, even if consumers face switching costs, Apple’s treatment of developers is constrained by the risk that restrictive terms could push developers toward Android-first or Android-only strategies, reducing the attractiveness of the iOS ecosystem.

The Tribunal briefly acknowledges switching costs arising from interoperability and compatibility preferences, noting that consumers prefer compatible smart devices and may therefore become “locked in.”[86] That observation has merit. But the Decision never examines actual substitution patterns in Brazil, nor does it adequately consider that interoperability itself is a dimension of competition that can enhance consumer welfare.

Later, the Tribunal appears to endorse the SG’s view that Apple’s ecosystem design itself demonstrates market power because Apple controls “all the rules and conditions of competition” within iOS.[87] This reasoning risks becoming circular. Every platform owner controls participation rules on its platform. That fact alone does not establish dominance. Market power depends on whether users had viable alternatives when deciding whether to participate. As discussed above, they did.

The circularity becomes explicit in the Tribunal’s conclusion that Apple “has market power in the form of the ability to unilaterally alter the conditions of competition in the markets related to its smart mobile digital ecosystem.”[88] In other words, Apple is dominant because it controls its own ecosystem.

This problem did not escape notice within the Tribunal itself. As discussed above,[89] Commissioner Freitas de Lima rejected “from the outset the thesis of the so-called ‘monopoly of oneself,’” arguing that product differentiation and brand exclusivity do not themselves create monopoly power. Whatever the merits of his proposed alternative market definition, his critique mirrors the central problem identified here: Apple’s market power cannot be established by definitional fiat.

A broader market definition would better align with established antitrust methodology and with cases such as Epic Games v. Apple, where the court rejected a single-brand iOS market and instead analyzed competition across mobile platforms.[90] Similarly, the U.S. Department of Justice recently defined a market for “performance smartphones” in its case against Apple.[91] Even that narrower framework—which still includes competing devices such as Samsung’s Galaxy S26 Ultra and Google’s Pixel 10 Pro XL—is substantially broader than CADE’s single-brand market.

The distinction matters because Apple’s position in Brazil looks very different under a broader market definition. Apple’s share of Brazilian smartphone sales is estimated at roughly 15-20% by units sold—well below levels typically associated with dominance under Law 12,529/2011.[92] The gap between Apple’s actual market position and the Tribunal’s dominance finding illustrates the consequences of adopting an excessively narrow market definition.

A careful reading of the Decision suggests that CADE could have pursued its case without invoking ecosystem theory at all. The problem is not merely that ecosystem theory was unnecessary. It is that the framework pushed the analysis toward a tautological finding of dominance based on Apple’s control over its own platform.

V.                    Evidentiary Standards and the Limits of Ex Post Enforcement

As discussed in the previous section, the Tribunal’s single-brand market definition for iOS app distribution and in-app payments lacks support under established antitrust methodology. That defect alone should substantially weaken the case: absent dominance, the vertical agreements governing the App Store and Apple’s IAP system cannot constitute abuse of dominance. Even so, CADE’s characterization of Apple’s conduct as exclusionary discrimination (“self-preferencing”) and unlawful tying warrants closer examination.

The Tribunal’s decision upholding interim measures against Apple combines ecosystem theory with traditional antitrust doctrines, including tying, self-preferencing, and foreclosure of rival distribution channels. Commissioner José Levi Mello do Amaral Júnior Fernandes presents these theories as mutually reinforcing: ecosystem theory supplies the descriptive framework, while conventional antitrust doctrines provide the legal structure. This section argues that the Decision falls short of the evidentiary standard that abuse-of-dominance cases should require, even at the interim-measures stage.

Interim measures are valuable enforcement tools, particularly in digital markets where proceedings often move more slowly than the markets themselves. But they still require reasonably robust evidence, especially where the remedy would radically alter market structure. Here, the Decision relies primarily on theories and evidence from foreign cases, rather than Brazil-specific evidence. It also gives remarkably little attention to the procompetitive effects of the conduct it condemns.

That omission is striking because the Decision itself acknowledges the evidentiary gap. At several points, the Tribunal concedes that definitive conclusions would require “supplementing the evidentiary investigation” and “a more detailed analysis of the competitive conditions” in the relevant markets.[93]

The reliance on foreign material is structural, not incidental. After describing Mercado Livre’s claims and testimony from the Public Hearing on Mobile Digital Ecosystems[94]—much of it from complementors raising concerns similar to Mercado Livre’s—the Tribunal simply asserts that “these practices would have the effect of impairing competition and restricting the promotion of alternatives, while Apple’s own apps do not face the same costs arising from these bans.”[95] The Decision provides no concrete evidence of actual or likely harm to competition in Brazil.

The tying analysis in Section 5.3.4 is similarly thin. Even assuming the Tribunal correctly concluded that Apple’s operating system and IAP system are “separate products,”[96] the evidence of competitive harm again comes almost entirely from theory and foreign precedent. The Decision relies on the U.S. Supreme Court’s Jefferson Parish framework, the European Commission’s Article 82 Guidance,[97] and academic literature describing how tying arrangements can reduce welfare.[98]

The Brazilian precedents cited—Comgás, PTI, Steel Placas, and the rapporteur’s opinion in Google Shopping—do little more than establish that tying and discriminatory conduct can, in principle, violate Article 36 of Law 12,529/2011. None involved vertically integrated mobile ecosystems, and none supplied the empirical record the Tribunal purported to apply.

To be clear, competition authorities may properly rely on academic literature and foreign jurisprudence. Comparative analysis is often useful, especially in cases involving novel business models or market structures. The Tribunal deserves credit for engaging seriously with international debates and precedent.

The problem is that the Decision uses those authorities mainly to establish the theoretical plausibility of its theory of harm. Plausibility is not proof that challenged conduct harms competition in the relevant market. The Decision devotes substantial attention to what courts and scholars in the United States and Europe have said about Apple’s business model, while doing far less to test those propositions against Brazil’s competitive realities.

The Decision is also notably thin on procompetitive effects. Modern antitrust systems, including Brazilian competition law, generally require effects-based analysis that weighs anticompetitive risks against efficiencies before condemning vertical restraints. The Tribunal acknowledges that principle in the abstract,[99] including CADE precedent recognizing that bundled or joint sales can “reduce the impact of double profit margins and enable economies of scale and scope arising from joint production or distribution.” Yet the analysis never meaningfully returns to those efficiencies.

Instead, the 30% commission, App Store-IAP integration, and anti-steering rules are treated only as potential sources of harm. The countervailing benefits—security investment, fraud prevention, curation, cross-subsidization of the more than 80% of apps that pay no commission,[100] and reduced transaction costs for users and developers—receive little comparable attention. An interim measure that effectively restructures a multi-sided platform should, at minimum, rest on a record that seriously evaluates both sides of the competitive ledger.

Apple’s procompetitive justifications are substantial and have been recognized even by courts skeptical of other aspects of its conduct. In Epic Games v. Apple, Judge Yvonne Gonzalez Rogers concluded that Apple’s integrated approach generated real consumer benefits in privacy, security, and curation—benefits that operate as dimensions of interbrand competition against Android’s more open model.[101]

The economic literature on vertical integration in technology markets reaches similar conclusions. Integration among hardware, software, and services often reduces transaction costs, improves quality control, and internalizes externalities that arise in modular systems.[102]

Randal Picker identifies another dimension the Tribunal largely overlooks: security competition between ecosystems. Apple and Google compete through markedly different security architectures. According to Picker, forcing Apple to permit open app distribution would reduce security competition by pushing iOS toward Android’s more permissive—and demonstrably more vulnerable—model.[103]

The Tribunal itself recognizes that Apple competes partly through privacy and security, while Google emphasizes openness and advertising monetization.[104] Yet the Decision never fully confronts the implications of that observation. If Apple’s closed architecture is a differentiated competitive offering, mandating openness may reduce, rather than increase, consumer choice at the ecosystem level.

Implicit in the Decision is the notion that digital ecosystems should operate as “open squares”—spaces where platform operators cannot meaningfully restrict access. But that is not an economic conclusion. It is a regulatory preference. Some jurisdictions, most notably the European Union through the Digital Markets Act (DMA),[105] have adopted that preference expressly through legislation.

The relevant question, then, is whether ex post antitrust enforcement can properly reach the same result through abuse-of-dominance doctrine—and whether it should.

The Tribunal’s theory of harm also understates the costs of the remedies it contemplates. Requiring Apple to permit third-party app stores and alternative payment systems on iOS would necessarily alter the platform’s security architecture.[106] The Tribunal acknowledges the sideloading debate and Apple’s claims about malware differentials between iOS and Android.[107] But it ultimately dismisses those concerns as pretextual or addressable through less restrictive alternatives, without the detailed analysis such a conclusion requires.

The literature on forced interoperability suggests those costs are not trivial. Research by the Center for European Policy Analysis (CEPA) argues that interoperability mandates create new attack surfaces and increase opportunities for malicious actors, potentially undermining the end-to-end security associated with vertically integrated systems.[108] Picker likewise suggests that mandated openness would not merely alter Apple’s business model; it would eliminate a dimension of competition consumers value.[109]

Innovation incentives are also at stake. Sven Völcker and Daniel Baker argue that Apple’s vertically integrated model encourages continuous investment in platform improvements—including security updates, APIs, and app-review infrastructure—because Apple captures part of the value the ecosystem creates through commissions.[110] If forced openness reduces or eliminates that revenue stream, investment incentives may weaken.

That concern is not merely theoretical. Early experience under the DMA suggests that compliance with interoperability and openness mandates can divert substantial engineering resources away from innovation and toward regulatory compliance.[111]

VI.                    Broader Implications for Brazilian Competition Law

The Apple proceedings carry implications well beyond the parties involved. The case tests the boundaries of Brazil’s competition framework along three dimensions. Analytically, it raises whether the Tribunal’s market-definition methodology and evidentiary standards risk overenforcement, particularly in digital-platform cases involving vertical integration and ecosystem-based theories of harm. Institutionally, it raises whether Brazil’s existing ex post competition framework is sufficient for digital markets or whether the case instead supports a new layer of ex ante regulation. Remedially, it highlights the tension between antitrust enforcement and sector-specific regulation, especially where competition remedies become detailed and prescriptive.

These issues are closely related. Lower evidentiary standards and expansive theories of dominance can facilitate increasingly intrusive remedies, blurring the line between competition enforcement and ongoing regulation. The Apple case therefore serves both as a test of the analytical discipline imposed by the consumer welfare standard and as a case study in the institutional limits of competition law in fast-moving digital ecosystems.

A.      Error Costs and the Consumer Welfare Standard

Brazilian competition law, as codified in Law 12,529/2011, uses an effects-based, efficiency-sensitive framework rather than formalistic prohibitions. The statute defines anticompetitive conduct by reference to its object or effects and recognizes that efficiencies and consumer benefits may justify conduct or transactions that otherwise raise concerns.[112] Consistent with that framework, CADE’s decisional practice has generally focused on consumer-welfare variables, asking whether challenged conduct or mergers are likely to affect prices, output, quality, or innovation, and weighing potential harms against cognizable efficiencies.[113] That orientation appears in CADE’s guidelines and in Brazil’s submissions to the Organisation for Economic Co-operation and Development (OECD), which emphasize that enforcement remains focused on conduct’s economic effects on consumers, even when broader policy considerations enter the discussion.[114]

The Apple proceedings risk departing from that tradition. By defining a single-brand market, the Tribunal collapses the distinction between property rights and market power: any firm that controls its own platform becomes dominant within it by definition. That lowers the threshold for finding dominance. By invoking ecosystem theory—even as a framing device—the Tribunal also lowers the threshold for finding competitive harm, overstating lock-in while underweighting the well-documented benefits of integration, including quality, security, and innovation.[115]

This move is not costless. As Frank Easterbrook showed in his foundational error-cost analysis, false positives in antitrust enforcement—the mistaken condemnation of procompetitive conduct—impose costs that markets cannot easily correct.[116] False negatives may eventually be disciplined by entry and innovation. False positives, by contrast, can chill efficient conduct and deter beneficial innovation across markets operating under the shadow of precedent. Joshua Wright and Murat Mungan applied this insight to digital markets, concluding that platform cases warrant a stronger evidentiary standard than preponderance of the evidence, precisely because the costs of mistaken intervention are amplified in fast-moving sectors with novel business models and rapidly shifting competitive dynamics.[117]

Geoffrey Manne has likewise argued that erroneous enforcement is both more likely and more costly in digital markets, where “antitrust decisions are made in innovative, fast-moving, poorly understood, or novel market settings” and where “erroneous interventions against innovation and the business models used to deploy it threaten to deter subsequent innovation.”[118] The CADE proceedings illustrate that risk. The most exhaustive judicial treatment of the same conduct concluded that Apple’s relatively closed ecosystem generates significant procompetitive benefits in privacy, security, and quality assurance. Condemning that model—or restructuring it through interim relief before a final merits determination—risks exactly the kind of false positive the error-cost literature warns against: suppressing an efficient vertical-integration strategy based on an unsupported theory of dominance.

The implications for Brazilian competition law are significant. If authorities can establish dominance by defining a market in which the respondent is the sole supplier by construction, and infer competitive harm from ecosystem structure rather than rigorous economic analysis, the consumer welfare standard no longer meaningfully constrains enforcement discretion. It becomes an ex post rationalization rather than an ex ante analytical discipline—the very outcome the error-cost framework is meant to prevent.

B.       The Case Against New Ex Ante Regulation

There is, however, a second—and clearer—reading of the Apple proceedings. Whatever the Decision’s analytical shortcomings, the case demonstrates that Brazil’s competition framework—Law 12,529/2011, CADE’s institutional structure, and the procedural tools available to the General Superintendence and Tribunal—already provides the mechanisms necessary to address competition concerns in digital markets. CADE opened an investigation, developed a theory of harm, imposed interim measures, conducted adversarial proceedings with full due process, and ultimately resolved the dispute through a negotiated settlement that imposed structural remedies on a global technology firm.

That is a significant institutional achievement. It should inform Brazil’s ongoing legislative debate over digital-market regulation. Bill 4,675/2025, currently before Congress, would create a new Digital Markets Superintendence within CADE and establish an ex ante regulatory framework for firms designated as “systemically relevant” in digital markets.[119] Proponents argue that traditional ex post antitrust enforcement cannot adequately address structural features of digital platforms, such as network effects, switching costs, and data advantages, and that ex ante regulation is therefore necessary to prevent harm before it occurs.[120]

The Apple case complicates that narrative. It shows that CADE can intervene aggressively against a global platform using its current ex post tools, including interim measures that, whatever their analytical merits, produced in months what years of litigation failed to achieve elsewhere. The real question is not whether Brazil’s competition framework can address digital markets, but whether it should do so through the existing case-by-case adjudicatory model or through a new layer of ex ante regulation.

The latter approach carries substantial risks. As Dario Oliveira Neto argues, it could “subtly sweep consumer welfare under the rug” by introducing regulatory objectives—such as “protection of the competitive process” and “reduction of barriers to entry”—that may sound consistent with competition policy but, in practice, diverge from the consumer welfare standard in both logic and application.[121]

The more prudent path is to preserve the existing adjudicatory model while insisting on the analytical rigor the consumer welfare standard requires.[122] The Apple case should serve both as proof of concept and as a cautionary tale: proof that Brazil’s current framework can address digital-platform disputes, and a reminder that its legitimacy depends on the quality of the analysis applied within it.

Lowering evidentiary standards in the name of addressing novel market structures does not strengthen competition enforcement. It weakens the institutional credibility on which effective enforcement depends.

C.      Antitrust Remedies or Sectoral Regulation?

The Decision did not merely resolve a precautionary dispute. It also laid out a remedial blueprint that anticipated—and effectively invited—the highly prescriptive outcomes later incorporated into the December 2025 settlement. That matters because the analytical weaknesses identified above were not merely theoretical. They shaped the remedies ultimately imposed on Apple and may do the same in future cases involving digital platforms.

The Decision identified three “principal pillars” of relief: (i) removal of anti-steering provisions; (ii) unbundling payment processing from app distribution; and (iii) enabling alternative app distribution channels, including third-party app stores and sideloading.[123] Standing alone, none of these remedies is especially unusual. The first closely mirrors the injunction upheld in Epic Games v. Apple,[124] while the second and third resemble obligations imposed under the European Union’s DMA.[125] What distinguishes the Decision is its remedial ambition: it pursued all three simultaneously through a precautionary measure issued before any merits determination.

The Tribunal was not unaware of the tension. Commissioner José Levi Mello do Amaral Júnior Fernandes acknowledged that, if Apple’s distribution “monopoly” were maintained, “CADE would have to arbitrate the prices charged by the monopolist, which is typical of a regulatory body and outside the regime of Law 12,529/2011.”[126] He also recognized that, at the precautionary stage, “it would be premature and methodologically inappropriate to establish ex ante the definitive set of appropriate technical safeguards” for sideloading.[127]

Those admissions are revealing. The Decision simultaneously recognized that price-setting and detailed technical regulation fall outside antitrust’s proper domain while constructing a framework that almost inevitably led there.

That outcome arrived seven months later. The December 2025 settlement operationalized the Tribunal’s three principles by specifying what the Decision itself acknowledged it could not: precise commission rates of 10% or 20% for App Store transactions, depending on the payment processor; a 5% “core technology commission” for apps distributed through alternative stores; and a 15% fee on transactions initiated through steering links.[128] CADE also required any security warnings shown to users to use “neutral and objective” language, effectively dictating the content of Apple’s consumer communications.[129] The settlement imposed a 105-day implementation timeline and penalties of up to R$150 million for noncompliance. Notably, it ultimately omitted the sideloading requirement.

The specificity of these obligations inverts the conventional hierarchy of antitrust remedies. Herbert Hovenkamp argues that antitrust relief should proceed incrementally: first targeted injunctions against challenged conduct, then quasi-structural remedies such as interoperability mandates, and structural interventions only as a last resort.[130]

The logic is institutional as much as economic. Courts and competition authorities generally lack the continuous supervisory capacity characteristic of sector-specific regulators, including detailed industry expertise, access to firm-level information, and the administrative infrastructure necessary to oversee ongoing compliance. Remedies prescribing prices or business architecture may therefore produce greater welfare losses than the conduct they seek to correct.[131]

Comparisons with other jurisdictions underscore the point. In the United States, the most exhaustive judicial treatment of the same conduct—Epic Games v. Apple—produced a single, narrowly tailored injunction against anti-steering provisions, leaving Apple’s commission structure, payment system, and distribution model intact.[132] In the European Union, the DMA’s obligations emerged from a legislative process involving stakeholder participation, impact assessments, and continuing administrative oversight by the European Commission.[133] Brazil’s settlement achieved outcomes more expansive than either regime—mandating alternative app stores, alternative payment systems, and specific fee structures—through a precautionary proceeding resolved before any merits determination.

As Dennis Carlton and Randal Picker explain, antitrust and regulation reflect different institutional competencies.[134] Antitrust promotes competition through generally applicable principles without prescribing prices or mandatory dealing terms. Regulation, by contrast, can set prices and mandate access conditions, but only where supported by the ongoing administrative capacity and sector-specific expertise regulatory agencies are designed to provide. Where markets are not characterized by structural failures, such as natural monopoly or insurmountable barriers to entry, antitrust is generally preferable precisely because it avoids the costs and distortions of prescriptive regulation.

The CADE settlement blurs that distinction. By codifying specific fee percentages—including the 15% steering fee and 5% core technology commission—it performs a price-regulation function without the institutional infrastructure that price regulation requires: continuous monitoring, periodic adjustment, cost analysis, and revision as market conditions evolve.[135] The more detailed and prescriptive an antitrust remedy becomes, the greater the risk of regulatory ossification and unintended consequences—especially in digital markets, where business models evolve faster than regulatory frameworks.

None of this suggests CADE’s remedial objectives were illegitimate. If the competitive concerns identified in the proceedings were well founded, meaningful relief might indeed require substantial structural or behavioral intervention. But the path from diagnosis to remedy matters. The Tribunal adopted a contestable market definition, relied on a theory of harm without clear limiting principles, and applied a relaxed evidentiary standard. On that foundation, it constructed a remedial framework that ultimately produced one of the most prescriptive antitrust settlements in global digital-platform enforcement.

The mismatch between the rigor of the analysis and the ambition of the remedy should concern anyone interested in the institutional integrity of competition enforcement.

VII.                    Conclusion

CADE’s decision to uphold interim measures against Apple, and the case’s later settlement, mark significant moments in Brazilian competition enforcement. The proceedings show CADE’s willingness to address digital-market concerns and to engage with emerging theories of platform competition. They also show that traditional ex post antitrust tools can produce many of the changes that legislative proposals such as Bill 4,675/2025 and Bill 882/2026 would impose through ex ante regulation.

In that respect, the case weakens the argument for new conduct-specific digital-market regulation. Existing prohibitions on abuse of dominance proved capable of reaching the alleged harms involving steering, in-app payments, and app distribution. The interim measures and final settlement also confirm that Brazil’s competition regime has tools to prevent market harm when the evidence and circumstances warrant intervention.

But the case’s importance should not be confused with analytical soundness. The Decision rests on a single-brand market definition for iOS app distribution that is difficult to reconcile with established antitrust methodology, even though some decisions in the United States and, especially, the European Union have adopted similar approaches. The Tribunal’s ecosystem theory of harm may offer useful descriptive vocabulary, but it has not shown that it adds meaningful analytical value beyond traditional doctrines of tying, leveraging, and foreclosure.

The characterization of CADE’s decision as the “first true ecosystem theory of harm” may be accurate as a matter of terminology. It may be the first case in which a competition authority explicitly used the language of the academic ecosystem literature as the organizing framework for its analysis. But if the underlying economics are the same—if the real issues remain tying, leveraging, and foreclosure—then the innovation is rhetorical rather than substantive. The Brazilian experience thus appears to confirm what the European experience suggests: ecosystem theory has so far added little to antitrust analysis, and where it produces different outcomes, it often does so by being misunderstood or misapplied rather than by expanding the analytical toolkit.

Analytical frameworks matter. Competition enforcement protects consumers when it rests on rigorous economic analysis, properly defined markets, and falsifiable theories of harm. As applied here, the ecosystem framework does not yet meet that standard. It exaggerates Apple’s market power, gives too little weight to legitimate business justifications for vertical integration, and contributes to an unwarranted finding of anticompetitive conduct.

The deeper lesson is institutional. Brazil’s competition framework is flexible enough to address digital-market disputes without abandoning the consumer welfare standard or importing regulatory presumptions through antitrust procedure. That flexibility is a strength only if paired with discipline. Otherwise, ex post enforcement risks becoming sectoral regulation by another name.

Developers may understandably want a larger share of the ecosystem’s returns. But dividing platform rents is not—and should not become—the purpose of antitrust law.

[1] Projeto de Lei No. 4.675, de 2025, Câmara dos Deputados (Braz.), https://www.camara.leg.br/proposicoesWeb/fichadetramitacao?idProposicao=2562481.

[2] Projeto de Lei No. 882/2026, Câmara dos Deputados (introduced Mar. 3, 2026) (Braz.), https://www.camara.leg.br/proposicoesWeb/fichadetramitacao?idProposicao=2606171. The bill would amend Law No. 12,529/2011 and Law No. 8,078/1990 to classify restricting a competitor’s access to an application-distribution system as an abuse-of-dominance offense. As of May 2026, the bill remained pending before the Consumer Defense Commission. Deputy Clodoaldo Magalhães (PV-PE) was designated rapporteur on Apr. 17, 2026, and the amendment period closed May 4, 2026, with no amendments filed.

[3] The description of the facts and procedural history below derives, unless otherwise noted, from the English-language version of the CADE Tribunal’s May 14, 2025 decision upholding the interim measure issued by the agency’s General Superintendency [hereinafter Decision]. The full English-language decision is available at https://www.gov.br/cade/en/matters/news/cade-upholds-interim-measure-against-apple (last visited Apr. 4, 2026). The case file is available in Portuguese at https://sei.cade.gov.br/sei/modulos/pesquisa/md_pesq_processo_exibir.php?1MQnTNkPQ_sX_bghfgNtnzTLgP9Ehbk5UOJvmzyesnbE-Rf6Pd6hBcedDS_xdwMQMK6_PgwPd2GFLljH0OLyFUsJpu5C1ProV1zLrv9dPsgwllnq3_Q7zJg7Cr5_JvAc.

[4] CADE’s Tribunal appeared to acknowledge the dispute’s redistributive character in observing that Apple’s alleged conduct could be “exploitative”: “While orchestration is crucial for coordination in the ecosystem, design and governance decisions can generate functional and distributive failures, resulting in a loss of value for all participants and hindering innovation. The framing of these decisions under competition laws blurs the boundaries between the exclusionary and exploitative nature of abusive conduct.” Decision, supra note 2, ¶ 50.

[5] See Commission Decision C(2024) 1307 final of 4 March 2024 relating to a proceeding under Article 102 of the Treaty on the Functioning of the European Union and Article 54 of the EEA Agreement (AT.40437 – Apple – App Store Practices (music streaming)), 2024 O.J. (C/2024/3556), https://ec.europa.eu/competition/antitrust/cases1/202419/AT_40437_10026012_3547_4.pdf [hereinafter Eur. Comm’n Apple Music Streaming Decision].

[6] See Epic Games, Inc. v. Apple Inc., 559 F. Supp. 3d 898 (N.D. Cal. 2021) (Gonzalez Rogers, J.), https://www.courtlistener.com/docket/17442392/epic-games-inc-v-apple-inc.

[7] SG Order No. 24/2024, SEI 1476083 (Nov. 25, 2024), https://sei.cade.gov.br/sei/modulos/pesquisa/md_pesq_documento_consulta_externa.php?HJ7F4wnIPj2Y8B7Bj80h1lskjh7ohC8yMfhLoDBLddbbVtapRDlsIRSGO8EfHGTjlAvVixwsF8RWAxY7K1lOCwlw9mGBV9xj89bT7n1aVU7T0oE2YGB0rHOHC39OrFWc. The order adopted the reasoning set out in a technical note issued by the SG staff. See Technical Note No. 63/2024/CGAA11/SGA1/SG/CADE, SEI 1475850 (Nov. 25, 2024), https://sei.cade.gov.br/sei/modulos/pesquisa/md_pesq_documento_consulta_externa.php?HJ7F4wnIPj2Y8B7Bj80h1lskjh7ohC8yMfhLoDBLddZGjmDYkx3_EXIVLLLrA_C3ojklC750gYvLk4Wjzp2CQAzNjE5yiDgT6lb0_1xdyihsVVs3J1xFcXVJMWUJOcf9.

[8] See Commission Decision C(2018) 4761 final of 18 July 2018 relating to a proceeding under Article 102 of the Treaty on the Functioning of the European Union and Article 54 of the EEA Agreement (AT.40099 – Google Android), https://ec.europa.eu/competition/antitrust/cases/dec_docs/40099/40099_9993_3.pdf [hereinafter Eur. Comm’n Google Android Decision].

[9] See Eur. Comm’n Apple Music Streaming Decision, supra note 5.

[10] See Epic Games v. Apple, supra note 6.

[11] See Epic Games, Inc. v. Google LLC, No. 3:20-cv-05671-JD (N.D. Cal. Dec. 11, 2023) (Donato, J.), https://www.courtlistener.com/docket/17443962/epic-games-inc-v-google-llc.

[12] 14th Civil Court of the Judicial Section of the Federal District, Civil Writ of Mandamus No. 1097967-08.2024.4.01.3400, (Dec. 4, 2024)

[13] CADE, Press Release, CADE Discusses Competition in Mobile Ecosystems (Feb. 5, 2025), https://www.gov.br/cade/en/matters/news/cade-discusses-competition-in-mobile-ecosystems-1. A video of the hearing is available at CADE, Audiência Pública: Aspectos Concorrenciais dos Ecossistemas Digitais de Sistemas Operacionais Móveis (Feb. 19, 2025), https://www.youtube.com/watch?v=ehpGbkb8reA.

[14] CADE also requested written submissions. ICLE’s submission is available at Geoffrey A. Manne, Dirk Auer & Mario A. Zúñiga, ICLE Comments to Brazil’s CADE on Competition in Digital Ecosystems of Mobile Devices (Int’l Ctr. for L. & Econ., Feb. 2025), https://laweconcenter.org/resources/icle-comments-to-brazils-cade-on-competition-in-digital-ecosystems-of-mobile-devices.

[15] Decision, supra note 3.

[16] Marcela Mattiuzzo & Barbara De Sa Neves, Beyond Anti-Steering: Brazil’s Landmark Case Against Anticompetitive Conduct in Apple’s App Store, ABA Antitrust Newsletter (Oct. 1, 2025), https://www.americanbar.org/groups/antitrust_law/resources/newsletters/beyond-anti-steering.

[17] See Eur. Comm’n Google Android Decision, supra note 8.

[18] See Section III infra, particularly notes 61-63 and accompanying text.

[19] Decision, supra note 3, ¶ 374.

[20] See Bruno Carballa-Smichowski, Brazil’s Competition Authority Introduced the First True Ecosystem Theory of Harm, Kluwer Competition L. Blog (June 30, 2025), https://competitionlawblog.kluwercompetitionlaw.com/2025/06/30/brazils-competition-authority-introduced-the-first-true-ecosystem-theory-of-harm.

[21] CADE, Appeal No. 08700.009932/2024-18, Concurring Opinion of Commissioner Gustavo Augusto Freitas de Lima (May 2025), https://sei.cade.gov.br/sei/modulos/pesquisa/md_pesq_documento_consulta_externa.php?HJ7F4wnIPj2Y8B7Bj80h1lskjh7ohC8yMfhLoDBLdda_nlCzwUp7VFa9nLtVGuLOhLXyYKIuYhgQTUoVw8sFn9ghwmMNdUEzhJUqKTiW4ZBpKBg8O6ocATHg0QfSgGGn [hereinafter Freitas Concurring Opinion].

[22] Id. ¶ 7.

[23] Id. ¶ 14.

[24] CADE, Press Release, SG Recommends Apple’s Conviction for Anticompetitive Conduct (Jul. 7, 2025), https://www.gov.br/cade/en/matters/news/sg-recommends-apple2019s-conviction-for-anticompetitive-conduct. The SG’s technical note largely follows the Decision, although it does not rely on the “ecosystem theory of harm.” See Technical Note No. 51/2025/CGAA11/SGA1/SG/CADE, Administrative Proceeding No. 08700.009531/2022-04, SEI 1583153 (June 30, 2025), https://sei.cade.gov.br/sei/modulos/pesquisa/md_pesq_documento_consulta_externa.php?HJ7F4wnIPj2Y8B7Bj80h1lskjh7ohC8yMfhLoDBLddY0IIJxEsPG8Jm55kZvdBeZPpUBM3OXRcEQOClOvvDdT0EyPZqE632Y4bjFovY5rBb1Pw-jErxa5dpYf8mlN7XL.

[25] CADE, Cease and Desist Agreement, Administrative Proceeding No. 08700.007177/2022-04 (Dec. 23, 2025), https://sei.cade.gov.br/sei/modulos/pesquisa/md_pesq_documento_consulta_externa.php?HJ7F4wnIPj2Y8B7Bj80h1lskjh7ohC8yMfhLoDBLddagMC0PGtHHNJ5OLjWmXqqKoNDHq57WbWDJKa4eAKK0enGmrJphBHiCJbWXKXvI8K_s-573DMQSRm2YO5d-Z_Bw.

[26] See Dario Oliveira Neto & Mario Zúñiga, Apple in Brazil: Ex Post Antitrust Meets Ex Ante Ambitions, Truth on the Mkt. (Feb. 4, 2026), https://truthonthemarket.com/2026/02/04/apple-in-brazil-ex-post-antitrust-meets-ex-ante-ambitions.

[27] Giuseppe Colangelo, Do Ecosystems Exist in EU Competition Law? 2-4 (Nov. 6, 2025) (forthcoming in Eur. L. Rev.), https://ssrn.com/abstract=5408647.

[28] Id. at 9-12.

[29] Id. at 6.

[30] Id. at 7.

[31] Christopher S. Yoo, Ecosystems in Competition Law: A U.S. Perspective, Network L. Rev. (Spring 2025), https://www.networklawreview.org/yoo-ecosystems.

[32] See generally Francine Lafontaine & Margaret E. Slade, Transaction Cost Economics and Vertical Market Restrictions—Evidence, 55 Antitrust Bull. 587 (2010). (“… the empirical evidence is consistent and convincing. Combined with evidence on where VRs are found, as described above, we conclude that the empirical literature supports Williamson’s contention that ‘[c]ontrary to the inhospitability tradition, contractual constraints can and often do serve legitimate economic purposes. Specifically, vertical constraints may be needed lest subgoal pursuit by the individual parts destroy the viability of the system.’ The logical implication, as he suggests, is not that their use should never be questioned, but that the presumption should not be that they are detrimental to consumers.”) Id. at 608-09 (quoting Oliver E. Williamson, Assessing Vertical Market Restrictions: Antitrust Ramifications of the Transaction Cost Approach, 127 U. Pa. L. Rev. 953, 992 (1979)).

[33] See, e.g., Michael G. Jacobides & Ioannis Lianos, Ecosystems and Competition Law in Theory and Practice, 30 Indus. & Corp. Change 1199, 1209 (Oct. 2021), https://academic.oup.com/icc/article/30/5/1199/6428760.

[34] Yoo, supra note 31.

[35] Org. for Econ. Cooperation & Dev. (OECD), Efficiency Analysis in Vertical Restraints—Background Note by the Secretariat 9 (Latin Am. & Caribbean Competition F., Sept. 9, 2021), https://www.oecd.org/content/dam/oecd/en/publications/reports/2021/09/efficiency-analysis-in-vertical-restraints_7fe5fb55/7108965c-en.pdf.

[36] See, e.g., Copenhagen Econ., The Economic Rationale for Vertical Integration in the Tech Sector (2021), https://copenhageneconomics.com/wp-content/uploads/2021/12/copenhagen-economics-the-economic-rationale-for-vertical-integration-in-tech.pdf; Mario Zúñiga, Ms Vestager: Do Not Tear Down This Wall, Truth on the Mkt. (Aug. 5, 2024), https://truthonthemarket.com/2024/08/05/ms-vestager-do-not-tear-down-this-wall.

[37] Interim measures are undoubtedly a valuable antitrust-enforcement tool in digital markets, given both the typical length of antitrust proceedings and the speed at which such markets evolve. Even so, authorities should apply them only when there is a significant degree of certainty, both to ensure procedural fairness and to minimize the risk of costly enforcement errors, particularly in abuse-of-dominance cases.

[38] At least as a matter of policy. This brief does not address whether the standard applied is consistent with CADE precedent or Brazilian law more generally.

[39] Decision, supra note 3, ¶ 45.

[40] Id. ¶ 46.

[41] Id. ¶¶ 49-50.

[42] Colangelo, supra note 27, at 6.

[43] See United States v. Google LLC, No. 1:23-cv-00108, slip op. (E.D. Va. Apr. 17, 2025).

[44] Ohio v. Am. Express Co., 585 U.S. 529, 541-45 (2018) (holding that multisided transaction platforms must be evaluated as a whole, rather than by examining one side in isolation). See also David S. Evans & Richard Schmalensee, The Antitrust Analysis of Multi-Sided Platform Businesses 18 (Coase-Sandor Inst. for L. & Econ. Working Paper No. 623, 2012).

[45] See Lafontaine & Slade, supra note 32, and accompanying text.

[46] See, e.g., Eur. Comm’n, Decision of 18 July 2018, Case AT.40099—Google Android ¶ 557 (acknowledging multihoming, while concluding that “it [is] not sufficient to make Google Android and iOS devices homogenous”).

[47] Decision, supra note 3, ¶ 197.

[48] Apple’s system is relatively closed. The original iPhone employed a fully closed model: it lacked both an application store and any means to install additional applications (then commonly referred to as “programs”). Apple later relaxed that approach. In June 2007, it allowed Web 2.0 applications that looked and behaved like native iPhone applications, and in 2008 it launched the App Store. With the App Store, the iPhone effectively became a platform connecting consumers and application developers. Apple sells devices, software, and software updates to consumers, while selling developers access to users.

[49] Craig Conrath, Practical Handbook of Antimonopoly Law Enforcement for an Economy in Transition § 5, at 5 (1995).

[50] Decision, supra note 3, ¶¶ 100-139.

[51] Id. ¶¶ 102-106.

[52] Id. ¶ 104.

[53] Cal. Bus. & Prof. Code §§ 17200-17210.

[54] Epic Games v. Apple, supra 6, at 164, quoted in Decision, supra note 3, ¶ 104.

[55] Id. at 164.

[56] United Farmers Agents Ass’n v. Farmers Ins. Exch., 89 F.3d 832 (5th Cir. 1996) (table).

[57] Epic Games v. Apple, supra note 6, at 128.

[58] See Conrath, supra note 49.

[59] Benjamin Klein, Market Power in Antitrust: Economic Analysis After Kodak, 3 Sup. Ct. Econ. Rev. 53 (1993).

[60] Decision, supra note 3, ¶ 139.

[61] Id. ¶¶ 5.3.3 (analyzing anticompetitive discrimination), 5.3.4 (analyzing tying of the App Store and IAP).

[62] Id. ¶ 205.

[63] Epic Games v. Apple, supra note 6, at 46.

[64] David Evans, Multisided Platforms, Dynamic Competition, and the Assessment of Market Power for Internet-Based Firms 12 (Univ. of Chi. Coase-Sandor Inst. for L. & Econ. Research Paper No. 753, Mar. 10, 2016), https://ssrn.com/abstract=2746095.

[65] Geoffrey A. Manne, Dirk Auer & Mario Zúñiga, Comments of the International Center for Law & Economics on CMA’s Proposal to Designate Apple and Google with Strategic Market Status 2 (Aug. 20, 2025), https://laweconcenter.org/wp-content/uploads/2025/08/ICLE-CMA-Apple-Google-Designation-comments.pdf.

[66] “It’s now possible, if you’ve got its latest Pixel or Samsung phones, to ask the Gemini app to order an Uber or a meal on DoorDash. I tried it today and it worked flawlessly, if slowly. (I ordered paper towels via DoorDash, and once Gemini had arranged the transaction, it sent me to DoorDash to click the order button). OpenAI’s ChatGPT can’t yet offer quite the same thing. Score one for Google. This feature—what it calls task automations—is an example of an AI agent actually fulfilling its promise. Notably, Google hasn’t made a big splash about this so far, other than this low-key announcement last month.” Martin Peers, Google’s Gemini Steals a March on OpenAI, The Information (Mar. 23, 2026), https://www.theinformation.com/newsletters/the-briefing/googles-gemini-steals-march-openai. With this product, Google competes primarily with ChatGPT, Anthropic, and other artificial-intelligence services. It also competes with Apple by offering Android devices an at-least-temporarily exclusive feature.

[67] Decision, supra note 3, ¶ 210.

[68] See Statcounter Glob. Stats, Mobile Operating System Market Share Brazil (Feb. 2025-Feb. 2026), https://gs.statcounter.com/os-market-share/mobile/brazil (last visited Mar. 29, 2026). The Decision specifically notes that Apple’s market share is below 20%, which cuts against a presumption of dominance.

[69] Decision, supra note 3, ¶¶ 211-212.

[70] See IDC: Smartphone Shipments Up 2.3% in Q4 2025, Apple Is the Clear Winner, GSMArena (Jan. 14, 2026), https://www.gsmarena.com/idc_smartphone_shipments_up_23_in_q4_2025_apple_is_the_clear_winner_-news-71096.php (reporting global smartphone shipments of 1.26 billion units, with Apple accounting for 24.2%, Samsung for 18.2%, and Xiaomi for 11.2%); see also Malcolm Owen, Apple Dominated 2025 Smartphone Market with a 20% Share, AppleInsider (Jan. 12, 2026), https://appleinsider.com/articles/26/01/12/apple-dominated-2025-smartphone-market-with-a-20-share.

[71] Apple and Google, of course, compete differently with the aforementioned device manufacturers because most manufacturers use Google’s Android—or a forked version of Android—as their operating system.

[72] Decision, supra note 3, ¶ 212.

[73] See Gregory J. Werden, Market Delineation and the Justice Department’s Merger Guidelines, 1983 Duke L.J. 514 (1983) (explaining that substitution patterns, rather than surface-level product similarities or differences, determine market boundaries under the hypothetical-monopolist test, and that low cross-elasticity of demand between different tiers of the same product may justify separate markets).

[74] Counterpoint Rsch., Global Premium Smartphone Share Climbs to 25% in 2024 as Premiumization Continues (Feb. 17, 2025), https://counterpointresearch.com/en/insights/post-insight-research-notes-blogs-global-premium-smartphone-share-climbs-to-25-in-2024-as-premiumization-continues.

[75] Mobile World Live, Premium Smartphone Market Hits Record in H1 (Sept. 8, 2025), https://www.mobileworldlive.com/devices/premium-smartphone-market-hits-record-in-h1.

[76] Malcolm Owen, Apple Dominated 2025 Smartphone Market with a 20% Share, AppleInsider (Jan. 12, 2026), https://appleinsider.com/articles/26/01/12/apple-dominated-2025-smartphone-market-with-a-20-share.

[77] Mobile World Live, supra note 75.

[78] From US$1,199 to US$1,299 and above, depending on specifications. See Aleksander Anastasov, Samsung Galaxy S25 Ultra vs Google Pixel 9 Pro XL: A Friendly but Heated Rivalry, PhoneArena (Mar. 26, 2025), https://www.phonearena.com/reviews/galaxy-s25-ultra-vs-pixel-9-pro-xl_id6440; Google Store (U.S.), Pixel 10 Pro, https://store.google.com/us/config/pixel_10_pro?hl=en-US&selections=eyJwcm9kdWN0RmFtaWx5IjoiY0dsNFpXeGZNVEJmY0hKdiJ9 (last visited Apr. 1, 2026); Apple Store, Buy iPhone 17 Pro, https://www.apple.com/shop/buy-iphone/iphone-17-pro/6.9-inch-display-256gb-silver-unlocked (last visited Apr. 1, 2026); Samsung Store, Galaxy S26 Ultra, https://www.samsung.com/us/smartphones/galaxy-s26-ultra/buy/galaxy-s26-ultra-1tb-unlocked-sku-sm-s948uzvfxaa (last visited Apr. 1, 2026). Prices in Brazil are often higher, and the price gap between the iPhone and Samsung’s premium models is somewhat larger. Vivo offers the Galaxy S26+ at R$9,899 (approximately US$1,920), https://store.vivo.com.br/lancamento-s26/c (last visited Apr. 1, 2026), while the iPhone 17 Pro starts at R$11,499 (approximately US$2,230), https://www.apple.com/br/shop/buy-iphone/iphone-17-pro. The premium-smartphone segment, however, continues to grow. See Daniela Braun, Premium Smartphones Gain Ground in Brazil as Unit Sales Decline, Valor Int’l (May 6, 2025), https://valorinternational.globo.com/business/news/2025/06/05/premium-smartphones-gain-ground-in-brazil-as-unit-sales-decline.ghtml (“In the world’s fourth-largest smartphone market, manufacturers are betting that Brazilian consumers are willing to spend more for advanced devices with premium features and eye-catching design.”). Samsung, moreover, increasingly relies on AI features to compete in the premium-smartphone segment. See João Luiz Rosa, Samsung Doubles Down on AI in Brazil’s Crowded Smartphone Market, Valor Int’l (Feb. 26, 2026), https://valorinternational.globo.com/business/news/2026/02/26/samsung-doubles-down-on-ai-in-brazils-crowded-smartphone-market.ghtml (“The main appeal of the Galaxy S26 series is its advanced integration of artificial intelligence.”).

[79] Counterpoint Rsch., supra note 74.

[80] Decision, supra note 3, ¶ 214.

[81] Simon Baker & Laurence Wu, Applying the Market Definition Guidelines of the European Commission, NERA Topics 21 (Feb. 1998), https://www.cea.fi/course/material/NERA.pdf.

[82] Geoffrey A. Manne, Dirk Auer & Mario A. Zúñiga, Comments of the International Center for Law & Economics: Competition in Digital Ecosystems of Mobile Devices (iOS and Android) 4 (Int’l Ctr. for L. & Econ., Feb. 12, 2025), https://laweconcenter.org/resources/icle-comments-to-brazils-cade-on-competition-in-digital-ecosystems-of-mobile-devices.

[83] Dirk Auer, Making Sense of the Google Android Decision 20 (Int’l Ctr. for L. & Econ., Feb. 25, 2020), https://laweconcenter.org/wp-content/uploads/2020/02/Auer-Making-Sense-of-the-Google-Android-Decision-White-Paper.pdf (the Commission’s decision presaged CADE’s position that Apple holds a monopoly over nonlicensable mobile operating systems).

[84] See Accent, Consumer Purchasing Behaviour in the UK Smartphone Market for the CMA’s Mobile Ecosystems Market Study: Final Report iii (June 2022), https://assets.publishing.service.gov.uk/media/62a1cb0b8fa8f50395c0a0e7/Consumer_purchasing_behaviour_in_the_UK_smartphone_market_-_CMA_research_report_new.pdf.

[85] David S. Evans & Michael D. Noel, Defining Antitrust Markets When Firms Operate Two-Sided Platforms, 2005 Colum. Bus. L. Rev. 667, 681-90 (2005); see also Herbert Hovenkamp, Antitrust and Platform Monopoly, 130 Yale L.J. 1952, 1962 (2021).

[86] Decision, supra note 3, ¶ 213.

[87] Id. ¶ 255.

[88] Id. ¶ 271.

[89] See Freitas Concurring Opinion, supra note 21.

[90] Epic Games v. Apple, supra note 6, at 132 (“As a result, neither consumers nor developers are ‘locked-in’ to the App Store for digital mobile game transactions—they can and do pursue game transactions on a variety of other mobile platforms and increasingly other game platforms. Although the state of the wider gaming market is not at a level where the entirety of these gaming platforms can truly be characterized as competing for purposes of antitrust law (e.g., substitutes), the continued rise of cross-platform games, technologies, and innovative ways in which to reach consumers only demonstrate that these differing platforms are converging and ever intertwining. In sum, with seasoned antitrust counsel at the helm, Epic Games created a market definition which theoretically made a strong showing within the Newcal and Eastman Kodak framework. For the reasons explained above, the market definition was fundamentally flawed, and in any event, does not satisfy all four of the Newcal factors.”).

[91] Complaint ¶¶ 165-181, United States v. Apple Inc., No. 2:24-cv-04055 (D.N.J. Mar. 21, 2024).

[92] Apple advanced this argument, but the SG rejected it “because (…) the relevant market in which the conduct originated was defined as the non-licensable smart mobile operating system iOS, in which Apple is a monopolist.” Decision, supra note 3, ¶ 254.

[93] Decision, supra note 3, ¶¶ 317, 363.

[94] Id. ¶¶ 298-322.

[95] Id. ¶ 323.

[96] Id. ¶¶ 333-350. The Tribunal’s conclusion that Apple’s IAP system and app-distribution platform constitute “separate products” for purposes of tying doctrine rests primarily on their technical separability—i.e., that app distribution can exist without mandatory in-app payment processing. The Decision quotes the U.S. Supreme Court’s statement in Jefferson Parish that whether two items are separate products “turns not on the functional relation between them, but rather on the character of demand for the two items.” Jefferson Parish Hosp. Dist. No. 2 v. Hyde, 466 U.S. 2, 19 (1984). The Court also explained, however, that “[t]he requirement that two distinguishable product markets be involved follows from the underlying rationale of the rule against tying.” Id. at 20. In other words, the relevant question is whether the arrangement forecloses competition in a distinct product market in which separate provision would be efficient. Here, it has not been established that it would be efficient for Apple to offer its IAP separately, particularly given that Apple does not charge developers merely to distribute applications through iOS; commissions apply only when developers sell digital goods or services through the platform.

[97] Id. ¶ 104.

[98] Id. ¶ 364.

[99] Id. ¶ 326.

[100] Apple Inc., Addressing Spotify’s Claims (Mar. 15, 2019), https://www.apple.com/newsroom/2019/03/addressing-spotifys-claims (“A full 84 percent of the apps in the App Store pay nothing to Apple when you download or use the app.”).

[101] Epic Games v. Apple, supra note 6, at 105-10 (finding cognizable procompetitive justifications for Apple’s restrictions based on security and privacy).

[102] Copenhagen Econ., supra note 36, at 42-47 (finding that vertical integration in digital markets is associated with efficiency gains and enhanced consumer welfare).

[103] Randal C. Picker, Security Competition and App Stores, Network L. Rev. (Aug. 2021), https://www.networklawreview.org/picker-app-stores (arguing that open-distribution mandates risk exposing iPhone users to malware levels comparable to Android, where sideloading is permitted).

[104] Decision, supra note 3, ¶¶ 58-59 (contrasting Apple’s and Google’s business models and ecosystem strategies).

[105] Id. ¶¶ 132, 348, 389 (expressly citing the comparison as a benchmark).

[106] Apple, Building a Trusted Ecosystem for Millions of Apps: A Threat Analysis of Sideloading (Oct. 2021), https://www.apple.com/privacy/docs/Building_a_Trusted_Ecosystem_for_Millions_of_Apps_A_Threat_Analysis_of_Sideloading.pdf (reporting that Android devices permitting sideloading experience 15 to 47 times more malware infections than iPhones).

[107] Decision, supra note 3, ¶¶ 63-68 (discussing sideloading risks and the debate over whether security concerns justify exclusive app-store distribution).

[108] Heather West, Europe’s DMA: A Cybercriminal’s Paradise?, Ctr. for Eur. Pol’y Analysis (Nov. 13, 2025), https://cepa.org/article/europes-dma-a-cybercriminals-paradise (analyzing security vulnerabilities created by forced interoperability in messaging and payment systems).

[109] Picker, supra note 103.

[110] Sven Völcker & Daniel Baker, Why There Is No Antitrust Case Against Apple’s App Store: A Response to Geradin & Katsifis 23-24 (SSRN Working Paper, 2020), https://papers.ssrn.com/sol3/papers.cfm?abstract_id=3660896 (arguing that Apple’s commercial model creates procompetitive incentives for platform investment).

[111] Mikolaj Barczentewicz, The Digital Markets Act as an EU Digital Tax: When Compliance Costs Dwarf Regulatory Estimates, Truth on the Mkt. (July 8, 2025), https://truthonthemarket.com/2025/07/08/the-digital-markets-act-as-an-eu-digital-tax-when-compliance-costs-dwarf-regulatory-estimates (documenting how DMA compliance diverts resources from innovation toward regulatory negotiation).

[112] Act No. 12.529 (Nov. 30, 2011), arts. 36, 88, § 6. An English-language version is available at CADE, https://cdn.cade.gov.br/portal-ingles/topics/leniency%20program/Applicable%20Laws/law-no-12529-2011-english-version-from-18-05-2012.pdf; see also Dario Oliveira Neto, Will Brazil Subtly Sweep Consumer Welfare Under the Rug?, Truth on the Mkt. (Oct. 28, 2025), https://truthonthemarket.com/2025/10/28/will-brazil-subtly-sweep-consumer-welfare-under-the-rug.

[113] CADE, Guide for Horizontal Merger Review 7-8 (2016), https://cdn.cade.gov.br/Portal/centrais-de-conteudo/publicacoes/guias-do-cade/GUIDE%20FOR%20HORIZONTAL%20MERGER%20REVIEW.pdf (referring to consumer welfare as a criterion for evaluating the potential anticompetitive effects of both unilateral and coordinated conduct in horizontal mergers); CADE, Guidelines for the Analysis of Non-Horizontal Mergers 7-8 (2024), https://cdn.cade.gov.br/Portal/centrais-de-conteudo/publicacoes/guias-do-cade/V%2B%20Guide%20in%20English%20-%20Final%20version%202.pdf (requiring analysis of harm to final consumers when evaluating potential competitive harm from vertical mergers).

[114] Org. for Econ. Cooperation & Dev. (OECD), Advantages and Disadvantages of Competition Welfare Standards—Note by Brazil, OECD Doc. DAF/COMP/WD(2023)27, at 2-5 (June 15, 2023), https://one.oecd.org/document/DAF/COMP/WD%282023%2927/en/pdf (“stating that consumer welfare drives antitrust policy in Brazil”); see also Oliveira Neto, supra note 112.

[115] See infra Sections III-IV (analyzing the Tribunal’s market definition and theory of harm).

[116] Frank H. Easterbrook, The Limits of Antitrust, 63 Tex. L. Rev. 1, 2-3 (1984) (“If the court errs by condemning a beneficial practice, the benefits may be lost for good. Any other firm that uses the practice faces sanctions in the name of stare decisis, no matter the benefits. If the court errs by permitting a deleterious practice, though, the welfare loss decreases over time. Monopoly is self-destructive. Monopoly prices eventually attract entry.”).

[117] Joshua D. Wright & Murat C. Mungan, The Easterbrook Theorem: An Application to Digital Markets, 130 Yale L.J.F. 622, 636-43 (2021).

[118] Geoffrey A. Manne, Error Costs in Digital Markets, in Report on the Digital Economy 33, 35 (Glob. Antitrust Inst. ed., 2020), https://papers.ssrn.com/sol3/papers.cfm?abstract_id=3733662.

[119] See supra note 1. For a critical analysis of the bill, see Geoffrey A. Manne, Dario Oliveira Neto & Dirk Auer, Digital Overreach: A Premature Turn to Ex Ante Regulation in Brazil (Int’l Ctr. for L. & Econ. Apr. 7, 2026), https://laweconcenter.org/resources/digital-overreach-a-premature-turn-to-ex-ante-regulation-in-brazil.

[120] See Victor Oliveira Fernandes, Brazil’s Calibrated Revolution in Digital Competition, ProMarket (Nov. 12, 2025), https://www.promarket.org/2025/11/12/brazils-calibrated-revolution-in-digital-competition.

[121] Oliveira Neto, supra note 112.

[122] Dario Oliveira Neto & Mario Zúñiga, Beyond the Bark: Brazil’s Prudent Path in Digital Regulation, Truth on the Mkt. (Sept. 19, 2025), https://truthonthemarket.com/2025/09/19/beyond-the-bark-brazils-prudent-path-in-digital-regulation.

[123] Decision, supra note 3, ¶ 407.

[124] Epic Games v. Apple, supra note 6.

[125] Regulation (EU) 2022/1925 of the European Parliament and of the Council of 14 Sept. 2022 on Contestable and Fair Markets in the Digital Sector and Amending Directives (EU) 2019/1937 and (EU) 2020/1828 (Digital Markets Act), arts. 5(4)-(5), 6(4), 6(7), 2022 O.J. (L 265) 1, https://eur-lex.europa.eu/legal-content/EN/TXT/?uri=CELEX:32022R1925 [hereinafter Digital Markets Act].

[126] Decision, supra note 3, ¶ 423.

[127] Id. ¶ 433.

[128] CADE, Cease and Desist Agreement, supra note 25; see also Marcus Mendes, Apple Settles Brazilian Antitrust Case with App Store Policy Overhaul, 9to5Mac (Dec. 23, 2025), https://9to5mac.com/2025/12/23/apple-settles-brazilian-antitrust-case-with-app-store-policy-overhaul.

[129] CADE, Cease and Desist Agreement, supra note 25.

[130] Herbert J. Hovenkamp, Structural Antitrust Relief Against Digital Platforms, 7 J.L. & Innovation 1, 47 (2024).

[131] Abbott B. Lipsky, Jr. & J. Gregory Sidak, Essential Facilities, 51 Stan. L. Rev. 1187, 1195 (1999) (“Mandatory access remedies . . . are the stuff of regulatory bodies, not courts.”)

[132] Epic Games, Inc. v. Apple Inc., No. 4:20-cv-05640 (N.D. Cal. Sept. 10, 2021) (permanent injunction), aff’d in relevant part, 67 F.4th 946 (9th Cir. 2023).

[133] Digital Markets Act, supra note 125, arts. 5–7.

[134] Dennis W. Carlton & Randal C. Picker, Antitrust and Regulation (Nat’l Bureau of Econ. Rsch., Working Paper No. 12902, 2007), https://www.nber.org/papers/w12902.

[135] See Oliveira Neto & Zúñiga, supra note 26.

Executive Summary

On May 27, the European Commission is expected to propose the Cloud and AI Development Act (CAIDA), the centerpiece of its Tech Sovereignty package. The proposal arrives at a pivotal moment for artificial intelligence. The gap between frontier AI systems and the next-best alternatives is widening, and access to frontier compute and model APIs increasingly determines who can train, serve, and use the most capable systems.

That makes CAIDA’s treatment of frontier AI compute the package’s most consequential issue for European users.

The political pressure on CAIDA pulls in two directions. On the categorical side, 25 European cloud CEOs coordinated by CISPE, along with France, have urged the Commission to import European Defence Fund-style ownership and “effective control” criteria into the cloud market. Their approach would reserve procurement shares for European providers and exclude providers exposed to extraterritorial law. On the risk-based side, the Commission’s pre-proposal framing, along with the Nordics, Baltics, and the Netherlands, reportedly favors risk mitigation over a strict European-only rule.

Three market facts cut against the categorical approach.

First, Europe does not lead the top tier of rentable AI-cluster providers. On SemiAnalysis’s independent benchmarking, only three EU-headquartered providers appear anywhere in the top three tiers—and one, Nebius, exists in its current form only because of the 2024 corporate split from Yandex, a Russian-origin company.

Second, the supply chains needed to change that geography are severely constrained through at least 2027. Leading-edge logic wafers, high-bandwidth memory, networking silicon, and advanced packaging cannot be produced simply because Brussels wants more of them. The relevant bottlenecks sit in TSMC cleanrooms, Samsung and SK Hynix yield curves, Nvidia allocation decisions, and long-term contracts already signed by the largest buyers.

Third, the existing rental and offtake market is already locked up. AI labs and hyperscalers have committed to multi-year deals for frontier capacity, often at scales that smaller European buyers cannot match. In that market, European users enter as price-takers.

A categorical compute-sovereignty rule would therefore impose its costs mostly on European users, not on the non-EU providers it excludes. At the frontier—base-model pretraining, frontier-scale inference, and access to proprietary frontier-model APIs—the cost of being pushed to a lower tier is measured in slower research cycles, weaker serving economics, and lost access to the highest-productivity AI tools.

Nor would the stricter version of the sovereignty claim deliver what it promises. The “immunity from non-EU law” test already embedded in EUCS High+ assumes that EU headquarters and EU-located storage can shield data from extraterritorial legal reach. King v. OVH shows otherwise. In that case, an EU-headquartered provider holding EU-located data faced a Canadian production order. Legal sovereignty is not legal immunity.

This brief argues that the EU should answer the cloud-and-AI question by building, not prohibiting. CAIDA should remain risk-based at the EU level, while preserving member-state subsidiarity for stricter public-administration rules where genuinely necessary. In general, existing tools—the GDPR, defined national-security exceptions, and proportionality rules in public procurement—already provide a better framework than provider-nationality exclusions.

The build agenda is where EU policy has real leverage: faster data-center permitting, grid expansion, power availability, corporate-law reform, deeper capital markets, and demand aggregation around capable EU-located providers. Europe should strengthen its position inside the Western capability-controls architecture, not define sovereignty as withdrawal from the only frontier-compute stack that currently exists.

I. Europe’s Sovereignty Turn Meets the Frontier-AI Bottleneck

On May 27, the European Commission is expected—after two earlier delays[1] —to propose the Cloud and AI Development Act (CAIDA),[2] the European Union’s first attempt to write a binding definition of “sovereign cloud” into law. The goal is to move cloud-sovereignty requirements from voluntary certification schemes into mandatory rules, with adoption targeted for the fourth quarter of 2027.

CAIDA sits at the center of the broader Tech Sovereignty package, a legislative bundle led by Executive Vice-President Henna Virkkunen. The package also includes a revised Chips Act (“Chips Act 2,” targeted for adoption in the second quarter of 2027), an open-source strategy, and a digitalization-and-artificial-intelligence roadmap for the energy sector.

CAIDA arrives amid a sharp shift in artificial-intelligence capability—one that makes this regulatory cycle different from earlier European debates over cloud sovereignty. Over the past year, the gap between the economic value of frontier-quality AI tokens and merely good ones has widened substantially. Tokens are the units in which model output is measured and priced—roughly, fragments of words generated by an AI model.

The AI specialist consulting firm SemiAnalysis estimates that Anthropic’s annualized revenue grew from $9 billion at the end of 2025 to more than $44 billion by spring 2026.[3] During the same period, gross margins on inference—the process by which a trained AI model produces an answer to a user prompt—rose from below 40% to more than 70%. That is why CAIDA’s treatment of frontier-AI compute and frontier-model application-programming interfaces (APIs) may matter more than any other part of the package. Compute is the processing power needed to train and run AI models. APIs are the access points through which businesses use proprietary models like Claude, GPT-5, or Gemini without building those models themselves. Who may train, who may serve, and who may buy access will determine what European users can actually do with AI.

Political pressure on CAIDA now pulls in two directions.

On one side, 25 European cloud chief executive officers, coordinated by Cloud Infrastructure Services Providers in Europe (CISPE), called in March for CAIDA to import European Defence Fund-style ownership and “effective control” requirements into the cloud market.[4] Their proposal included reserved procurement shares for European providers and the explicit exclusion of providers subject to extraterritorial legal obligations. In plain terms, they want “European cloud” to mean not just that data sits in Europe, but that non-European governments cannot reach the provider through foreign law. France has pushed in the same direction. Its 2024 Sécuriser et Réguler l’Espace Numérique (SREN) law already requires sovereign-cloud hosting for sensitive workloads.

On the other side, pre-proposal reporting suggests the Commission’s current approach would limit restrictions to sensitive public-sector data, including financial, judicial, and health information, while leaving private-sector cloud use alone.[5] Nordic and Baltic member states, along with the Netherlands, reportedly favor risk-mitigation measures over strict European-only requirements.

The drift toward a categorical sovereignty rule was already visible in last month’s Cloud Sovereignty Framework procurement.[6] The procurement—worth as much as €180 million over six years—covers fully managed cloud services, commonly known as Platform-as-a-Service (PaaS). Under a PaaS model, the provider manages the servers, storage, networking, and operating systems. The customer brings the application. It is cloud computing with most of the plumbing hidden from view.

The Commission evaluated bids under its Sovereignty Effectiveness Assurance Levels (SEAL) framework, which measures legal and operational sovereignty. Three of the four winning providers achieved SEAL-3 “Digital Resilience” status, meaning the services “cannot be blocked by a non-EU third party.” The fourth awardee, operating through S3NS—a joint venture between Thales and Google Cloud—received only SEAL-2 “Data Sovereignty” status. European cloud trade associations have already criticized the arrangement as institutionalized “sovereignty washing.”[7]

At roughly the same time, a separate technical assessment of AI-cloud providers tells a very different story. SemiAnalysis’s ClusterMAX 2.1 evaluates providers not on legal sovereignty, but on the operational factors that determine whether frontier-AI work gets done: cluster reliability, networking quality, storage speed, and overall system efficiency.[8] A cluster is a large group of connected processors—usually advanced graphics-processing units (GPUs)—that work together to train or run AI models. For frontier AI, the question is not just whether a provider has chips. It is whether those chips, networks, and storage systems can reliably operate as one high-performance machine.

Of the four providers selected under the European Union’s sovereign-cloud procurement, only one appears in the top three tiers of the ClusterMAX rankings.

The gap between these two pictures—the SEAL-framework procurement and the ClusterMAX rankings—highlights what the European debate has not fully confronted. Legal sovereignty and AI capability are not the same thing. A cloud provider can check the right sovereignty boxes, while still falling short of the operational standard needed for frontier AI.

FIGURE 1: Reliable AI Compute, by Country and Tier

NOTE: Number of providers in each tier of the SemiAnalysis ClusterMAX 2.1 ranking (April 2026), grouped by country of headquarters. The European Union member-state category contains one Gold-tier provider (Nebius, the Dutch successor entity to Yandex), one Silver-tier provider in France (Scaleway), one in Luxembourg (GCORE), and all remaining providers in the “Not Recommended” category. Country classifications by the author, not SemiAnalysis.

At this point, three market realities point in the same direction.

First, the top tier of rentable AI-compute providers is not European-led. The European Union has only a small number of providers in the Platinum-through-Silver tiers.

Second, the semiconductor supply chains needed to change that reality—including advanced GPUs, memory, and packaging capacity—will remain severely constrained through at least 2027. Packaging refers to the advanced manufacturing step that connects chips, memory, and other components so they can operate at high speed. It is unglamorous, indispensable, and hard to expand quickly.

Third, the existing rental and compute-offtake market is already tied up in multi-year contracts, with buyers that have both deep pockets and little price sensitivity. Offtake contracts are long-term commitments to buy future capacity. In this market, the best capacity is not sitting on a shelf waiting for Europe’s next procurement rule.

Taken together, these constraints suggest that the categorical-sovereignty approach now visible in the CISPE position, the French policy line, and the SEAL procurement deserves another look. A provider-nationality rule that pushed European users away from non-European compute providers and frontier-model APIs would not create European capability quickly enough to matter. It would mostly make scarce capability harder for Europeans to use.

This brief uses “frontier” to refer to three related capabilities: frontier-scale base-model pretraining, frontier-scale inference, and access to proprietary frontier-model APIs. Base-model pretraining is the initial, resource-intensive process of building a general-purpose AI model from massive datasets. Inference is the later process of running that trained model to answer prompts or perform tasks. Proprietary frontier-model APIs are the paid access channels to the most capable closed models.

The analysis differs for less demanding commercial workloads, such as fine-tuning or inference using smaller open-weight models. Fine-tuning means adapting an existing model for a narrower task, while open-weight models are models whose parameters are publicly available for others to run or adapt. For those uses, the costs of sovereignty restrictions may be smaller. At the frontier, by contrast, a categorical sovereignty rule would mainly ration scarce capability away from European users.

That point carries an important corollary. The alternative to a categorical sovereignty regime is not just a narrower, workload-specific version of the same rule. It is a different policy posture altogether—one focused on building capability, rather than restricting access. The European Union has more leverage to shape cloud-and-AI markets through investment, procurement, energy policy, and capacity expansion than current sovereignty debates often acknowledge.

Given the economics of frontier AI, exclusion would impose two distinct costs on European actors.

For a European AI lab trying to compete at the frontier, a sovereignty rule that forced the lab one ClusterMAX tier lower would likely mean slower research cycles and months of lost engineering time. For a European business using Claude, GPT-5, or Gemini through an API, losing access to frontier models could mean exclusion from the highest-productivity tier of commercially available AI systems. That is not sovereignty. It is self-rationing with better branding.

This brief examines the three market constraints in turn: geography, supply chains, and booked capacity. It then evaluates the implicit costs of a categorical compute-sovereignty rule for European users, situates the issue within the broader Western export-control and capability-governance framework in which European firms already operate, and concludes by outlining a build-rather-than-prohibit alternative that existing EU policy instruments could plausibly support.

The analysis draws primarily on SemiAnalysis’s 2026 research, which currently offers the public market one of the most coherent accounts of AI-infrastructure supply, demand, and operational quality.

II. What Counts as ‘Compute,’ and Why the Frontier Is Different

Before going further, it is worth clarifying a term that policy debates often flatten: “compute.” In artificial intelligence, compute is not one thing. It can mean a single GPU running a small model, or it can mean a massive cluster of thousands of GPUs training a frontier model. A sovereignty rule would not affect those uses in the same way.

The argument that follows applies chiefly to three frontier uses.

First is frontier base-model pretraining: the tightly coordinated, multi-thousand-GPU process by which a small number of labs train genuinely new, general-purpose flagship models.

Second is frontier-scale inference: serving those flagship models at production scale. Inference is the process by which a trained model responds to prompts.

Third is access to frontier proprietary models through APIs, such as Anthropic’s Claude, OpenAI’s GPT, and Google’s Gemini. APIs are the access layer through which most European businesses and researchers encounter frontier AI. For those users, the key sovereignty constraint is not which cluster they personally run. It is whether they retain legal access to the API at all.

Much useful AI work falls outside this frontier category, and the distinction matters. Some labs publish open-weight models, meaning models whose trained parameters can be downloaded and run by others, rather than accessed only through a proprietary API. Many of the strongest such models currently come from China, including DeepSeek, Kimi, and Qwen, although Google, Meta, and Mistral also release important open-weight models.

With sufficient infrastructure—anything from an older gaming GPU to clusters costing millions of euros—open-weight models can be downloaded, fine-tuned for specific tasks, run at home or in an office, or served from European clouds at quality levels that work well for many uses. For specialized applications, or for tasks that do not require frontier-level performance, an EU location requirement may therefore impose only a modest tax.

The argument is not that Europe should have no compute strategy, nor that every European AI workload depends on non-European infrastructure. It is narrower: at the frontier (in base-model training, frontier-scale inference, and access to the best proprietary APIs), compute access matters most, and the cost of exclusion compounds fastest.

If EU persons faced severe limits on frontier-pretraining infrastructure, no European frontier model could be built without working around the rule. Mistral and any successors would either train on non-EU compute, defeating the sovereignty objective, or train at smaller scale and lower quality, falling further behind the frontier each cycle. That is especially costly in a market where the gap between the best models and the next tier has widened, rather than narrowed.

If EU persons faced severe limits on frontier-scale inference infrastructure, no European platform could serve frontier-quality models at competitive economics. Frontier serving would consolidate further around the U.S. providers that already dominate it.

And if EU persons faced severe limits on access to frontier proprietary-model APIs, European businesses and researchers would lose access to the productivity gains they now get from Claude, GPT, and Gemini. On SemiAnalysis’s accounting, those gains can run five to 10 times the cost of the underlying tokens.[9] Their non-European competitors would keep the same tools.

The cost of frontier exclusion, in other words, is not the loss of “AI” in some general sense. It is the loss of the specific capability tier where the economic gap between the best option and the next-best alternative is unusually large, and continuing to widen.

A note on method: This brief uses provider headquarters as the first-pass proxy for “European” or “non-European” because that is the proxy the European sovereignty debate itself uses. The SEAL framework,[10] SecNumCloud,[11] the European Cybersecurity Certification Scheme for Cloud Services (EUCS) High+ debate, and the recent CISPE objection to S3NS being recognized as “sovereign” all rest on headquarters-and-ownership criteria.

Moreover, the Commission’s SEAL framework and SemiAnalysis’s ClusterMAX do not measure the same thing. That is precisely why both matter.

SEAL grades legal and operational sovereignty: whether a service can be blocked by a non-EU third party, whether the provider is incorporated and controlled in the EU, and whether the supply chain runs through the EU from end to end.

ClusterMAX grades operational AI-compute capability: whether a cluster can reliably produce useful work at frontier scale, how well its interconnect performs, and how quickly customers can get a large training job running. An interconnect is the high-speed networking fabric that lets thousands of processors act less like separate machines and more like one large system.

The two rankings answer different questions about different layers of the same stack. A provider can be SEAL-3 and still fail to appear anywhere on ClusterMAX’s serious-tier list. One recurring claim of this brief is that European policy debates often treat legal sovereignty and technical capability as substitutes. They are not.

The April ClusterMAX 2.1 rankings (Figure 1) make the capability gap plain. SemiAnalysis grades providers from Platinum at the top through Gold, Silver, Bronze, and “Not Recommended.” Higher tiers correspond to better measured cluster reliability, networking, storage, and customer experience.

Exactly one provider (CoreWeave, headquartered in New Jersey) holds the Platinum rating. Of the five Gold-tier providers, three (Oracle, Microsoft Azure, and Crusoe) are headquartered in the United States. The fourth (Nebius) is incorporated in Amsterdam, although its origin as the renamed Yandex N.V. is likely to matter to any sovereignty-minded reader. The top tiers are U.S.-dominated, not EU-led, with the Dutch entry carrying the Yandex provenance discussed below.

The Silver tier improves the European picture only slightly. Of the providers SemiAnalysis grades at Silver, exactly two are headquartered in the EU: Scaleway (a subsidiary of the French Iliad Group) and GCORE (headquartered in Luxembourg). The remaining Silver-tier providers include eight U.S. firms (Together AI, Lambda, Google Cloud, Amazon Web Services, Cirrascale, Vultr, Voltage Park, and TensorWave); one Australian firm with Singaporean origins (Firmus); and one Japanese firm (GMO). Across the entire Platinum-through-Silver range—the only tiers where serious frontier work is reliably done—the EU contributes Scaleway, GCORE, and the Yandex-descended Nebius.

The contrast sharpens when one looks at the providers a European policymaker might most naturally cite. Mistral AI, OVHcloud, Sesterce, and Mithril are all ranked “Not Recommended” in the SemiAnalysis assessment. So are Hetzner and dstack in Germany, Clore.AI in the Czech Republic, Tatra SuperCompute in Slovakia, and Telenor in Norway.

SemiAnalysis distinguishes within “Not Recommended” between “Underperforming” providers, where hands-on testing identified specific operational gaps, and “Unavailable” providers, where capacity either could not be tested in the relevant way or was not available for third-party rental. Mistral, Tatra SuperCompute, and Telenor fall in the “Unavailable” bucket. OVHcloud, Sesterce, Mithril, Hetzner, dstack, and Clore.AI fall in “Underperforming.”

Nebius deserves its own parenthetical. The highest-rated EU-located provider exists in its current form only because of the 2024 corporate split from Yandex N.V., undertaken to separate the international business from the Russian one.[12] That complicates any story in which “EU-headquartered” suffices as a proxy for sovereignty. Whatever else one says about the European cloud landscape, its highest-tier EU entry is neither a longstanding European champion nor an organic European startup.

The picture is sharper still one layer up the stack, where frontier-scale inference is served. Artificial Analysis, an independent benchmarking firm, tracks performance across more than 500 inference endpoints from more than 20 API providers, using metrics that include intelligence, output speed, and price.[13] Its provider universe overlaps substantially with ClusterMAX, but it is not identical, because serving inference is a different business from running training clusters. On the inference side, the EU contributes two names to the serious-inference universe: Scaleway and Nebius. The rest of the cohort is essentially American.

A European user serving a small open-weight model from a single eight-GPU node on commodity infrastructure is not meaningfully dependent on this list. A European platform trying to serve frontier-quality output at competitive economics is.

That kind of platform must run large, sharded model architectures, such as mixture-of-experts models, in which only part of the model activates for any given token. It also must use advanced serving techniques that depend heavily on cluster quality and scale. For those workloads, the EU’s position comes down to two providers—Scaleway and Nebius—one of which is the post-Yandex Dutch entity already discussed.

This is why tiers matter. SemiAnalysis’s Cluster TCO and Goodput methodology shows that, even when GPU-hour pricing is held constant across providers, a Silver-tier cluster has roughly 15% higher total cost of ownership (TCO) than a Gold-tier cluster for a representative large-language-model (LLM) pretraining workload.[14] Hyperscalers—the large cloud providers such as Amazon Web Services (AWS), Microsoft Azure, and Google Cloud—have about 10% higher TCO.

The cost differences come from “goodput” losses. Goodput is the share of paid computing time that actually produces useful work. Failures, slow detection, slow replacement, long setup times, weaker storage performance, and ongoing debugging all reduce goodput. Network-performance tuning can take “weeks to months” on a hyperscaler’s Elastic Fabric Adapter (EFA)—AWS’ proprietary high-throughput cluster networking system—compared with near-zero on Gold-tier InfiniBand or RoCE systems, which are open-standard networking technologies used for high-performance clusters. Weaker storage also lengthens job-initialization times after every interruption. In frontier AI, downtime is not merely annoying. It is expensive at industrial scale.

A European user effectively limited to EU-headquartered providers is therefore choosing from a very short list. A sovereignty advocate may reasonably answer that Europe should not accept this distribution, but change it—by building more reliable European clouds.

The next two sections explain why that path is unlikely to change the frontier-compute geography quickly enough to matter for the current regulatory cycle.

III. The Supply Chain Will Not Move on Brussels’ Timeline

The natural response to the geographic picture is that geography is itself a policy variable. Build more European clouds, the argument goes, and the ClusterMAX distribution will look different in three years.

That argument deserves a serious answer. It is sincerely held, and in the abstract, not obviously wrong. The problem is timing. In 2026, the inputs needed to build any large AI cloud—European or otherwise—are severely constrained by global supply chains. Those constraints sit upstream of both capital and political will.

The first bottleneck is access to leading-edge logic wafers, especially Taiwan Semiconductor Manufacturing Co.’s (TSMC) N3 family. “N3” refers to TSMC’s 3-nanometer generation, the most advanced chip-manufacturing process currently in volume production. These are the manufacturing lines used to produce many of the most capable AI accelerators—the specialized chips that train and run frontier models.

SemiAnalysis’s “Great AI Silicon Shortage” analysis shows that nearly all major AI-accelerator families have converged on N3, including Nvidia Rubin, AMD MI400, Google TPU v7 and v8, AWS Trainium3, and Meta MTIA.[15] On SemiAnalysis’s modeling, AI demand accounts for roughly 60% of N3 wafer output in 2026 and is projected to reach 86% in 2027, even as TSMC adds capacity. Effective N3 utilization is expected to exceed 100% in the second half of 2026, meaning TSMC will be pushing existing lines beyond their nominal capacity to serve buyers with multi-year demand visibility and few alternatives.

Why can TSMC not simply add more N3 capacity? Because the binding constraint is not just money. It is cleanroom space: the specialized facilities that must be built before lithography tools can be installed and new wafer starts brought online. Construction timelines run in years, not months. Even if European industrial policy could produce unlimited capital tomorrow, the cleanroom calendar would still move on its own schedule. For practical purposes, N3 capacity for 2026 and 2027 is already spoken for.

Memory is the next constraint, and arguably the more dramatic one. SemiAnalysis describes the current dynamic as a “once-in-four-decades” memory supercycle.[16] High-bandwidth memory (HBM) is specialized stacked dynamic random-access memory (DRAM) packaged directly with AI accelerators. The latest generation, HBM4, is central to frontier-AI systems because these models need enormous amounts of data to move quickly between memory and processors.

HBM is also expensive in capacity terms. It consumes roughly three times as much wafer capacity per bit as commodity DRAM, rising to nearly four times as much with HBM4. Every additional HBM bit therefore crowds out commodity memory. Prices in some memory segments have risen roughly sixfold year over year, while the number of major DRAM suppliers has fallen from more than 20 in the mid-1990s to three or four today: Samsung, SK Hynix, Micron, and a much smaller fringe.

The constraint here is not whether Brussels is willing to invest. It is the HBM4 yield curve at SK Hynix and Samsung, Micron’s much-discussed lag at higher pin speeds, and the long-term agreements that customers now use to lock in HBM allocation.[17] Nvidia, in particular, has reportedly secured substantial HBM supply across multiple generations of its rack-scale GPU systems.

The bottleneck extends beyond logic and memory. Networking silicon—the Ethernet switches and network-interface cards that connect modern AI clusters—is part of the same supply chain and follows the same priority queue. Advanced packaging—the process that integrates logic chips, memory, and other components into high-performance systems—remains tight, although it is easing in part because TSMC has aligned packaging investment with front-end wafer constraints, rather than overshooting them.

There are workarounds at the margins, including Intel’s Embedded Multi-Die Interconnect Bridge (EMIB) packaging technique and outsourced semiconductor assembly-and-test providers such as ASE/SPIL and Amkor. But none changes the basic picture. The marginal AI cluster, European or otherwise, is competing for scarce inputs in a global queue.

This point corrects a recurring misconception in the European policy debate. At the frontier, compute supply is not simply a function of capital and political will. It depends on TSMC’s cleanroom calendar, Samsung and SK Hynix’s HBM yield curves, and the long-term agreements that major U.S. AI labs and hyperscalers have already signed with foundries, memory suppliers, and original equipment manufacturers (OEMs). These are not levers the European Commission or member-state governments can pull on a useful timetable.

That matters because the European Union’s standard response to a structural disadvantage—investment, regulation, and certification—runs into a wall those tools were not designed to move. European policy can affect electricity supply, permitting, demand aggregation, procurement, and EU-located capacity. Those are real levers. But they cannot, on the current horizon, conjure TSMC cleanroom space, HBM4 yield, Nvidia allocation, or uncontracted GB300 fleets at the scale needed to erase the frontier-compute gap.

That leaves a third question: Even if Europe cannot rebuild the frontier-compute stack quickly, can it rent enough existing capacity in the meantime?

IV. Renting the Frontier Is Not Much Easier

SemiAnalysis’s “Great GPU Shortage” analysis reports that on-demand graphics-processing-unit (GPU) rental capacity is sold out across Nvidia’s previous and current AI-chip generations: Hopper and Blackwell.[18] “All capacity coming online until August to September 2026 has already been booked,” and lead times for new Blackwell deployments now stretch into June or July. Renting a cluster, one observer remarked, is starting to look less like booking the last seat on a crowded flight and more like trying to buy a controlled substance.

Pricing tells the same story. SemiAnalysis’s H100 1-year contract-price index, now publicly available, shows prices rising from $1.70 per GPU-hour in October 2025 to $2.35 per GPU-hour by March—a 40% jump in five months for a two-generation-old chip.[19] Hopper contracts that were supposed to expire this year are being renewed at their original rates from two or three years ago and, in some cases, extended through 2028. This is what tight markets do: They ration access through price and long commitments, with the remaining flexibility going to buyers that can lock up capacity the longest.

The long-term market is dominated by AI labs buying at scale. Four- and five-year deals now often reach 50 to 100 megawatts, which is enough electricity to power tens of thousands of current-generation GPUs under a single contract. Hyperscalers often backstop these deals. They act as the direct offtaker, meaning the party that signs a long-term contract to buy a defined amount of compute from a neocloud, then resells that compute to an AI lab. A neocloud is a newer cloud provider built mainly around GPU capacity for AI workloads.

That matters because the marginal European cluster is not really competing with other European clusters for compute. It is competing with Anthropic, OpenAI, Google, Meta, and Microsoft, and with any hyperscaler willing to stand behind their offtake agreements.

These buyers can commit at this scale because token demand now substantially exceeds token supply. SemiAnalysis’ “AI Value Capture” analysis shows that the economics of frontier models have pulled away from the next tier, as the Anthropic revenue and margin figures discussed in Section I suggest.[20] Frontier buyers have a high—and currently price-insensitive—willingness to pay for compute. Suppliers know it.

A European actor entering this market—sovereign or otherwise—therefore enters as a price-taker in a market tilted decisively toward suppliers and the largest buyers.

V. The Cost of Sovereignty Depends on the Workload

The cost of a sovereignty rule depends on which workload it reaches. The three frontier tiers—pretraining, inference, and API access—would bear the cost differently.

The recent Cloud Sovereignty Framework procurement makes the operational stakes concrete.[21] As noted above, the Commission’s tender focused on fully managed cloud services—especially PaaS, developer experience, and automation. The Commission evaluated providers under the SEAL framework, not by frontier-AI cluster capability.

That distinction matters. A PaaS-oriented sovereign-cloud tender need not reproduce SemiAnalysis’ ClusterMAX rankings. But a sovereignty framework now being normalized in public procurement can easily be mistaken for evidence of AI-compute sovereignty. It is not. Legal sovereignty and frontier-compute capability are different things.

At the frontier-pretraining tier, a categorical preference for EU-headquartered providers would constrain the small number of European labs trying to train frontier base models. The TCO impact is workload-specific. In SemiAnalysis’s LLM pretraining scenario, Silver-tier neocloud TCO is roughly 15% higher than Gold-tier TCO at equal GPU-hour pricing.[22] The gap widens for “Not Recommended” providers that lack basic operational infrastructure.

That is the relevant comparison for large, long-running training jobs. It is not a universal inference tax. In SemiAnalysis’ modeled inference-endpoint scenario, equal-priced Gold and Silver neoclouds differ by less than 1%. For a European frontier-model effort, a sovereignty rule that pushed the lab even one ClusterMAX tier lower would likely cost months of engineering time.

At the frontier-inference tier, the cost would look different. The European platforms trying to serve frontier models at scale—chiefly Scaleway and Nebius, along with a small number of European model labs (most prominently Mistral) that operate their own infrastructure—compete against U.S. providers with the dominant inference economies of scale.

A categorical sovereignty rule would push European inference customers onto a shorter provider list, with weaker pricing power, smaller capacity reserves, and slower iteration on the advanced serving techniques that make frontier inference economical. The severity would depend on the workload. Serving a small open-weight model from a single node can work well on Silver-tier infrastructure. Serving very large mixture-of-experts models at competitive cost is a different business.

At the API tier—the layer where most European users actually encounter AI—the cost would be different again. It would be less about cluster tier than about access.

A European business, researcher, or institution using Anthropic’s Claude or OpenAI’s GPT-5 through an API consumes inference on the frontier lab’s own infrastructure stack. That stack may include internal capacity, hyperscaler partnerships, custom silicon such as Google’s Tensor Processing Units or Amazon’s Trainium chips, and rented neocloud capacity. Not all of that capacity appears directly in ClusterMAX, but it belongs to the frontier-grade class this brief addresses.

The binding sovereignty constraint at the API tier would not be which cluster the user runs on. It would be whether the user may legally use the API at all. A categorical bar on transferring personal data to non-EU controllers or processors, or on contracting with non-EU service providers even with safeguards, would in practice foreclose access to those APIs for many ordinary commercial workloads. SemiAnalysis’s analysis shows that the gap between frontier proprietary models and the next-best alternatives, including the best open-weight models, has widened rather than narrowed over the past year.[23]

The implicit cost of a categorical compute-sovereignty rule would therefore fall less on the providers it excludes than on the European users it pushes onto worse infrastructure or cuts off from better outputs. The structure is not new. The same pattern appears in critiques of the EU’s recent GDPR “fix”[24] and earlier work on industrial-policy approaches to “digital sovereignty.”[25]

The legal mechanisms that could operationalize a sovereignty preference are familiar: data-transfer rules and Schrems II-derived adequacy concerns under the GDPR; the EUCS high-assurance debates over immunity from non-EU law;[26] and procurement preferences such as the SEAL framework now being normalized at the Commission level.

Most of these instruments were not designed as compute-sovereignty tools. But policymakers may be tempted to read them as imposing categorical restrictions on non-EU compute or APIs for ordinary commercial AI workloads. That reading is not inevitable. Still, the Cloud Sovereignty Framework already builds end-to-end EU-supply-chain criteria into Commission procurement.

CAIDA, which will be proposed (along with the Chips Act 2) on May 27 as part of the Tech Sovereignty package[27] (after delays from late March and mid-April),[28] is widely expected to move sovereignty requirements from voluntary certification into binding legislation. Adoption is targeted for the fourth quarter of 2027.

Based on the most recent pre-proposal reporting,[29] the Commission’s framing emphasizes “sensitive services,” including financial, judicial, and health data processed by public-sector organizations, rather than ordinary commercial workloads. The package reportedly would not regulate private-sector cloud use. Whether that line holds through member-state negotiation—and whether the same instruments are later read more expansively in implementation—is one of the central questions facing the package.

VI. Compute Sovereignty in an Export-Controlled World

One point gets surprisingly little airtime in Europe’s sovereignty debate: The global compute market is not a neutral commodity market with national preferences layered on top. It is a strategic market that the United States and a tightening group of partners—including the Netherlands, Japan, South Korea, Taiwan, the United Kingdom, and, to varying degrees, Australia and Singapore—have spent the past three years shaping through export controls.

The result is a capability gradient. Access to frontier-AI compute depends not only on price, capacity, and contract terms, but also on alignment with rules governing who may use the technology and for what purposes.

The architecture is familiar in outline. The U.S. Bureau of Industry and Security’s October 2022 controls on advanced AI chips[30] were expanded repeatedly in 2023, 2024, and 2025. HBM controls were added in December 2024, and the short-lived AI Diffusion Framework was issued and then rescinded in early 2025.[31] The Netherlands has constrained ASML lithography exports. Japan has constrained Tokyo Electron and Nikon equipment. The United Kingdom has aligned on outbound-investment review. The design treats AI compute much as earlier regimes treated nuclear materials or strong cryptography: as a strategic input to be managed bilaterally and multilaterally, not left entirely to commodity markets.

Two features of this architecture matter for Europe.

First, export controls have not denied near-frontier AI capability to their targets. DeepSeek’s V3 and R1 models, released in early 2025, demonstrated training and inference close to—though still behind—the contemporaneous Western frontier, despite running on legally restricted hardware.[32] Subsequent reporting on chip flows through Singapore and other intermediary jurisdictions documented material leakage of nominally controlled inventory. China’s domestic ecosystem—including SMIC’s N+2 process, Huawei’s Ascend family, and CXMT in commodity DRAM—has accelerated rather than stalled on most public assessments.

The export-control regime has slowed China’s frontier without foreclosing it, while accelerating China’s push to substitute domestic alternatives. In other words, the regime is neither succeeding as advertised nor failing entirely.

Second, that same architecture defines the terms on which European users access frontier compute. The chips, clusters, model weights, and APIs discussed above all sit inside a gradient that runs from the United States and its closest partners outward. EU member states sit near the inside, but not at the center—and not uniformly. The original AI Diffusion Framework placed several member states in a capped middle group and required active diplomacy to lift those caps. The framework was rescinded, but its underlying logic survives: allies are differentially trusted, and frontier-compute access depends on that trust.

That means “sovereignty” framed as decoupling from the Western stack would choose the worst available position. The Chinese stack is sanctioned and, in any event, not legally available to European users. The Western stack is where the frontier sits, and access to it depends on alignment with the export-control regime. A European policy that defined itself as an autonomous third pole between the two would, in practice, place Europe downstream of both: unable to use the Chinese stack and less trusted within the Western one because it had defined itself in opposition to it.

The Commission’s own Joint Research Centre reached much the same conclusion earlier this month.[33] Its policy brief argued that “digital sovereignty cannot be equated with autarky, as pursuing total autonomy across the entire technology stack is both infeasible and undesirable.” It instead called for economic statecraft calibrated between “insulation and interdependence,” with autonomy pursued through “selective rebalancing rather than full-fledged decoupling.” That narrower case—for European capacity formation, foreign-direct-investment incentives, and demand aggregation around EU-located Gold and Silver providers—remains intact. The case for categorically excluding U.S. compute and APIs does not.

The most ambitious version of the categorical-exclusion approach is the “immunity from non-EU law” test already embedded in EUCS High+ and likely to appear in CAIDA or a successor sovereignty-certification scheme. The test treats EU headquarters and EU-located processing as sufficient protection against the reach of non-EU law into EU data. It would not deliver what it implies.

The Canadian case King v. OVH demonstrates why.[34] In September 2024, the Ontario Court of Justice upheld a production order requiring OVHcloud—the canonical European sovereign-cloud champion—to disclose subscriber data stored on servers in France, the United Kingdom, and Australia. The court applied a “virtual presence” doctrine and treated France’s blocking statute as presenting a low risk of prosecution. OVH’s appeal to the Ontario Superior Court has been pending since late 2024 and remains undecided as of early 2026.

That should give sovereignty maximalists pause. The most prominent live extraterritorial-production order of the past 18 months runs against an EU-headquartered provider holding EU-located data. EU headquarters and EU-located storage do not, by themselves, dissolve extraterritorial legal reach. A categorical test that promises otherwise is making a claim the law does not support.

The harder question—and the one Europe’s AI-sovereignty debate has not yet seriously engaged—is what the EU can actually influence directly. A build-not-prohibit strategy would require more than swapping one certification label for another. It would require asking where EU policy can expand real capacity, where it can reduce dependence without pretending away interdependence, and where its own reflexes toward regulation, procurement rules, and formal sovereignty tests may themselves be part of the bottleneck.

VII. Build, Do Not Prohibit

The EU’s cloud-and-AI debate has focused mostly on whether and how to restrict access to non-EU providers. The evidence in the preceding sections points the other way: At the frontier, exclusion offers little upside and imposes high costs. The better question is what the EU can actually influence. The answer is a build agenda the current debate underweights.

The first category is capacity formation. Data-center shells, grid connections, transmission, cooling, water, planning approval, and time to energize all fall within European policy’s reach. The marginal AI cluster is increasingly gated by power and permitting, not just silicon. The gap between, say, Texas and the Netherlands on time to energize is measured in years.

The AI Factories initiative,[35] planned AI Gigafactories, and the Chips Act 2’s harmonized permitting regime for semiconductor facilities all belong in this category. They are more promising than sovereignty filters because they address a real constraint. The supply-chain bottlenecks discussed earlier will still determine how quickly these initiatives translate into capability. And, as the Draghi report and subsequent commentary make clear, capacity additions will not solve Europe’s broader competitiveness problem if the surrounding regulatory environment keeps imposing compounding compliance costs.[36]

The second category is capital and legal form. European venture capital remains structurally smaller than its U.S. counterpart at every stage after Series A. The “Delaware flip”—that is, European startups reincorporating in the United States to access more predictable corporate law and deeper capital markets—continues to drain European value capture.

The Commission’s March 18 proposal for an EU Inc., or 28th-regime, corporate form responds to that problem.[37] The proposal would offer 48-hour digital registration, €1 minimum capital, and a harmonized corporate form across the single market. It followed an October 2024 grassroots petition signed by more than 22,000 European founders and investors, including Stripe’s Patrick Collison, Wise’s Taavet Hinrikus, Supercell’s Ilkka Paananen, Wolt’s Miki Kuusi, and Pigment’s Eléonore Crespo, along with Index Ventures, Sequoia, and Seedcamp.[38] The economic case is strong, and the proposal belongs squarely inside the build agenda this brief endorses.

But the EU Inc. proposal, as drafted, appears too weak to do the job. Article 4 of COM(2026) 321 final provides that “matters that are not covered by this Regulation or by the articles of association shall be governed by national law,” with each member state designating which national legal form fills the gaps. As Luca Enriques, Casimiro A. Nigro, and Tobias H. Tröger put it, the result is “twenty-seven different versions of the EU Inc., each with its own national legal substrate.”[39]

That is a serious problem. The instrument meant to give European founders Delaware-style legal certainty risks becoming yet another optional regime layered on top of national variation. The dilution reflects the EU’s structural reluctance to override member-state legal autonomy, even where the Commission and European founders broadly agree on the goal. The pattern is not new; the same critique applied to EU industrial policy in 2022.[40]

The harder question is how the build agenda relates to the sovereignty agenda. The two are usually treated as complementary: Invest in European capacity, restrict access to non-European capacity, and the bloc will rise. The evidence above does not support that pairing.

A categorical-sovereignty interpretation of CAIDA would impose a capability tax on the very European users a build agenda is supposed to help. The EU Inc. dilution shows that the EU’s instinct to compromise with member-state autonomy is already weakening the build side, even where consensus on the destination is unusually strong.

Those two impulses—sovereignty-as-prohibition and build-side dilution—are different symptoms of the same underlying problem. The EU’s policy machinery tends to overdeliver on prohibition and underdeliver on building. A serious build agenda has to be defended against both.

In practice, “build, not prohibit” means treating the EU’s job as removing structural barriers to European AI-cloud capacity: power, permitting, capital, legal form, talent, and regulatory friction. It does not mean restricting European users’ access to the only frontier-compute stack that currently exists.

The Tech Sovereignty package that the Commission adopts on May 27, and the member-state negotiations that run through Q4 2027, will not be the last chance to make that pivot. But the posture set now will shape the next legislative cycle. On the present evidence, the prohibition side is much better resourced than the build side. The question worth pressing in the days before adoption and in the months of negotiation that follow is whether the EU is prepared to defend the build agenda not only against sovereignty-as-prohibition, but also against its own instinct to dilute the build instruments it proposes.

VIII. Conclusion: Build, Don’t Ration

Categorical exclusion of non-EU compute and APIs would impose a capability tax on the European users CAIDA is supposed to serve, without delivering the legal immunity it implies. A build-not-prohibit alternative would require a different posture from the Commission, member states, and the EU’s existing data-protection toolkit.

A. Don’t Tax the Frontier

Do not adopt EU-wide categorical sovereignty rules at any level of the AI stack. EU-level sovereignty rules should be risk-based, not categorical, across both public-sector and commercial workloads. Member states that judge stricter measures necessary for their own public-administration uses should remain free to adopt them under subsidiarity. But the EU-level instrument should not bind commercial users—or any single workload category—to provider-nationality or “effective control” tests across the bloc. The CISPE position and the French line both push in the opposite direction. The Council negotiation through Q4 2027 is where this will be settled.

Do not treat “immunity from non-EU law” as a categorical exclusion test, whether in EUCS High+, CAIDA, or any successor sovereignty-certification scheme. The OVH order requiring release of EU user data to Canadian authorities shows that EU headquarters alone does not dissolve extraterritorial legal reach. A categorical test that promises immunity it cannot deliver would impose capability costs on European users without buying the legal protection its proponents have in mind.

B. Build What Europe Lacks

Pursue the structural reforms that a build-not-prohibit posture actually requires. Two areas matter immediately.

The first is corporate law, where the 28th-regime, or EU Inc., proposal risks dilution through deference to member-state legal autonomy—recapitulating earlier patterns in EU industrial policy.

The second is the financial single market. Europe needs reforms that enable larger later-stage allocations from institutional and venture investors. Without deeper capital markets, Europe can simplify incorporation and still watch its most promising firms leave for jurisdictions with more predictable law, deeper financing, and faster scaling paths.

For genuinely sensitive workloads, rely on the instruments that already exist. Article 9 of the GDPR, defined national-security exceptions, and the proportionality discipline that applies to public-sector procurement already supply the answer for the narrow set of cases where residual extraterritorial concern is genuinely live. In those cases, the right level of protection is workload-specific and legal-procedural, not provider-nationality exclusion.

Stay inside the Western capability-controls architecture. The Commission’s own Joint Research Centre concluded earlier this month that “digital sovereignty cannot be equated with autarky” and that economic statecraft should be “a careful calibration between insulation and interdependence, in which autonomy is pursued through selective rebalancing rather than full-fledged decoupling.”[41] The current trajectory of the CISPE position and the French line risks defining the EU as an autonomous third pole, which in practice would place it downstream of both the Western and Chinese stacks. Member states supporting the package should resist that framing in Council negotiations.

C. Match Rules to Workloads

CAIDA’s definitions should distinguish between cloud services and AI services, and within AI services between frontier and downstream tiers. Frontier services should include base-model pretraining, frontier-scale inference, and access to frontier proprietary-model APIs. Downstream services should include fine-tuning, open-weight-model inference, and on-premises serving.

If any sovereignty-grade tests survive the negotiation, workload calibration will matter more than the abstract question of whether to impose such tests at all. The proposal’s definitions would benefit from explicit attention to this tier distinction.

Public-procurement frameworks should also decouple legal-sovereignty assessment from operational-capability assessment. SEAL measures the first. An independent benchmark of the ClusterMAX type, or a Commission-developed equivalent, would measure the second.

The Cloud Sovereignty Framework procurement awarded three SEAL-3 contracts and one SEAL-2 contract, but only one of the four providers ranks in the top three tiers of independent operational benchmarking. Procurement design should surface that divergence explicitly when a workload’s value depends on AI-compute capability, not legal-operational sovereignty alone.

[1] Maximilian Henning, EU’s Flagship Tech Sovereignty Package Delayed Again, Euractiv (Mar. 17, 2026), https://www.euractiv.com/news/eus-flagship-tech-sovereignty-package-delayed-again

[2] Cloud and AI Development Act, Eur. Parl. Legisl. Train Schedule (updated Apr. 20, 2026), https://www.europarl.europa.eu/legislative-train/theme-a-new-plan-for-europe-s-sustainable-prosperity-and-competitiveness/file-cloud-and-ai-development-act.

[3] Daniel Nishball et al., AI Value Capture: The Shift to Model, SemiAnalysis (Apr. 30, 2026), https://newsletter.semianalysis.com/p/ai-value-capture-the-shift-to-model.

[4] Joint Letter from European Cloud CEOs, Joint Call for a Sovereign and Resilient European Cloud Policy, Cloud Infrastructure Services Providers in Europe (CISPE) (Mar. 17, 2026), https://www.cispe.cloud/website_cispe/wp-content/uploads/2026/03/Cloud-CEOs-Joint-Letter-March-17-2026-FINAL.pdf.

[5] Kai Nicol-Schwarz, EU Mulls Restricting Use of U.S. Cloud for Sensitive Government Data: Sources, CNBC (May 7, 2026), https://www.cnbc.com/2026/05/07/eu-commission-cloud-sensitive-data.html.

[6] Eur. Comm’n, Commission Advances Cloud Sovereignty Through Strategic Procurement (Apr. 17, 2026), https://commission.europa.eu/news-and-media/news/commission-advances-cloud-sovereignty-through-strategic-procurement-2026-04-17_en.

[7] Dan Robinson, Europe Picks 4 Sovereign Cloud Providers, But One Has Google, The Register (Apr. 20, 2026), https://www.theregister.com/2026/04/20/europe_picks_4_sovereign_cloud.

[8] Jordan Nanos et al., ClusterMAX 2.1: How Much Do GPU Clusters Really Cost?, SemiAnalysis (Apr. 20, 2026), https://newsletter.semianalysis.com/p/how-much-do-gpu-clusters-really-cost.

[9] Nishball et al., supra note 3.

[10] Eur. Comm’n, SEAL Framework (Oct. 2025), https://commission.europa.eu/document/download/09579818-64a6-4dd5-9577-446ab6219113_en.

[11] Agence Nationale de la Sécurité des Systèmes d’Information (ANSSI), SecNumCloud — Référentiel Pour les Fournisseurs de Services Cloud, https://cyber.gouv.fr/secnumcloud-pour-les-fournisseurs-de-services-cloud (last visited May 14, 2026).

[12] Nebius Group, Wikipedia, https://en.wikipedia.org/wiki/Nebius_Group (last visited May 14, 2026).

[13] AI Provider Leaderboard, Artificial Analysis, https://artificialanalysis.ai/leaderboards/providers (last visited May 14, 2026).

[14] Cluster TCO and Goodput Methodology, ClusterMAX, https://www.clustermax.ai/tco (last visited May 14, 2026).

[15] Ivan Chiam et al., The Great AI Silicon Shortage, SemiAnalysis (Mar. 12, 2026), https://newsletter.semianalysis.com/p/the-great-ai-silicon-shortage.

[16] Dylan Patel et al., Memory Mania: How a Once-in-Four-Decades Supercycle Began, SemiAnalysis (Feb. 6, 2026), https://newsletter.semianalysis.com/p/memory-mania-how-a-once-in-four-decades.

[17] Nishball et al., supra note 3.

[18] Daniel Nishball et al., The Great GPU Shortage: Rental Capacity, SemiAnalysis (Apr. 2, 2026), https://newsletter.semianalysis.com/p/the-great-gpu-shortage-rental-capacity.

[19] H100 1-Year Contract Price Index, SemiAnalysis, https://semianalysis.com/gpu-pricing-index (last visited May 14, 2026).

[20] Nishball et al., supra note 3.

[21] Eur. Comm’n, Strategic Procurement, supra note 6.

[22] Cluster TCO and Goodput Methodology, supra note 14.

[23] Nishball et al., supra note 3.

[24] Mikołaj Barczentewicz, The EU’s GDPR ‘Fix’ Misses the Point Entirely, Truth on the Mkt. (June 24, 2025), https://truthonthemarket.com/2025/06/24/the-eus-gdpr-fix-misses-the-point-entirely.

[25] Kristian Stout & Mikołaj Barczentewicz, How Not to Use Industrial Policy to Promote Europe’s Digital Sovereignty, Truth on the Mkt. (Oct. 5, 2022), https://truthonthemarket.com/2022/10/05/how-not-to-use-industrial-policy-to-promote-europes-digital-sovereignty.

[26] Eur. Union Agency for Cybersecurity (ENISA), European Cybersecurity Certification Scheme for Cloud Services (EUCS) (Dec. 22, 2020), https://www.enisa.europa.eu/publications/eucs-cloud-service-scheme.

[27] Cloud and AI Development Act, supra note 2.

[28] Henning, supra note 1.

[29] Nicol-Schwarz, supra note 5.

[30] U.S. Bureau of Indus. & Sec., Press Release, Commerce Implements New Export Controls on Advanced Computing and Semiconductor Manufacturing Items to the People’s Republic of China (Oct. 7, 2022), https://www.bis.gov/press-release/commerce-implements-new-export-controls-advanced-computing-semiconductor-manufacturing-items-peoples.

[31] Framework for Artificial Intelligence Diffusion, 90 Fed. Reg. 4543 (Jan. 15, 2025), https://www.federalregister.gov/documents/2025/01/15/2025-00636/framework-for-artificial-intelligence-diffusion.

[32] Dylan Patel et al., DeepSeek Debates, SemiAnalysis (Jan. 30, 2025), https://newsletter.semianalysis.com/p/deepseek-debates.

[33] Samuele Fratini et al., Digital Sovereignty as a Geopolitical Strategy: Navigating Dependencies, Power, and Democratic Resilience in a Changing Global Order, JRC146878, Eur. Comm’n (May 8, 2026), https://publications.jrc.ec.europa.eu/repository/handle/JRC146878.

[34] Shanzay Pervaiz, What Canada’s King vs. OVH Case Reveals, and Affirms, About Cross-Border Data Access, Privacy Across Borders (Feb. 3, 2026), https://privacyacrossborders.org/2026/02/03/what-canadas-king-vs-ovh-case-reveals-and-affirms-about-cross-border-data-access.

[35] AI Factories, Eur. Comm’n Digit. Strategy, https://digital-strategy.ec.europa.eu/en/policies/ai-factories (last visited May 14, 2026).

[36] Mario Zúñiga, Draghi Report Highlights Why to Be Wary of the ‘Brussels Effect’, Truth on the Mkt. (Sept. 26, 2024), https://truthonthemarket.com/2024/09/26/draghi-report-highlights-why-to-be-wary-of-the-brussels-effect.

[37] Eur. Comm’n, Press Release IP/26/614, Commission Presents Proposal for EU Inc.—Unlocking the Full Potential of the Single Market for Europe’s Entrepreneurs (Mar. 18, 2026), https://ec.europa.eu/commission/presscorner/detail/en/ip_26_614.

[38] EU Inc. Petition, EU-Inc.org, https://www.eu-inc.org/petition (last visited May 14, 2026).

[39] Luca Enriques, Casimiro A. Nigro & Tobias H. Tröger, Why the 28th Regime Proposal Falls Short of Europe’s Challenge, Oxford Bus. L. Blog (Mar. 2026), https://blogs.law.ox.ac.uk/oblb/blog-post/2026/03/why-28th-regime-proposal-falls-short-europes-challenge.

[40] Stout & Barczentewicz, supra note 25.

[41] Fratini et al., supra note 33.

Executive Summary

Payment-card networks are two-sided platforms that create value by connecting cardholders and merchants. Interchange fees—the small amounts retained by cardholders’ banks when payment cards are used—serve as balancing payments that help optimize participation on both sides of the market. Networks therefore set differentiated interchange fees that vary by card type, transaction channel, merchant category, geography, and whether the transaction is domestic or cross-border. Those differences reflect real variation in cost, fraud risk, demand, and transaction value.

Many governments have imposed price controls on interchange fees, compressing those differentials and impairing networks’ ability to balance the two sides of the market. The consequences are well documented: reduced card rewards, higher cardholder or account fees, higher interest margins, limited pass-through of merchant savings to consumers, and diminished investment in payment innovation. Consumers bear much of the harm, but merchants may suffer as well.

This issue brief focuses on a less-studied effect: the application of interchange-fee caps to cross-border card payments. Cross-border transactions are costlier and riskier than domestic transactions because they involve higher fraud rates, currency-conversion costs, cross-jurisdictional compliance, and more complex processing and settlement. When regulators apply caps calibrated to domestic conditions to foreign-card transactions, they create a mismatch between issuer revenue and issuer cost.

That mismatch can lead issuers to tighten authorization criteria, increasing the likelihood that legitimate cross-border transactions will be declined. Industry evidence showing materially lower authorization rates for cross-border payments than for domestic ones is consistent with that mechanism.

The harm is especially acute for high-value tourism spending and cross-border e-commerce. Both depend heavily on cards, including credit and premium cards, and both involve transaction types that carry higher fraud and authorization risks. Cross-border interchange caps therefore threaten not only issuer economics, but also consumers’ ability to transact and merchants’ ability to complete sales. The paper concludes that policymakers should exempt cross-border transactions from interchange-fee-cap regimes or, at minimum, adopt separate, higher caps that reflect the genuine incremental costs those transactions impose.

I.            Interchange-Fee Regulation and Cross-Border Distortions

Payment cards are integral to modern economies. Over the past several decades, card networks have transformed how consumers buy goods and services, replacing cash and checks with faster, safer, and more convenient electronic payments. The growth of cross-border e-commerce and international travel has extended that transformation across national borders, making payment cards a central medium for international consumer transactions.

Governments around the world have increasingly intervened in the pricing of payment-card services by imposing price controls on interchange fees—the small amounts retained by cardholders’ banks when consumers use payment cards. Regulators typically justify these caps by claiming that interchange fees are “too high” and that lower fees would benefit merchants and, ultimately, consumers. The economic literature shows why that claim is incomplete at best.[1]

Payment-card networks are two-sided platforms that serve both cardholders and merchants. Platform efficiency and welfare depend not only on the total level of fees, but also on how those fees are allocated between the two sides of the market.[2] Interchange fees operate as balancing payments, enabling networks to subsidize participation on the more price-sensitive side—typically cardholders—while recovering more of the system’s costs from the side with less elastic demand.[3]

The theoretically optimal interchange fee varies across jurisdictions, card types, merchant categories, transaction channels, and domestic versus cross-border transactions. Those differences reflect variation in fraud risk, demand elasticities, issuer costs, regulatory obligations, and the value each side derives from network participation. In practice, network-set interchange fees can only approximate the optimal schedule. But rigid price controls make that approximation worse. A cap calibrated for domestic debit-card transactions at grocery stores is unlikely to fit cross-border credit-card transactions at hotels. No regulator has the information needed to set the right fee for every transaction type.

This informational problem has both static and dynamic consequences. Interchange-fee caps compress differentiated fee schedules, distort the allocation of costs and benefits between merchants and cardholders, and predictably induce issuers to recover lost revenue through reduced rewards, higher account fees, and diminished cardholder benefits. They also weaken incentives to invest in fraud detection, tokenization, real-time authorization infrastructure, and other payment technologies that are especially important for cross-border transactions. And when caps apply unevenly across network models, they can distort competition by pushing consumers, merchants, and fintech firms toward less-regulated products or organizational forms.

This issue brief focuses on a dimension of interchange-fee regulation that has received too little attention: the application of interchange-fee caps to cross-border transactions. Cross-border card payments are inherently more costly and risky than domestic payments. They involve higher fraud rates, currency-conversion costs, additional processing and settlement costs, and compliance with multiple regulatory regimes. When regulators calibrate interchange caps to domestic cost structures, they create a systematic mismatch between the revenue issuers receive and the costs they bear on cross-border transactions.

That mismatch grows when the caps apply to transactions by foreign cardholders whose issuers operate under cost structures, fraud environments, and competitive conditions different from those the regulator considered. At the margin, the predictable result is more declines of legitimate cross-border transactions—a prediction consistent with industry evidence showing materially lower authorization rates for cross-border payments than for domestic ones.

A.             Interchange Fees Reflect Transaction Differences

Payment networks do not set a single, uniform interchange fee. Instead, they maintain complex fee schedules that vary across multiple dimensions, including card type (debit versus credit, standard versus premium, consumer versus commercial), transaction channel (point-of-sale versus online, card-present versus card-not-present), merchant category, and whether the transaction is domestic or cross-border. These distinctions reflect differences in costs, risks, and demand conditions across transaction types.

The distinction between debit and credit cards illustrates the point. Interchange fees fund a range of cross-side subsidies—benefits provided to cardholders and financed, in part, by merchant-paid fees. For credit cards, those subsidies include rewards programs, such as cashback, points, and miles; the interest-free grace period between purchase and repayment; and issuers’ assumption of default and collection risk.

For debit cards, interchange revenue has historically funded rewards programs and enabled banks to offer free or low-cost checking accounts to consumers with lower balances. In effect, interchange revenue cross-subsidizes account holders who might otherwise be unprofitable to serve. When governments impose price controls on debit-card interchange fees, banks often respond by curtailing rewards programs, increasing checking-account fees, and raising minimum-balance requirements for free accounts.[4]

Credit-card transactions generally impose higher costs on issuers than debit-card transactions because they involve extending credit, guaranteeing zero liability for fraud, and providing more substantial cardholder benefits. Credit-card interchange fees therefore tend to exceed debit-card interchange fees. Premium and commercial cards typically carry even higher fees because they offer enhanced benefits, such as travel insurance and extended payment terms, that attract high-spending cardholders whose transactions merchants especially value.

Cross-border transactions also typically carry higher interchange fees than domestic transactions. That differential reflects measurable cost differences. Cross-border transactions involve currency-conversion risk, higher fraud rates, compliance with multiple regulatory regimes, and additional processing and settlement costs. Higher interchange fees compensate issuers for those incremental costs and help ensure that networks can authorize transactions that might otherwise be uneconomic to process.

B.             Price Controls Distort Interchange-Fee Structures

When regulators impose interchange-fee price controls, they compress the differentiated fee structures that payment networks have developed to balance costs and incentives across heterogeneous transaction types. The European Union’s Interchange Fee Regulation (IFR), for example, caps consumer debit-card interchange fees at 0.20% and consumer credit-card interchange fees at 0.30% of transaction value for both domestic and intra-European Economic Area (EEA) cross-border transactions.[5] Those caps sit far below pre-regulation interchange fees, which ranged from roughly 0.5% to more than 2%, depending on card type, jurisdiction, and transaction characteristics. Most important for present purposes, the IFR eliminates the cross-border differential.

The core problem is informational. Regulators cannot know the “optimal” interchange fee because it depends on variables that differ across jurisdictions, card types, merchant categories, and transaction channels. Payment networks have spent decades refining interchange-fee schedules to account for those differences. Price controls—and especially rigid caps—replace that flexible system with an arbitrary ceiling. Any fee above the cap disappears, limiting networks’ ability to calibrate fees to differing transaction conditions.

The problem becomes more acute in cross-border transactions. A price control designed around domestic conditions in one country may roughly reflect that country’s costs, competitive dynamics, and consumer preferences. Applying the same cap to transactions involving foreign cardholders creates a larger mismatch. Foreign issuers operate under different cost structures, face different fraud risks, and serve consumers with different spending patterns. In many cases, the regulated fee will fall below both the issuer’s domestic interchange rate and the higher costs associated with cross-border authorization and settlement.

Economic models of payment networks as two-sided markets underscore these risks. Rong Ding and Julian Wright, for example, analyze interchange fees in a framework that allows platforms to differentiate pricing across merchants and jurisdictions.[6] They find that restricting such differentiation through a single interchange fee can reduce welfare because it removes an important mechanism for aligning prices with heterogeneous costs and benefits. Price controls do not necessarily eliminate fee differentiation altogether, although very low caps may do so. Even where some differentiation remains, however, price controls substantially constrain networks’ ability to price efficiently and are therefore likely to generate the distortions Ding and Wright identify.

The theoretical literature also predicts compensating behavior throughout the payments ecosystem. Because issuing banks depend in part on interchange revenue, price controls predictably lead them to adjust other prices and expenditures. Banks may increase annual card fees, reduce rewards, scale back insurance and ancillary benefits, or reduce investment in innovation.[7]

The empirical evidence strongly supports those predictions. In our 2022 review paper, Geoffrey Manne, Todd Zywicki, and I document that, in every jurisdiction that has imposed interchange-fee price controls—including the European Union, Australia, Spain, and the United States—issuing banks responded with some combination of reduced card rewards, higher annual card fees, increased checking-account fees, and higher interest margins. At the same time, merchants generally did not meaningfully pass through their savings to consumers.[8]

C.             Price Controls Reduce Incentives to Innovate

Interchange-fee price controls do more than reallocate costs between cardholders and merchants. They also affect long-term investment and innovation in payment systems. Interchange revenue funds not only cardholder rewards, but also issuer investments in fraud detection, tokenization, real-time authorization infrastructure, and other technologies that benefit both sides of the market.[9] When regulators compress that revenue, issuers have fewer resources and weaker incentives to invest in such improvements.

Several important payment-card technologies emerged specifically to address cross-border transaction problems and may not have been developed absent the ability to monetize those investments through interchange revenue. For example:

• The EMV chip standard was developed in significant part to improve cross-border interoperability.[10]
• EMV 3-D Secure, a machine-learning-enhanced fraud-detection and prevention system, was developed largely in response to elevated cross-border fraud risks.[11]
• Dynamic Currency Conversion—which allows cardholders to view and pay for purchases in their home currency—was designed specifically for cross-border transactions.[12]

Markus Reisinger and Hans Zenger show that interchange caps set at the level of the “tourist test”—the point at which a hypothetical merchant is indifferent between accepting a card payment or cash—produce investment levels below the social optimum.[13] Such caps fail to account for the dynamic benefits of innovation financed by interchange revenue. In our review paper, we similarly document slower investment in payment technologies in jurisdictions that imposed interchange-fee price controls.[14]

Price controls also redirect innovation toward regulatory arbitrage rather than consumer welfare, security, speed, or product quality. The Durbin Amendment illustrates the point. The law capped debit-card interchange fees only for issuers with more than $10 billion in assets, while exempting smaller issuers. That structure created a regulatory wedge.

Despite possessing greater scale, infrastructure, compliance capacity, and data resources, covered banks became less attractive partners for fintech firms because interchange caps limited their ability to recover investments in product development, rewards, fraud prevention, and customer acquisition. Fintech firms therefore structured new debit-card and banking products through partnerships with smaller, Durbin-exempt banks, which could sustain higher interchange revenue than comparable programs issued by covered institutions.

The result was a distortion in both the direction and organization of innovation. Entrepreneurs and investment capital flowed toward exempt-charter structures and product designs that preserved interchange economics, rather than toward the partnerships or technologies that otherwise would have been most efficient, secure, or valuable to consumers.

These innovation effects are especially important in cross-border transactions, where fraud-prevention and real-time authorization costs are highest. If interchange revenue cannot sustain continued investment in those systems, the quality and security of cross-border payment processing will deteriorate over time. That dynamic would compound the direct effects of fee compression on authorization rates.

D.            Price Controls Distort Payment-Network Competition

Interchange-fee price controls can shift demand across payment networks, in addition to compressing issuer revenue. Where regulators primarily cap four-party schemes, consumers who value rewards programs and premium-card benefits have incentives to migrate toward less-regulated alternatives, especially three-party cards, where those benefits remain more sustainable.

Australia provides a useful example. Following the country’s initial interchange-fee reforms, three-party cards and companion-card arrangements gained market share. That share later declined after regulators extended comparable restrictions to American Express companion-card structures.[15] The European Union saw a similar pattern. Under the IFR, co-branded American Express products initially enjoyed a relative advantage over regulated four-party schemes. That advantage narrowed after the European Court of Justice held in 2018 that certain co-branded and agent-issued three-party products also fell within the interchange-cap regime.[16]

The broader implication is that interchange-fee price controls distort not only pricing, but also competition among network models. As the Durbin Amendment experience illustrates, such regulation encourages substitution toward less-regulated products and organizational forms, rather than competition based on efficiency, security, or consumer value.

II.            Cross-Border Transactions Carry Higher Costs and Risks

Cross-border card payments have grown rapidly with international tourism and e-commerce, but they remain inherently more costly and risky than domestic transactions. They involve currency-conversion and exchange-rate risk, overlapping regulatory and anti-money-laundering obligations, more complex settlement procedures, and substantially higher fraud exposure.

Those differences matter for interchange regulation. Caps calibrated to domestic conditions are unlikely to cover the higher costs of foreign-card transactions, increasing the risk that issuers will decline legitimate cross-border payments at the margin.

A.             Cross-Border Card Payments Have Expanded Rapidly

Cross-border card payments have grown rapidly alongside the expansion of international tourism and e-commerce—and, in many respects, have helped enable both. The Federal Reserve reports that, in 2022, U.S.-issued cards were used for 7.5 billion cross-border transactions totaling $470 billion, while foreign-issued cards were used for 2.7 billion transactions in the United States totaling $180 billion.[17] Globally, Visa reported 15% growth in cross-border volume in fiscal year 2025 across 258 billion total transactions, while Mastercard reported 18% growth in cross-border volume across 175 billion switched transactions.[18] Together, these figures reflect the continued expansion of international card usage.

Figure 1 shows that global outbound cross-border card volume has more than tripled over the past decade.

FIGURE 1: Global Outbound Cross-Border Card Volume

SOURCE: Author’s estimate (see note for details)[19]

B.             Cross-Border Transactions Carry Higher Fraud Costs

Cross-border card transactions impose several costs beyond those associated with domestic payments, including currency-conversion and exchange-rate risk, compliance with multiple regulatory and anti-money-laundering regimes, and more complex settlement procedures. Fraud, however, is the single largest additional cost driver. Multiple authoritative data sources show that fraud rates on cross-border card transactions substantially exceed domestic fraud rates, although the scale of the disparity varies across jurisdictions.

European Union. The disparity is especially pronounced in the euro area. The European Central Bank’s 2023 report on card fraud in the Single Euro Payments Area (SEPA) found that, between 2016 and 2021, cross-border transactions accounted for only 10% to 11% of total card-transaction value, but generated 63% to 65% of total fraud losses by value.[20] The joint ECB-European Banking Authority 2025 Report on Payment Fraud suggests that those proportions have remained similar or may have increased.[21]

Australia. AusPayNet data for 2024 show total card fraud losses of A$899 million, of which A$495.5 million—roughly 55%—involved Australian-issued cards used overseas. Card-not-present (CNP) fraud accounted for A$454 million of those overseas losses. Fraud involving foreign-issued cards used in Australia added another A$87 million.[22]

These data show that cross-border card transactions generate fraud costs several times higher than domestic transactions on a per-transaction basis. An interchange-fee cap calibrated to domestic fraud costs will therefore tend to undercompensate issuers for cross-border transactions, creating incentives to restrict authorization of higher-risk foreign transactions.

III.            Fee Caps Can Increase Cross-Border Declines

The preceding sections establish two key points: Interchange fees help balance two-sided payment markets, and the efficient fee varies by card type, transaction channel, and geography. Cross-border transactions also carry higher costs than domestic transactions, especially fraud costs.

This section connects those points to transaction authorization. When regulators cap interchange fees for cross-border transactions, they compress the revenue issuers use to offset higher fraud, compliance, currency-conversion, and settlement costs. That mismatch can make marginal cross-border transactions uneconomic to approve, especially in card-not-present channels, higher-risk merchant categories, and transactions involving unfamiliar foreign acquirers.

The result is predictable: Issuers have stronger incentives to tighten authorization criteria, increasing the risk that legitimate cross-border transactions will be declined. Industry evidence showing materially lower authorization rates for cross-border payments is consistent with that mechanism.

A.             Fee Compression Can Reduce Authorization Incentives

Each time a cardholder attempts a purchase, the issuing bank must decide in real time whether to authorize the transaction. Authorization means accepting potential fraud liability, funding the transaction in the case of credit cards, and bearing the processing costs of clearing and settlement. The issuer’s incentive to approve the transaction depends on whether expected marginal revenue—principally the interchange fee—exceeds expected marginal cost. If it does, authorization has a positive expected return. If not, the issuer faces a marginal loss.

For domestic transactions, local issuers can adapt to price-controlled interchange fees by adjusting other fees and expenses. Cross-border transactions are different. Fraud costs are several times higher than domestic levels, and issuers typically set interchange fees based on their own domestic conditions plus an additional amount to cover cross-border liabilities. A price control set by a foreign jurisdiction will therefore often fall short of the issuer’s costs.

The European Union illustrates the problem. The IFR caps interchange fees at 0.20% for debit cards and 0.30% for credit cards. Yet the European Central Bank’s 2021 data—the most recent comparable data available—show that cross-border transactions accounted for roughly 11% of card-transaction value but 63% of fraud losses by value.[23] That implies an average fraud-loss rate for cross-border transactions roughly 14 times that of domestic transactions. A 0.30% interchange fee on a €100 cross-border credit-card transaction yields only €0.30, which must cover elevated fraud risk, currency-conversion costs, cross-jurisdictional compliance, international settlement complexity, and cardholder benefits such as rewards and insurance.

One likely issuer response is to tighten authorization criteria for transactions where the shortfall is greatest: cross-border transactions, especially card-not-present transactions, higher-risk merchant categories, and purchases involving unfamiliar foreign acquirers. The issuer cannot recover the lost interchange revenue from the merchant, with whom it has no direct relationship. Nor can it easily recover the shortfall from the cardholder at the point of sale. In the short run, the issuer’s only realistic margin of adjustment is the authorization decision itself.

Over time, issuers may also seek to recover lost revenue through other channels, including annual fees, interest charges, and reduced cardholder benefits.

B.             Cross-Border Transactions Face Higher Decline Rates

No public dataset isolates the specific contribution of interchange-fee caps to cross-border decline rates—a gap in the empirical literature that warrants further study. But industry evidence consistently shows that cross-border transactions face substantially higher decline rates than domestic transactions. Payment processors and gateway aggregators regularly report materially higher authorization rates for domestic payments than for cross-border payments.[24]

Several factors drive that gap. Higher fraud risk on cross-border transactions triggers more conservative issuer-authorization algorithms. Currency mismatches between the cardholder’s home currency and the transaction currency can raise flags in automated fraud-detection systems. Issuer unfamiliarity with foreign acquirers and merchants can also reduce the confidence score assigned to a transaction.

Interchange-fee caps intensify these pressures. When caps compress the interchange revenue available to offset the higher costs of cross-border authorization, the issuer’s expected return on marginal cross-border transactions turns negative earlier in the risk distribution. As a result, transactions that would have been authorized under market-determined interchange fees may be declined under a capped regime. All else equal, jurisdictions that cap interchange fees should therefore see a wider authorization gap between domestic and cross-border transactions than jurisdictions that do not.

Research by Checkout.com and Oxford Economics found that merchants lose between 1% and 2.1% of revenue to false declines—legitimate transactions that issuer fraud-screening systems erroneously reject—and that cross-border payment complexity exacerbates the problem.[25]

IV.            Tourism and E-Commerce Bear the Heaviest Costs

The costs of cross-border interchange-fee caps do not fall evenly. They are most likely to affect transactions where issuer costs, fraud risk, and transaction values are highest—especially tourism spending and cross-border e-commerce.

Both categories rely heavily on card payments, including credit and premium cards, and often involve either foreign-card-present or card-not-present transactions. Those are precisely the transactions most vulnerable to tighter authorization standards when interchange-fee caps compress issuer revenue.

The result is more than a marginal inconvenience. In tourism, false declines can disrupt travel, cost merchants high-value sales, and damage the customer experience. In e-commerce, they can translate into significant lost commerce, with smaller merchants likely bearing a disproportionate share of the burden.

A.             Tourism Depends on Reliable Cross-Border Payments

The U.N. World Tourism Organization estimates that global tourism receipts reached $1.6 trillion in 2024.[26] Surveys suggest that tourists strongly prefer card payments, which offer greater convenience and lower risk than cash.[27] Except for domestic tourists, travelers with local bank accounts, or transactions acquired in the cardholder’s home jurisdiction—such as prepaid hotel bookings—tourism-related card payments are generally cross-border transactions.

Higher-value tourism spending often occurs on credit cards, including premium cards that offer rewards, no foreign-transaction fees, travel protections, and purchase protection.[28] When interchange-fee price controls apply to those transactions, they reduce issuers’ ability to recover the higher costs associated with international payments, including fraud risk, cross-jurisdictional compliance, and settlement complexity. Intra-European Economic Area transactions offer the clearest example because the EU’s IFR subjects them to the same caps as domestic transactions. The likely effect is not that every such transaction becomes uneconomic, but that margin compression increases pressure to tighten authorization for higher-risk transactions at the margin.

That friction falls directly on tourists, who are more likely to experience card declines, and on tourism-dependent merchants, who lose sales and may suffer reputational harm from the poor customer experience associated with declined transactions.

Given consumers’ strong preference for using cards abroad, a large share of the roughly $1.6 trillion in annual tourism spending likely occurs through card payments. Even a modest increase in decline rates would therefore translate into billions of dollars in lost commerce.

The welfare loss is likely larger for premium-card holders, who are disproportionately high spenders and thus among the tourists whose spending generates the greatest economic benefit. A declined $2,000 hotel booking or $500 restaurant bill imposes a much greater economic loss than a declined $20 transit fare.

B.             E-Commerce Magnifies the Costs of False Declines

In 2024, eMarketer projected that global retail e-commerce would reach $7.5 trillion in 2026.[29] UNCTAD, meanwhile, estimated that cross-border transactions accounted for 25% of retail e-commerce in 2019. If that share remains similar, cross-border e-commerce would total roughly $1.9 trillion.[30]

A very large share of those transactions are card-not-present (CNP) payments, which carry inherently higher fraud risk and trigger more conservative issuer-authorization policies. The European Central Bank reports that CNP fraud accounted for €1.28 billion of the €1.53 billion in total euro-area card fraud in 2021.[31] Online fraud losses have also risen rapidly: Juniper Research estimates that global e-commerce fraud losses grew from $17.5 billion in 2020 to $56 billion in 2025, a compound annual growth rate of more than 26%.[32]

When interchange-fee caps reduce the revenue issuers receive for authorizing CNP cross-border transactions, while fraud costs continue to rise, issuers rationally tighten authorization standards. That means accepting fewer marginal transactions where the fraud probability is elevated. The result is a higher decline rate in cross-border e-commerce, where every transaction is simultaneously CNP, cross-border, and subject to interchange-fee compression.

With cross-border e-commerce valued at roughly $1.9 trillion, even small increases in decline rates caused by interchange-fee compression would represent substantial losses in foregone commerce.

These losses also have distributional consequences. Smaller merchants, which often lack the technical sophistication to deploy advanced fraud-prevention tools or maintain relationships with multiple acquirers, are likely to face higher decline rates than large platforms that can negotiate bespoke authorization arrangements.

 V.            Policy Implications and Recommendations

The evidence assembled here points to several conclusions with direct policy relevance.

First, the differentiated structure of interchange fees is not merely an artifact of market power. It reflects the heterogeneous costs, risks, and demand characteristics of different transaction types. Uniform caps compress those differences and remove a key instrument of efficient price discrimination in two-sided markets, with predictable negative consequences for transaction volume and welfare.

Second, cross-border interchange-fee caps impose costs that differ in kind and scale from domestic price controls. Higher fraud rates, currency-conversion costs, and regulatory-compliance burdens make cross-border transactions more expensive to authorize and settle. Caps applied to those transactions are therefore likely to do greater harm than domestic caps, including by increasing decline rates and harming both consumers and merchants.

Cross-border caps may also distort competition among network models. Four-party card issuers—such as Visa and Mastercard issuing banks—may respond by reducing jurisdiction-specific benefits on affected transactions, including rewards accrual, fee waivers, or approval rates. Those changes could make capped-jurisdiction transactions less attractive to cardholders and increase the relative appeal of less-regulated three-party products, such as American Express and Discover.

Third, policymakers should pay particular attention to the disproportionate impact on high-value tourism and e-commerce transactions. Tourism provides a critical source of foreign-exchange earnings for many economies, and payment frictions directly reduce tourist spending. Cross-border e-commerce expands market integration and consumer access to global goods and services. Regulatory interventions that impede these transactions impose costs well beyond the payment system itself.

Fourth, policymakers considering interchange-fee regulation should exempt cross-border transactions or, at minimum, establish separate and higher caps that reflect the cost differences documented here. The European Union’s decision to apply the same cap to intra-EEA cross-border transactions as to domestic transactions—while leaving extra-EEA transactions uncapped—creates an incoherent regime. It penalizes intra-European trade while failing to address the higher costs of transactions involving non-EEA cards.

Fifth, regulators should proceed cautiously before extending interchange-fee caps to card types or transaction categories that remain unregulated. The U.S. experience with the Durbin Amendment shows that even a narrowly targeted debit-card cap can produce significant distortions. Extending caps to credit cards, as some legislative proposals would do, would compress the remaining unregulated margin that helps the system absorb the costs of premium services, fraud prevention, and cross-border processing.

VI.            Conclusion

Interchange fees are the critical balancing mechanism in two-sided payment-card markets. Their differentiated structure—varying by card type, transaction channel, merchant category, and geography—reflects real differences in cost, risk, and demand across heterogeneous transactions. Price controls that compress those differences impair that balancing function, with especially serious consequences for cross-border payments.

This issue brief has shown that cross-border card transactions carry higher costs and face higher decline rates than domestic transactions. The gap between capped interchange fees and the actual cost of cross-border processing creates systematic pressure on issuers to tighten authorization standards, particularly for higher-risk transactions. These are not abstract welfare losses. They mean real purchases not made, real sales not completed, and real tourists and online shoppers denied the seamless payment experience modern card networks were built to provide.

Put differently, price controls on cross-border interchange fees become, at the margin, restrictions on cross-border commerce. They function as a form of de facto export restriction on suppliers of export-oriented consumer goods and tourism services.

They also create cross-border transfers with no democratic mandate. Issuers that authorize transactions subject to foreign price-controlled interchange fees effectively transfer value to acquiring banks—and perhaps indirectly to merchants—in the regulating jurisdiction, at the expense of their own cardholders. That dynamic raises the risk of retaliation by jurisdictions whose issuing banks are subject to foreign interchange caps.

The contagion risk is compounded by regulatory benchmarking. Once some jurisdictions cap cross-border interchange, other regulators may treat those controlled rates as evidence of the “market” level. That can produce a downward ratchet divorced from the heterogeneous costs of fraud, compliance, settlement, card type, merchant category, and transaction channel.

If such tit-for-tat responses spread, the consequences for cross-border credit-card payments could be severe. A single jurisdiction’s cap may induce issuers to absorb losses at the margin, impose destination-specific foreign-transaction fees, tighten authorization rules, or reduce cardholder benefits. If many jurisdictions follow suit, the result would not be a series of isolated local transfers. It would be a cumulative erosion of the global revenue base that funds cross-border card acceptance, fraud prevention, travel benefits, and reliable authorization.

Under those conditions, issuers would have stronger incentives to deny marginal transactions, impose destination-specific fees, withdraw premium benefits in capped jurisdictions, or steer cardholders toward less-regulated payment channels. The apparent local saving would come at the cost of a less reliable and less innovative cross-border payments system, with losses borne not only by issuers and cardholders, but also by the merchants that depend on high-value tourism and cross-border e-commerce.

The evidence strongly supports treating cross-border interchange-fee caps as a particularly costly form of price control. Regulators seeking to reduce merchant costs should consider alternatives—such as enhanced transparency requirements, tax incentives, interoperable real-time payment systems, or other programs to encourage digital-payments adoption and acceptance—that do not artificially compress the interchange-fee differentials on which efficient two-sided card markets depend.

At minimum, regulators should explicitly carve cross-border transactions out of any interchange-fee-cap regime or subject them to separate, higher caps that reflect the genuine incremental costs those transactions impose.

[1] Julian Morris, Todd Zywicki & Geoffrey Manne, The Effects of Price Controls on Payment-Card Interchange Fees: A Review and Update, Int’l Ctr. for L. & Econ. (2022), https://laweconcenter.org/wp-content/uploads/2022/03/Payments-2021-Lit-Review.pdf; Jean-Charles Rochet & Jean Tirole, Platform Competition in Two-Sided Markets, 1 J. Eur. Econ. Ass’n 990 (2003).

[2] Julian Wright, Optimal Card Payment Systems, 47 Eur. Econ. Rev. 587 (2003).

[3] A large body of economic literature—beginning with Baxter’s seminal 1983 contribution and later developed by Rochet and Tirole, Wright, and others—shows that, when payment networks set interchange fees, those fees generally reflect each side’s relative costs and demand elasticities. William F. Baxter, Bank Interchange of Transactional Paper: Legal and Economic Perspectives, 26 J.L. & Econ. 541 (1983); Jean-Charles Rochet & Jean Tirole, Two-Sided Markets: A Progress Report, 37 RAND J. Econ. 645 (2006).

[4] Morris, Zywicki & Manne, supra note 1, at 8–15 (documenting the decline in debit-card rewards and free checking after interchange-fee caps). See also Todd J. Zywicki, Geoffrey A. Manne & Julian Morris, Price Controls on Payment Card Interchange Fees: The U.S. Experience, Mercatus Ctr., George Mason Univ., Working Paper No. 14-18 (2014), https://papers.ssrn.com/sol3/papers.cfm?abstract_id=2446080; Vladimir Mukharlyamov & Natasha Sarin, Price Regulation in Two-Sided Markets: Empirical Evidence from Debit Cards, J. Fin. Econ. (2025), https://www.sciencedirect.com/science/article/pii/S0304405X25001023.

[5] Regulation (EU) 2015/751 of the European Parliament and of the Council of Apr. 29, 2015, on Interchange Fees for Card-Based Payment Transactions, arts. 3–4 [hereinafter IFR], https://eur-lex.europa.eu/eli/reg/2015/751/oj/eng.

[6] Rong Ding & Julian Wright, Payment Card Interchange Fees and Price Discrimination, 65 J. Indus. Econ. 39 (2017).

[7] Wright, supra note 2; Julian Wright, The Determinants of Optimal Interchange Fees in Payment Systems, 52 J. Indus. Econ. 1 (2004); Richard Schmalensee, Payment Systems and Interchange Fees, 50 J. Indus. Econ. 103 (2002).

[8] Morris, Zywicki & Manne, supra note 1, at 3–5.

[9] Julian Morris, The Hidden Wealth of Payment Cards: How Innovations in Payments Transform Society, Truth on the Mkt. (Dec. 19, 2024), https://truthonthemarket.com/2024/12/19/the-hidden-wealth-of-payment-cards-how-innovations-in-payments-transform-society.

[10] EMVCo, Overview of EMVCo, https://www.emvco.com/about-us/overview-of-emvco (last visited Apr. 20, 2026).

[11] EMVCo, EMV® 3-D Secure, https://www.emvco.com/emv-technologies/3-d-secure (last visited Apr. 20, 2026).

[12] Mastercard, Dynamic Currency Conversion (DCC) Performance Guide—Merchant Edition (Nov. 11, 2024), https://www.mastercard.com/content/dam/mccom/shared/business/support/rules-pdfs/DCC-Guide-2025-Merchant-Version.pdf.

[13] Markus Reisinger & Hans Zenger, Interchange Fee Regulation and Service Investments, 66 Int’l J. Indus. Org. 40 (2019), https://www.sciencedirect.com/science/article/abs/pii/S0167718719300311.

[14] Morris, Zywicki & Manne, supra note 1.

[15] Reserve Bank of Austl., Payments System Board Annual Report 2016, at 19 (2016); Reserve Bank of Austl., Payments System Board Annual Report 2020, at 17 (2020). See also Julian Morris, Todd J. Zywicki & Geoffrey A. Manne, The Effects of Price Controls on Payment-Card Interchange Fees: A Review and Update 30–31, Int’l Ctr. for L. & Econ. (2022); Julian Morris, Geoffrey A. Manne, Ian Lee & Todd J. Zywicki, Punishing Rewards: How Clamping Down on Credit Card Interchange Fees Can Hurt the Middle Class 22–23, Macdonald-Laurier Inst. (Nov. 2017).

[16] Case C-304/16, Am. Express Co. v. Lords Comm’rs of Her Majesty’s Treasury, ECLI:EU:C:2018:66, ¶¶ 53–61 (Feb. 7, 2018); Press Release No. 12/18, Court of Justice of the Eur. Union, A Three Party Card Scheme Involving a Co-Branding Partner or an Agent Is Subject to the Same Restrictions as Those Applicable to Four Party Schemes with Respect to Interchange Fees (Feb. 7, 2018).

[17] Bd. of Governors of the Fed. Reserve Sys., The Federal Reserve Payments Study: 2024 Update (Nov. 2024), https://www.federalreserve.gov/paymentsystems/2024-November-The-Federal-Reserve-Payments-Study.htm.

[18] Visa Inc., Annual Report for Fiscal Year 2025 (2025); Mastercard Inc., Supplemental Operational Performance 2023Q4–2025Q4 (2026).

[19] The 2024 figure is anchored to Datos Insights’ estimate that global outbound cross-border card volume reached $4 trillion in 2024, up 21% year over year. Earlier figures are back-cast using the average of Visa’s constant-dollar cross-border-volume growth and Mastercard’s local-currency cross-border-volume growth as an industry proxy. The network universe follows Nilson’s global brand-card universe: Visa, UnionPay, Mastercard, American Express, JCB, and Discover/Diners Club. This is not an audited all-network time series.

[20] European Central Bank, Report on Card Fraud in 2020 and 2021 6 (2023), https://www.ecb.europa.eu/pub/pdf/cardfraud/ecb.cardfraudreport202305~5d832d6515.en.pdf (last visited Apr. 9, 2026) [hereinafter ECB Card Fraud Report].

[21] European Banking Authority, 2025 Report on Payment Fraud, EBA-REP-2025-40, at 11 (2025) (while the report does not provide precise figures, the bar chart in Chart 1a suggests the trend continued).

[22] AusPayNet, Payment Fraud Statistics: 1 January 2024–31 December 2024 (2025), https://www.auspaynet.com.au/resources/fraud-statistics.

[23] ECB Card Fraud Report, supra note 20, at 15.

[24] See, e.g., Worldpay, How To Accept Payments Internationally, https://www.worldpay.com/en/insights/articles/how-to-accept-payments-internationally (last visited Apr. 20, 2026).

[25] Checkout.com & Oxford Economics, High-Performance Payments: The Hidden Billion-Dollar Opportunity 6, 10 (2023) (finding that merchants lose 1–2.1% of revenue to false declines, with cross-border complexity serving as a key driver).

[26] U.N. World Tourism Org., World Tourism Barometer, Vol. 23, No. 1 (2025).

[27] Discover Global Network, 3 Travel Payment Trends Shaping the Way Travelers Spend in 2025 (2025) (reporting that most consumers planned to use credit cards (68%) and debit cards (52%), rather than cash, for personal travel expenses, and that 46% of travelers sometimes, often, or always abandon purchases when merchants do not accept their preferred credit card); Tim Zawacki, Travel Perks Continue to Drive Premium Credit Card Uptake, Survey Finds, S&P Glob. Mkt. Intel. (Apr. 12, 2022) (reporting survey evidence that 52% of annual-fee credit-card holders identified travel-related perks as the cards’ most important feature); American Express, 2025 Global Travel Trends Report 4 (2025) (reporting that 66% of global respondents said combining credit-card rewards with other loyalty benefits provides “the best value for international trips,” and that 61% of surveyed Millennials and Gen Z use credit cards to maximize travel rewards).

[28] See Visa Inc., How Travel Trends Are Reshaping Global Payments (2023), https://corporate.visa.com/en/sites/visa-perspectives/trends-insights/how-travel-trends-are-reshaping-global-payments.html (noting that payment cards are the most commonly used payment method for international travel); Mastercard, Inside the Wealthy’s Playbook: How the Affluent Are Mastering Their Money (2024), https://www.mastercard.com/news/ap/en-hk/newsroom/press-releases/en-hk/2024/inside-the-wealthy-s-playbook-how-the-affluent-are-mastering-their-money-with-financial-gymnastics (observing that affluent consumers disproportionately use credit cards for high-value and overseas purchases); American Express, Global Travel Trends Report (2023), https://www.americanexpress.com/en-au/travel/discover/get-inspired/global-travel-trends-en-au (reporting that consumers value credit-card rewards and benefits in international travel); see also JPMorgan Chase & Co., Chase Sapphire Reserve® Card Benefits, https://creditcards.chase.com/rewards-credit-cards/sapphire/reserve; American Express, Platinum Card® Benefits, https://www.americanexpress.com/us/credit-cards/card/platinum (describing travel-related benefits, including no foreign-transaction fees, travel insurance, and purchase protection).

[29] Arielle Feger, Worldwide Ecommerce Sales to Break $6 Trillion, Make Up a Fifth of Total Retail Sales, eMarketer (Mar. 19, 2024), https://www.emarketer.com/content/worldwide-ecommerce-sales-break-6-trillion.

[30] U.N. Conf. on Trade & Dev. (UNCTAD), Estimates of Global E-Commerce 2019 and Preliminary Assessment of COVID-19 Impact on Online Retail 2020, Technical Note on ICT for Development No. 18 (2021), https://unctad.org/system/files/official-document/tn_unctad_ict4d18_en.pdf (estimating that cross-border transactions accounted for 25% of retail e-commerce).

[31] ECB Card Fraud Report, supra note 20, at 6.

[32] Juniper Research, Press Release, Fraudulent E-Commerce Transactions to Surpass $131bn by 2029 (Feb. 3, 2025).

Executive Summary

California’s long-term-care (LTC) financing challenge is no longer prospective—it is immediate. Rapid population aging, unusually high care costs, and a Medi-Cal program already under strain are converging to increase demand for long-term services and supports (LTSS). In this environment, a functioning private LTC insurance market—including both standalone policies and hybrid or linked-benefit life/LTC products—is not optional. It is a necessary complement to public financing, helping households manage risk, preserve assets, and reduce long-term pressure on state budgets.

Section I shows that California faces a widening financing gap. LTC needs are common and often prolonged, costs in California exceed national averages, and Medi-Cal operates primarily as a safety net rather than a comprehensive solution for middle-income households. At the same time, the state’s primary public-private bridge—the Partnership for Long-Term Care—is no longer functioning as intended, and overall private coverage has declined.

Section II examines how California administers LTC insurance regulation. The state relies on a fragmented prior-approval system that combines extensive substantive requirements with divided review authority and limited coordination. Compared to national trends toward standardized and centralized review, California’s approach produces greater variability, longer timelines, and less predictable outcomes. Evidence from national data and carrier surveys suggests that these differences reflect institutional design, not simply stronger consumer protection.

Section III analyzes the operational consequences of that system. Review timelines in California routinely extend into years rather than months, filing fees are among the highest in the country, and overlapping review authority—especially for hybrid products—introduces delay and inconsistency. These frictions raise the cost of participation, discourage product entry, and create a sequencing problem in which products can become outdated before they reach the market.

Section IV shows how these delays translate into consumer harm. LTC insurance depends on timing: consumers typically purchase coverage during a midlife window when premiums are lower and underwriting is more permissive. Prolonged review narrows that window, reduces product choice, and limits access to newer product designs that have gained traction nationally. As a result, households are more likely to rely on savings drawdowns or unpaid caregiving. Over time, these dynamics shift costs onto families and increase the likelihood of eventual reliance on Medi-Cal. Regulatory delay is therefore not just an administrative issue—it is a consumer-welfare and fiscal-policy problem.

Using a law & economics framework, the brief shows that California’s outcomes diverge from those in other states with similarly strong consumer-protection regimes. The evidence is more consistent with institutional friction—rather than substantive regulatory stringency alone—as the primary driver of market weakness.

Section V outlines a set of practical reforms to restore market function without weakening consumer protection. These include establishing timeline accountability and escalation protocols, consolidating review authority for hybrid products, rationalizing filing fees, creating a pathway for limited-duration LTC coverage, and evaluating participation in interstate review mechanisms. Taken together, these reforms would reduce avoidable delay, improve regulatory predictability, and expand the range of products available to California consumers.

For a state confronting rapid demographic change and rising care costs, improving the administration of LTC insurance regulation is not a marginal adjustment. It is a necessary step toward a more sustainable long-term-care financing system.

I.               California’s Widening Long-Term Care Financing Gap

California is entering a new demographic era—one that will place sustained pressure on how the state finances long-term care. By 2040, the share of Californians age 65 and older will rise from 14% in 2020 to 22%. Over the same period, the old-age dependency ratio will increase from 24 to 38 older adults per 100 working-age adults. This shift will drive demand for long-term services and supports (LTSS), even as the working-age population that finances much of the system grows only modestly.[1]

FIGURE 1: California’s Aging Pressure, 2020–2040

SOURCE: PPIC

*Old-age dependency ratio is the number of adults 65+ per 100 working adults.

LTSS needs are common, not exceptional. Federal estimates indicate that a person turning 65 today has nearly a 70% chance of needing some form of LTSS, and 20% will require care for more than five years. Microsimulation estimates reinforce this point: 48% of older adults will receive paid LTSS over their lifetimes, and 24% will receive paid care for more than two years. Long-term-care risk is therefore not a tail event affecting a small minority of households. It is a predictable feature of aging.[2]

California’s cost structure makes that risk especially acute. AHIP’s 2025 state report, using 2024 data, estimates annual costs of about $86,944 for homemaker services, $89,232 for home health aide services, $88,200 for assisted living, $140,343 for a semi-private nursing-home room, and $182,135 for a private room. Each figure exceeds the national benchmark. Even relatively short care episodes can impose significant financial strain, particularly on middle-income households without substantial liquid assets.[3]

FIGURE 2: California v. United States Annual Care Costs by Setting

SOURCE: AHIP

Public programs cannot absorb this growing burden on their own. In 2025–26, the budget for Medi-Cal—California’s Medicaid program—is estimated at $197 billion, making it the largest program in the state budget on a total-funds basis. LTSS obligations within Medi-Cal are already expanding. From 2017 to 2022, the number of Medi-Cal enrollees using LTSS increased by 20%, and more than six times as many enrollees relied on home- and community-based services (HCBS) as experienced long-term-care stays in 2022. These trends show that LTSS financing is already a central fiscal issue—not a distant concern.[4]

Medi-Cal operates as a safety net, not a comprehensive financing solution for middle-income households. The California Department of Health Care Services (DHCS) applies asset limits to people age 65 or older, people with disabilities, nursing-home residents, and certain other non-MAGI enrollees. The current limit is $130,000 for an individual, plus $65,000 for each additional family member. For nursing-home LTSS, DHCS reviews asset transfers made during the prior 30 months, and transfers made on or after Jan. 1, 2026, may trigger penalties that delay eligibility.[5]

These rules have meaningful real-world effects. A 2025 JAMA Network Open study finds that 16.4% of nursing-home residents who initially lacked Medicaid coverage spent down assets and enrolled between 2018 and 2022.[6]

FIGURE 3: Medi-Cal LTSS Growth and Service Mix, 2017–2022

SOURCE: California Health Care Foundation. LTSS totals include overlap because some enrollees use both long-term care stays and home and community-based services (HCBS).

This framework creates a persistent financing gap. Households with assets to protect—but not enough wealth to self-insure—face the greatest exposure. The American Academy of Actuaries identifies long-term-care insurance purchasers as often falling into this category: individuals who would not qualify for Medicaid without reducing assets but who also cannot absorb extended care costs out of pocket.[7]

California’s Partnership for Long-Term Care reflects this logic. DHCS explains that Partnership policies offer Medi-Cal asset protection, allowing policyholders to shield one dollar of assets for every dollar paid in benefits. In principle, this public-private model allows households to avoid unnecessary impoverishment while reducing future pressure on Medi-Cal.[8]

In practice, the model is not functioning. DHCS’s current Partnership website states that no Partnership-approved insurers are selling policies.[9] This breakdown likely reflects both broader actuarial pressures and a structural mismatch between the Partnership program and evolving product design.

The Partnership framework centers on standalone long-term-care policies with specified benefit structures and consumer protections.[10] Newer asset-based and hybrid products—typically life insurance or annuities with long-term-care riders—do not fit this structure. Hybrid life insurance policies with LTC riders generally do not qualify for the California Partnership program,[11] and state law requires disclosures stating that they are not Partnership policies.[12] As hybrid and linked-benefit products gain traction nationally, California’s flagship public-private program remains tied to a narrower legacy model.

This disconnect both reflects and contributes to a broader contraction in private coverage. AHIP data show that total private LTC coverage in California declined from 750,963 covered lives in 2021 to 735,901 in 2024. Standalone coverage fell from 605,488 to 572,215 over the same period, and growth in extended-benefit and acceleration products did not offset the decline. California has therefore experienced both contraction in traditional LTC insurance and insufficient expansion in alternative products.[13]

This deterioration does not mean private financing is unavailable. Nationally, the standalone LTC insurance market has faced sustained pressure. Milliman reports that policy terminations have exceeded new policy issuances by roughly 127,000 individuals per year over the past decade. The market, however, has evolved. The American Academy of Actuaries reports that hybrid products now play a significant role, and short-term-care products offering up to 12 months of coverage appear to be growing.[14]

FIGURE 4: California Private LTC Coverage by Product Type, 2021 v. 2024

SOURCE: AHIP

For California, the key policy question is not whether national headwinds exist—they do. The question is whether the state allows sufficient flexibility for product redesign and private financing options to reach households exposed to catastrophic care costs.

California is already evaluating broader LTSS financing options. DHCS notes that the state commissioned feasibility and actuarial work for a public LTC option under the Budget Act of 2019, while a California Department of Insurance (CDI) task force advanced a separate feasibility report under Assembly Bill 567.[15] Regardless of the path chosen for public financing, California will still require a functioning private market—particularly for households unlikely to qualify for Medi-Cal until after significant asset depletion.

A durable strategy should treat private financing not as a substitute for public programs, but as a necessary complement.

Taken together, these trends point to a widening financing gap. California faces rapid population aging, high care costs, and a Medi-Cal program that is already fiscally significant. At the same time, the private market that could help households pool risk and protect savings has weakened. Later sections examine whether state-specific regulatory frictions have compounded national market pressures. Absent a more balanced financing strategy, more middle-income households will face delayed planning, asset depletion, and eventual reliance on Medi-Cal.[16]

II.             Administrative Frictions in California’s LTC Insurance System

California’s LTC insurance market is best understood by separating substantive consumer-protection rules from the administrative machinery that implements them. The state regulates LTC insurance through parallel prior-approval tracks. The California Department of Insurance (CDI) reviews policy forms and related materials before products reach the market, while the Insurance Code separately requires prior approval of both initial premium schedules and subsequent premium increases. This framework also applies to LTC benefits embedded in life insurance policies or annuity contracts.[17]

FIGURE 5: California LTC Insurance Prior-Approval Structure

CDI guidance highlights the administrative demands of this system. LTC filings typically require detailed review of policy labels, outlines of coverage, benefit triggers, inflation protection, nonforfeiture provisions, replacement notices, required coverages, prohibited terms, and other disclosure and suitability protections. Given that these products are often purchased by older consumers and designed to remain in force for decades, that scrutiny is justified. It carries a tradeoff, though. Outcomes depend not only on legal standards, but also on timely review, clear objections, and predictable resolution. When those elements break down, even well-designed rules can produce inefficient results.[18]

Institutional design amplifies these administrative pressures. Within CDI, review authority is divided by product type. Health-insurance units generally review standalone LTC policies, while hybrid life/LTC products often require review by both life and health divisions. In practice, this structure tends to produce sequential—not parallel—review, repeated requests for clarification, and inconsistent interpretations across bureaus. Each iteration can reset or extend approval timelines, increasing uncertainty for insurers and delaying product availability.

California’s approach is not unique in imposing substantive regulation, but it stands apart in how it administers that regulation. National practice increasingly relies on shared standards to reduce duplication and improve consistency. NAIC Model Regulation 641 establishes common rules on rates, nonforfeiture, suitability, and marketing. The National Association of Insurance Commissioners’ 2022 Long-Term Care Insurance Multistate Rate Review Framework seeks to streamline and harmonize multistate rate review.[19] The Interstate Insurance Product Regulation Compact—now covering 48 jurisdictions, 40 of which have adopted LTC standards—provides centralized review of individual LTC product standards, including core policy terms and modified rate schedules. California does not participate in the compact.[20]

Comparative evidence suggests that regulatory outcomes vary for reasons beyond underlying cost differences or a uniform commitment to consumer protection. The 2021 NAIC LTC data call, based on submissions from 19 insurers across all 50 states, reported a nationwide average cumulative approved rate increase of 112%, alongside substantial cross-state variation. The same analysis found that nursing-home costs did not appear to be a primary driver of state-level LTC experience. Similarly situated policyholders can therefore face materially different outcomes depending on jurisdiction.[21]

Recent carrier surveys reinforce this pattern. The 2025 Milliman survey, covering 17 companies representing more than 75% of LTC premium volume, reports an average approval time of six months from submission to disposition for 2024 filings, down from seven months in the 2021 survey. In both years, carriers identified California as one of the most resource-intensive jurisdictions, alongside Florida, New Jersey, New York, and Texas. The same survey finds that the most common reason for reduced approvals or disapprovals is a political cap or other nonactuarial factors. Administrative practices continue to diverge across states, even within a national market.[22]

California’s institutional structure introduces another source of variability. The insurance commissioner is elected.[23] Jessica Liu and Weiling Liu, analyzing identical insurer requests submitted simultaneously across states, find that approval rates increase by 13% and approved amounts rise by 4% in the period following an election year. As elections approach, commissioners approve fewer requests, approve them less frequently, and grant smaller increases.[24]

Figure 6 illustrates this dynamic through New York Life Insurance Co.’s experience. After seeking roughly 20% premium increases in 2013 for LTC policies sold nationwide, the company received divergent outcomes: a 30% increase in New Hampshire, but outright denials in California, Connecticut, Rhode Island, and Vermont.[25]

FIGURE 6: Average Rate Increase Granted to NY Life Insurance

SOURCE: Liu & Liu

Because California vests rate-review authority in an elected office, these findings are directly relevant—even if they do not isolate a California-specific effect. They underscore how political incentives can shape regulatory outcomes.

CDI itself has acknowledged concerns about administrative capacity and review timelines. In Bulletin 2024-7, Commissioner Ricardo Lara states that the department’s rate-application review process exceeded statutory timelines for multiple reasons and announces reforms to improve speed, transparency, and certainty. The bulletin addresses property-casualty insurance rather than LTC specifically, but it reflects a broader issue: prior-approval systems can become slower and less predictable than intended.[26]

Taken together, the evidence points to institutional friction—not simply “strong regulation”—as the key explanatory factor. California combines extensive substantive requirements, an elected regulator, and nonparticipation in a major interstate review mechanism, all within a national market that increasingly relies on shared standards and coordinated processes. These features do not make current outcomes inevitable. They make them contingent—and therefore open to reform.[27]

III.           Delay, Cost, and Product Obsolescence in California’s LTC Market

Section I documented contraction in California’s private long-term-care (LTC) insurance market. This section examines a plausible institutional explanation: the cost and duration of product review. The claim is not that every carrier exit or product non-entry stems from a single regulatory decision. It is that California combines long review periods, high filing fees, and a sequencing problem that can leave products outdated before they reach the market.[28]

In a line of insurance where firms can reallocate capital, product-development staff, and distribution across jurisdictions, these frictions carry real economic consequences.

A.   Review Timelines

Public benchmarks suggest that California’s timing problem does not simply reflect the complexity of LTC insurance. New York’s life-insurance SERFF guidance illustrates a different model. Under its “deemer” procedure, the department generally has 90 days to act, with additional 45-day response periods if objections arise. These benchmarks are not perfectly comparable to California LTC form review, but they show that multi-year review is not inherent to the product.[29]

Survey evidence reinforces this point. A 2024 member survey by the Association of California Life and Health Insurance Companies (ACLHIC) finds that average review times for LTC products in California approach or exceed 600 days, making the state a clear outlier.[30] Comparable filings in other states typically receive approval in fewer than 85 days.[31] California’s timelines are roughly seven times longer than the national average and, in some cases, exceed peer-state timelines by more than an order of magnitude.

FIGURE 7: California LTC SERFF Filings by Type, 2020–2025

SOURCE: Authors analysis of California SERFF LTC03I/03G filings.

We examined 368 product filings submitted in California between 2020 and 2025 using NAIC System for Electronic Rate and Form Filing (SERFF) data for LTC03I/03G filings (individual LTC insurance).[32] LTC03I accounts for 93%–95% of filings and provides a representative sample. Of the 368 filings, 255 were marketing-related and another 70 involved minor requests, such as form variations.[33]

The remaining 43 filings were substantive. Average time to disposition was 905 days for rate filings, 951 days for form filings, and 957 days for combined rate/form filings. Fourteen of the 43 filings—nearly one-third—took more than three years to reach final disposition.[34]

The distribution of review times underscores the severity of the delay. Multi-month review is common in complex insurance lines. California routinely subjects LTC filings to timelines measured in years. Some filings remain pending for more than four years.[35] These delays persist even when products are not novel, have already been approved in other states, and rely on familiar actuarial assumptions and policy structures.

FIGURE 8: Avg. Days to Disposition for Substantive California LTC Filings

SOURCES: Authors analysis of California SERFF LTC03I/03G filings; Milliman.

Comparison with other states highlights the gap. Connecticut—identified in the 2024 Society of Actuaries survey as a high-effort jurisdiction[36]—received 69 LTC rate-increase requests between 2020 and 2022. It approved 16 as submitted and 42 at reduced levels, rejecting only six.[37]

These differences are too large to attribute solely to case mix. A process measured in roughly 2.5 years rather than six months reflects a qualitatively different administrative environment. The relevant question is not whether delay is justified in some cases. It is whether California’s system has shifted from scrutiny to bottleneck.

Approval data from 2020 to 2022 illustrate that bottleneck. Between Jan. 1, 2020, and Feb. 21, 2022, CDI reportedly approved only two new LTC policy forms that included application materials.[38] One filing remained under review for 1,741 days—nearly five years—before approval.[39] For most insurers, a timeline of that length is functionally equivalent to a denial, given the opportunity cost of capital and the pace of product development elsewhere.

B.    Filing Fees

California also imposes some of the highest filing fees in the United States for LTC products. Depending on the submission, fees for policy forms, riders, and related materials range from roughly $5,000 to $50,000 per filing.[40] Insurers incur these costs regardless of whether a filing receives timely approval, undergoes repeated review cycles, or remains pending.

Under Title 10, Section 2202 of the California Code of Regulations, LTC filings are assessed by document category. Posted fees include $4,960 for a policy, $2,480 for a certificate, $1,110 for a rider or insert page, $1,540 for an application, $1,110 for new-issue rates, and $2,600 for rate changes. Separate charges apply to advertising and certain compliance documents. Life-insurance or annuity products with LTC benefits—such as accelerated death benefits or enhanced annuity features—face similarly substantial document-specific fees.[41]

These charges make even routine submissions expensive before regulators reach the merits. A standalone LTC filing with one policy, one application, and three riders generates $9,830 in posted California fees, excluding any separate rate filing. Texas, by comparison, charges $100 for a filing with one policy, three riders, and an application, and $100 for an LTC rate filing. The comparison is not exhaustive, but the direction is clear: California converts product review into a significantly larger fixed-cost commitment than at least some peer states.[42]

Because these costs are fixed rather than usage-based, they matter most at the margin—where firms decide whether to test a niche product, update an existing form, or allocate development resources to a jurisdiction. Larger insurers may absorb these costs more easily than smaller or specialized firms. Even for large carriers, though, high fees combined with multi-year review periods materially reduce the expected return on filing in California.

C.   Pending Filings and Product Obsolescence

Delay also creates dynamic costs. California’s Checklist for Long-Term Care Insurance Policies, revised in May 2025, applies to both standalone LTC policies and LTC benefits embedded in life-insurance or annuity products. When review stretches into years, routine updates to disclosure requirements, form standards, or supporting materials can force insurers to revise pending filings before launch.[43]

The result is product lag. When a review cycle outlasts a product-development cycle, regulators are no longer evaluating the market frontier. They are evaluating an earlier version of it. Approval may arrive only after a carrier has redesigned the product, updated materials, or shifted distribution elsewhere. At that point, approval has limited commercial value. It authorizes a stale filing rather than enabling timely market entry. This dynamic helps explain why prolonged delay can deter participation even without formal denial.

FIGURE 9: How Delay Can Turn Review into Product Lag

The evidence in this section points to a consistent conclusion. California’s LTC regime is not problematic simply because it is protective. It is problematic because, in practice, it is slower and more costly than necessary. These administrative frictions deter entry, discourage product redesign, and reduce sustained participation. This does not prove that regulation alone drives market contraction. It does show that California has made participation materially more burdensome than in faster-moving jurisdictions.[44]

IV.          The Consumer Costs of Regulatory Delay

Section III shows that regulatory delay imposes real and compounding costs on California consumers. Those costs extend beyond administrative timelines. They shape household finances and public budgets, as individuals make long-term decisions with fewer—and often inferior—options.[45]

Long-term-care (LTC) insurance is not a commodity purchase that households can defer indefinitely. Timing is part of the product. Consumers typically purchase coverage while they remain insurable, before premiums rise materially, and before product design evolves again. When a state slows product entry and redesign, the result is not simply fewer offerings. Delay narrows options during the key purchase window, limits access to newer products, and increases the likelihood that households will rely on savings drawdowns or unpaid caregiving.[46]

A.   Reduced Product Choice

California consumers face a narrower—and increasingly outdated—set of LTC insurance options. Traditional standalone policies are scarce, and hybrid or linked-benefit products remain limited, even as they dominate new sales elsewhere. Some of this reflects national retrenchment: the number of companies writing new LTC business fell from more than 100 in 2000 to fewer than a quarter of that total.[47] California nonetheless stands out. As of 2022, much-smaller Virginia reported 13 companies writing new LTC business,[48] while California had just eight.[49]

Limited choice weakens competition, reduces pressure on premiums, and discourages product differentiation.[50] These effects are especially costly in a heterogeneous market, where consumers differ widely in income, health status, and care preferences.

California has also lagged in the transition to newer product forms. AHIP data show that total private LTC coverage in California fell from 750,963 covered lives in 2021 to 735,901 in 2024. Over the same period, combined extended-benefit and acceleration coverage rose from 145,475 to 163,686—an increase of about 12.5%. Nationally, those categories grew from 799,313 to 1,121,800, an increase of roughly 40.4%. The national market evolved. California adopted those changes more slowly.[51]

Faster review of new forms and new-issue rates would reduce the risk that products reach market with stale pricing assumptions. That concern is particularly acute for hybrid or asset-based LTC products. These products combine life-insurance or annuity value with LTC benefits, often preserve a death benefit if care is never needed, and are frequently sold on a single-premium basis with greater price certainty than traditional coverage.[52] Their appeal depends on locking in value and pricing at issue, rather than returning later for class-wide increases.

FIGURE 10: California Lags Shift Toward New LTC Product Forms

SOURCE: AHIP

Multi-year review delays do more than postpone access. They erode the usefulness of the approved pricing itself. Even for those guaranteed-renewable products that can later seek rate adjustments, a multi-year process risks approving rates that no longer reflect current market conditions.

That lag matters because newer product forms are no longer marginal. The American Academy of Actuaries reports that hybrid products now play a significant role in the LTC market, and the American Association for Long-Term Care Insurance (AALTCI) reports that linked-benefit policies now outsell traditional standalone coverage. For California consumers, delay means not just fewer policies, but reduced access to the forms of coverage gaining traction elsewhere.[53]

Delay is especially costly because the purchase window is concentrated in midlife. AALTCI reports that 55% of traditional LTC buyers are between ages 55 and 65, while only 18% are age 66 or older. Another AALTCI summary finds that more than 76% of new buyers fall between ages 50 and 69 and identifies ages 55 to 65 as the “sweet spot,” when underwriting is easier and premiums are more affordable.[54]

FIGURE 11: The Premium Cost of Waiting to Buy LTC Insurance

SOURCE: AALTCI. Rates shown are the organization’s Illinois benchmark and are not California-specific

Missing that window carries financial and medical consequences. AALTCI’s 2025 price index shows that an illustrative couple purchasing traditional coverage with a $165,000 initial benefit pool for each insured and 3% annual growth would pay about $5,050 at age 55, $5,800 at age 60, and $7,150 at age 65, based on Illinois benchmark rates. Underwriting also tightens with age. About 38.2% of applicants ages 65 to 69 are declined, and 47% of applicants ages 70 to 75 are either declined or deferred. Delay does not simply postpone purchase—it pushes consumers into a market that is more expensive and harder to enter.[55]

B.    Product Modernization and Service Innovation

Regulatory delay also limits access to product modernization. The American Academy of Actuaries’ 2025 issue brief reports that hybrid products combining life or annuity benefits with LTC coverage now play a significant role in the market. It also notes growth in short-term-care products offering up to 12 months of coverage and increased insurer investment in data analytics and policyholder engagement for claims and pre-claim management. The NAIC likewise reports efforts to facilitate innovative product offerings.[56]

Modern LTC products increasingly compete on features that matter directly to consumers: financing structure, benefit flexibility, access to home- and community-based care, and claims-support models. NAIC guidance notes that LTC policies may cover home health care, respite care, hospice care, personal care in the home, assisted living, and adult day care. The relevant question is no longer whether policies include noninstitutional benefits. It is whether consumers can access newer combinations of benefits, funding structures, and support features while those products remain current.[57]

When review cycles stretch into years, California risks approving products that no longer reflect current market design. That harms consumers even when approval issues, because the approved form may no longer align with insurer strategy or with more responsive products available elsewhere. Delay thus converts regulatory caution into product obsolescence.

FIGURE 12: How Regulatory Delay Becomes Consumer Harm

These harms extend beyond individual applicants. When households cannot secure timely private coverage, costs shift to spouses, adult children, and other unpaid caregivers. AARP’s 2021 caregiver-cost study finds that family caregivers spend an average of $7,242 annually out of pocket. AARP’s 2026 caregiving valuation report estimates that family caregivers of adults provide 49.5 billion hours of care annually, valued at about $1.01 trillion. Regulatory delay is therefore not merely an administrative issue. It is a household-welfare problem that shifts cost and risk from the insurance market back onto families.[58]

V.             Practical Reforms for California’s LTC Insurance Market

The preceding sections point to a clear conclusion: California needs a practical reform agenda. The state’s LTC insurance problem is not excessive consumer protection. It is that administrative friction, fragmented review paths, and high filing costs have become structural barriers to product entry and modernization.

Reform should focus on how review is organized, managed, and measured. If California can reduce avoidable delay without weakening substantive protections, it can expand product availability while preserving the consumer-protection commitments that justify prior approval.

A.   Establish Timeline Accountability and Escalation

California lacks a management framework for aging filings. CDI does not need rigid statutory deadlines, which could encourage superficial review. It does need internal benchmarks that distinguish routine processing from abnormal delay.

Graduated targets—such as 90 days for routine filings, 120–180 days for complex or novel products, and mandatory escalation beyond those ranges—would function as management tools rather than legal mandates. A workable system would define standard review windows and require supervisory escalation when filings exceed them.

FIGURE 13: Proposed Timeline Accountability and Escalation Framework

In August 2024, CDI acknowledged the need to improve speed, transparency, and certainty in other insurance lines.[59] The same logic applies here. When filings exceed internal benchmarks, responsibility should shift from individual reviewers to supervisory or cross-functional teams empowered to resolve disputes, consolidate requests, and narrow the scope of review. Without escalation, filings stall due to staff turnover, shifting priorities, or risk aversion.

Escalation matters because it changes how organizations handle delay. A filing pending for hundreds of days should not remain with the original examiner without intervention. It should move to a higher level where managers can consolidate issues, narrow disagreements, and determine whether continued review improves consumer protection or simply prolongs uncertainty.

California should also report aggregate timing data for LTC filings, including average and median days to disposition, counts of pending filings by age bucket, and the share exceeding defined thresholds. Transparency changes incentives. It turns delay into a measurable outcome rather than a default condition.

B.    Consolidate Review Authority for Hybrid Products

Hybrid life/LTC and annuity/LTC products present a structural challenge. Because they combine long-duration life or annuity features with LTC benefits, review often fragments across units with different expertise and interpretive approaches. California’s checklist confirms that LTC rules apply to benefits embedded in life insurance or annuities. That breadth is defensible. Diffuse ownership is not.[60]

FIGURE 14: Consolidating Review Authority for Hybrid Life/LTC Products

California should designate a lead bureau for hybrid products, supported by targeted consultation from LTC specialists. The goal is not to reduce expertise, but to eliminate predictable inefficiencies when no single unit owns a filing from intake through disposition. A lead-bureau model would improve accountability, reduce handoffs, and make it easier to identify where and why a filing has stalled.

Consolidating authority within the CDI Life Bureau—consistent with practice in many states—would reduce these inefficiencies while preserving substantive oversight. The Life Bureau already has expertise in long-duration contracts, guaranteed benefits, actuarial pricing, and solvency analysis. Existing LTC protections can remain fully intact within a unified structure.

C.   Rationalize Filing Fees

Filing fees are a core barrier to market access. California’s PRAB I fee schedule imposes substantial document-specific charges. A filing with one policy, one application, and three riders generates $9,830 in fees before any rate filing; a standalone rate filing adds another $2,600. Texas, by comparison, charges $100 for a comparable submission and $100 for an LTC rate filing.[61]

These differences matter because filing fees are fixed costs. Insurers must incur them before knowing whether a filing will move quickly, undergo multiple review cycles, or remain pending. California need not eliminate fees. It should align them more closely with actual review effort.

A more rational structure would distinguish among routine amendments, standardized forms, novel products, and actuarially intensive filings. In a system already burdened by delay, lower and more proportional fees would make incremental product improvement materially easier.

D.   Consider Adopting the Limited LTC Insurance Model Act

California should consider adopting the NAIC’s 2018 Limited Long-Term Care Insurance Model Act and Model Regulation. The Model Act defines limited LTC insurance as coverage providing less than 12 consecutive months of benefits and aims to promote availability, flexibility, and consumer understanding alongside traditional protections.[62]

The Model Regulation includes disclosure requirements, standards for home- and community-based benefits where offered, inflation protection, application and replacement rules, marketing standards, and suitability requirements. It offers a tailored, consumer-protective framework for shorter-duration products.[63]

The strategic value is straightforward. The current market is thin, comprehensive coverage remains expensive, and many households will not purchase traditional policies even when available. A limited-LTC framework would create a pathway for lower-cost, shorter-duration products that expand awareness and provide an entry point for consumers. It would not solve the financing gap, but it would expand supply and engagement.

E.    Evaluate Joining the Interstate Compact

California could further reduce friction by joining the Interstate Insurance Product Regulation Compact (IIPRC). The Compact is an operational system with 48 member jurisdictions and a centralized filing process for life insurance, annuities, disability income, and LTC insurance.[64]

It already includes a developed LTC framework, adopted by 40 compact members, that covers core policy terms, application requirements, benefit features, advertising, and multiple rate-filing approaches. Participation would reduce duplicative state-by-state review while requiring California to operate within standardized product-form rules.[65]

FIGURE 15: IIPRC Average Business Days to Approval (2021-2025)

SOURCE: IIPRC

Historically, the Compact has delivered faster review. It reports average review times of 25 days in 2018, 33 days in 2019, 21 days in 2020, and 27 days in both 2021 and 2022—well below its 60-day rule. Turnaround times increased in 2025 due to higher volume and complexity, but the system remains faster and more standardized than California’s current approach.[66]

Compact participation does not require deregulation or surrender of state authority. California would retain control over rates, market conduct, and enforcement. The Compact governs product-form review, not pricing or substantive consumer-protection policy. It offers administrative efficiency without sacrificing state authority.

From a fiscal perspective, improved access to private LTC insurance supports earlier planning and risk pooling, reducing long-term pressure on Medi-Cal. As demographic aging accelerates, the Compact offers a practical tool to stabilize the market while protecting consumers and public finances.

F.    Strengthen Legislative Oversight and Pilot Authority

The Legislature should make these reforms measurable. Reporting requirements for LTC filings would make delay visible and create an external check on administrative drift. Informational hearings focused on filing age, disposition patterns, staffing, and backlog reduction would shift the default from passive tolerance to active oversight.

FIGURE 16: Reform Pathways and Expected Effects

The Legislature should also authorize targeted pilots. California does not need to redesign its approval system all at once. It can test reforms incrementally—expedited review for standardized filings, a lead-bureau model for hybrid products, or a limited-LTC pathway based on the NAIC framework. Well-designed pilots would generate evidence on whether faster review and strong consumer protection can coexist.

Taken together, these reforms operate as a governance package rather than isolated fixes. Timeline accountability improves discipline. Consolidated review improves ownership. Fee reform improves incentives. Limited-LTC legislation and Compact participation expand product pathways. California does not need to weaken consumer protection to restore a functioning LTC insurance market. It needs to administer that protection in a way that is timely, coherent, and compatible with modern product development.

VI.          Conclusion

California’s long-term-care challenge is no longer prospective—it is present. The population is aging rapidly, care costs are unusually high, and Medi-Cal already finances a large and growing share of long-term services and supports. In this environment, a functioning private LTC insurance market is not optional. It is a necessary complement to public financing if California aims to reduce household asset depletion and limit future pressure on public budgets.

The preceding sections show that California’s weak private LTC market does not stem solely from national actuarial headwinds. Insurers have adapted elsewhere by redesigning products, expanding hybrid and linked-benefit offerings, and introducing shorter-duration coverage. California has taken a different path. It combines extensive substantive requirements with prolonged review periods, high filing costs, fragmented oversight of hybrid products, and limited administrative accountability. The result is predictable: slower product entry, narrower consumer choice, and delayed access to innovation.

These outcomes impose tangible costs. For consumers, delay compresses the window when coverage is most affordable and underwriting is least restrictive. For families, it shifts risk to unpaid caregiving and self-financing. For the state, it increases the likelihood that households will spend down into Medi-Cal rather than plan through private risk pooling. Regulatory delay is not merely administrative. It directly affects consumer welfare and fiscal outcomes.

California does not need to weaken consumer protection to improve these outcomes. It needs a regulatory system that implements those protections in a timely, disciplined, and transparent way. The reforms outlined in this paper—timeline accountability, clearer ownership of hybrid-product review, more proportional filing fees, a pathway for limited-LTC products, and serious consideration of interstate coordination—will not resolve the LTC financing gap on their own. They will, however, remove avoidable barriers that constrain supply and delay access.

For a state confronting rapid population aging and rising care costs, removing those barriers is not a marginal improvement. It is a necessary condition for a more sustainable long-term-care strategy.

[1] Hans Johnson et al., California’s Aging Population 1 (Pub. Pol’y Inst. of Cal. Jan. 2025), https://ppic.org/wp-content/uploads/policy-brief-californias-aging-population.pdf.

[2] See How Much Care Will You Need?, Admin. for Cmty. Living, U.S. Dep’t of Health & Hum. Servs., https://acl.gov/ltc/basic-needs/how-much-care-will-you-need (last visited Apr. 20, 2026); see also Off. of the Assistant Sec’y for Plan. & Evaluation, U.S. Dep’t of Health & Hum. Servs., What Is the Lifetime Risk of Needing and Receiving Long-Term Services and Supports? 1 (Apr. 3, 2019), https://aspe.hhs.gov/reports/what-lifetime-risk-needing-receiving-long-term-services-supports.

[3] AHIP, Long-Term Care Insurance Coverage: State-to-State 2025 3, 8 (Nov. 2025), https://ahiporg-production.s3.amazonaws.com/documents/202511-AHIP_LTCStateData.pdf.

[4] See Legis. Analyst’s Off., Considering Medi-Cal in the Midst of a Changing Fiscal and Policy Landscape 1 (Oct. 24, 2025), https://lao.ca.gov/Publications/Report/5083; see also Cal. Health Care Found., Long-Term Care in California: 2024 Edition 1, 4 (Dec. 2024), https://www.chcf.org/wp-content/uploads/2024/12/LongTermCareAlmanac2024.pdf.

[5] Asset Limit Frequently Asked Questions, Cal. Dep’t of Health Care Servs., https://www.dhcs.ca.gov/Medi-Cal/Pages/Help/asset-limits-faqs.aspx (last visited Apr. 20, 2026).

[6] Gabriella Aboulafia, Amanda C. Chen & David C. Grabowski, Asset Spend-Down and Medicaid Enrollment in Nursing Homes, 8 JAMA Network Open e2546876 (2025), https://jamanetwork.com/journals/jamanetworkopen/fullarticle/2842303.

[7] Am. Acad. of Actuaries, The State of Long-Term Care Insurance—2025 2 (Feb. 2025), https://www.actuary.org/wp-content/uploads/2025/03/health-brief-2025-stateofltc.pdf.

[8] See The California Partnership for Long-Term Care, Cal. Dep’t of Health Care Servs., https://www.dhcs.ca.gov/services/ltc/Pages/CPLTC.aspx (last visited Apr. 20, 2026); see also Partnership Policies, Cal. Dep’t of Health Care Servs., https://www.dhcs.ca.gov/individuals/rureadyca/Pages/Partnership-Policies.aspx (last visited Apr. 20, 2026); see also Cal. Dep’t of Health Care Servs., Before You Buy 3, 5, https://www.dhcs.ca.gov/individuals/rureadyca/Documents/Before-You-Buy.pdf (last visited Apr. 20, 2026).

[9] Partnership Certified Companies, Cal. Dep’t of Health Care Servs., https://www.dhcs.ca.gov/individuals/rureadyca/Pages/Partnership-Certified-Companies.aspx (last visited Apr. 23, 2026).

[10] See Partnership Policies, Cal. Dep’t of Health Care Servs., https://www.dhcs.ca.gov/individuals/rureadyca/Pages/Partnership-Policies.aspx (last visited Apr. 23, 2026); see also Cal. Code Regs. tit. 22, §§ 58050, 58059 (2026).

[11] Traditional Long-Term Care Insurance vs. Hybrid Policies, ACSIA Partners, https://acsiapartners.com/traditional-long-term-care-insurance-vs-hybrid-policies (last visited Apr. 23, 2026).

[12] See Cal. Dep’t of Ins., Checklist for Long-Term Care Insurance Policies 1 (rev. May 2025), https://www.insurance.ca.gov/0250-insurers/0300-insurers/0100-applications/pab/upload/long-term-care-checklist-updated.pdf; see also Cal. Ins. Code §§ 10295, 10295.11 (2026).

[13] See AHIP, Long-Term Care Insurance Coverage: State-to-State 2023 8 (Nov. 2023), https://ahiporg-production.s3.amazonaws.com/documents/AHIP_LTC_State_Data_Report.pdf; see also AHIP, supra note 3, at 8.

[14] See Sam Smetek, Annie Gunnlaugsson, Christopher Giese & Quentin Clemens, The Long-Term Care Insurance Industry Through 2024: Summary Statistics and Observations from the Experience Reporting Forms, Milliman (Dec. 31, 2025), https://www.milliman.com/en/insight/ltci-2024-statistics-experience-reporting-forms; see also Am. Acad. of Actuaries, supra note 7, at 1.

[15] Cal. Dep’t of Health Care Servs., Long-Term Services and Supports Feasibility Study, https://www.dhcs.ca.gov/provgovpart/Pages/LTSS-Feasibility-Study.aspx (last visited Apr. 20, 2026).

[16] See Johnson et al., supra note 1, at 1–2; see also AHIP, supra note 3, at 8; see also Legis. Analyst’s Off., supra note 4, at 1; see also Smetek et al., supra note 14.

[17] See Cal. Ins. Code §§ 10236.11, 10236.13 (2025); see also Cal. Dep’t of Ins., Policy Regulation & Approval Bureau I, https://www.insurance.ca.gov (last visited Apr. 20, 2026); see also Cal. Dep’t of Ins., Guideline for California Long-Term Care Insurance Policies 1 (June 27, 2017).

[18] See Cal. Dep’t of Ins., supra note 17, at 1–3; see also Cal. Dep’t of Ins., Long Term Care Insurance Outlines of Coverage, https://www.insurance.ca.gov (last visited Apr. 20, 2026).

[19] See Long-Term Care Insurance Model Regulation § 1, Nat’l Ass’n of Ins. Comm’rs Model No. 641 (2017); see also Nat’l Ass’n of Ins. Comm’rs, NAIC Membership Adopts Framework to Address Long-Term Care Insurance Rate Approvals (Apr. 19, 2022).

[20] See Membership, Interstate Ins. Prod. Regul. Comm’n, https://www.insurancecompact.org (last visited Apr. 20, 2026); see also Interstate Ins. Prod. Regul. Comm’n, Core Standards for Individual Long-Term Care Insurance Policies (effective Oct. 10, 2017); see also Interstate Ins. Prod. Regul. Comm’n, Rate Filing Standards for Individual Long-Term Care Insurance—Modified Rate Schedules (effective Oct. 10, 2017).

[21] Matt Morton & Kirill Grin, NAIC LTC Data Call Workstream #6: Suggested Public Presentation 3–6, 15–18 (June 25, 2021).

[22] See Mike Bergerson, Andrew Duxbury & Courtney Williamson, Long-Term Care Rate Increase Survey 3–5, Milliman (Apr. 2025); see also Mike Hebig & Jeff Johnson, Highlights from a Rate Increase Survey, Soc’y of Actuaries Long-Term Care News (Feb. 2022); see also Cassi Noel & Courtney Williamson, LTC Rate Increase Landscape Update, Soc’y of Actuaries Long-Term Care News (Apr. 2025).

[23] Cal. Ins. Code § 12900 (2025).

[24] Jessica Liu & Weiling Liu, The Effect of Political Frictions on the Pricing and Supply of Insurance, 37 Rev. Fin. Stud. 1149, 1149–50 (2024).

[25] Id.

[26] See Ricardo Lara, Bulletin 2024-7: Revisions to Department Review of Complete Rate Applications 2–3, Cal. Dep’t of Ins. (Aug. 9, 2024); see also Cal. Dep’t of Ins., Commissioner Lara Moves to Implement New Insurance Rate Review Reforms (Aug. 9, 2024).

[27] See Long-Term Care Insurance Multistate Rate Review, Nat’l Ass’n of Ins. Comm’rs, https://content.naic.org (last visited Apr. 20, 2026); see also Interstate Ins. Prod. Regul. Comm’n, Membership, supra note 20; see also Bergerson et al., supra note 22, at 4–5.

[28] See generally Section I, supra; see also Cassi Noel & Courtney Williamson, LTC Rate Increase Landscape Update, Soc’y of Actuaries Long-Term Care News (Apr. 2025), https://www.soa.org/sections/long-term-care/long-term-care-newsletter/2025/ltc-2025-04-noel-williamson.

[29] See Noel & Williamson, supra note 22; see also Mike Hebig, Highlights from a Rate Increase Survey, Soc’y of Actuaries Long-Term Care News (Feb. 2022), https://www.soa.org/sections/long-term-care/long-term-care-newsletter/2022/february/ltc-2022-02-hebig; see also N.Y. Dep’t of Fin. Servs., Life Insurers: General SERFF Guidelines for Form Filings, https://www.dfs.ny.gov/apps_and_licensing/life_insurers/general_serff_guidelines_for_form_filings (last visited Apr. 21, 2026).

[30] Ass’n of Cal. Life & Health Ins. Cos., Long-Term Care Filing Experiences in California (Sept. 23, 2024).

[31] Id.

[32] State of Cal., System for Electronic Rate and Form Filing (SERFF), rate filings submitted between Jan. 1, 2020 and Dec. 31, 2025 for product types LTC03I and LTC03G, https://filingaccess.serff.com (last visited Feb. 3, 2026).

[33] Id.

[34] Id.

[35] Id.

[36] Noel & Williamson, supra note 22.

[37] Alex Reger, Long-Term Care Rate Request Data, 2010–2023, Conn. Off. of Legis. Rsch. (Sept. 27, 2023), https://www.cga.ct.gov/2023/rpt/pdf/2023-R-0226.pdf.

[38] Ass’n of Cal. Life & Health Ins. Cos., supra note 30.

[39] Id.

[40] Id.

[41] 10 Cal. Code Regs. § 2202(a)(5), (b) n.3 (2026).

[42] Life and Health Lines Office Filing Fees, Tex. Dep’t of Ins., https://www.tdi.texas.gov/life/lhfee.html (last visited Apr. 21, 2026).

[43] Cal. Dep’t of Ins., supra note 12.

[44] See Noel & Williamson, supra note 22; see also N.Y. Dep’t of Fin. Servs., supra note 29; see also SERFF, supra note 32; see also 10 Cal. Code Regs. § 2202; see also Tex. Dep’t of Ins., supra note 42; see also Cal. Dep’t of Ins., supra note 12.

[45] See Cass R. Sunstein, The Cost-Benefit Revolution 31–34 (MIT Press 2019); see also Council of Econ. Advisers, The Economic Effects of Federal Deregulation Since January 2017 (June 2019), https://trumpwhitehouse.archives.gov/wp-content/uploads/2019/06/The-Economic-Effects-of-Federal-Deregulation-Interim-Report.pdf.

[46] Ari Houser, Long-Term Services and Supports Are Becoming Even More Unaffordable for Middle-Class Americans, AARP Pub. Pol’y Inst. (Mar. 12, 2026), https://doi.org/10.26419/ppi.00400.001.

[47] Paul E. Forte, Writing Long-Term Care in a Short-Term World, Contingencies (Mar./Apr. 2024), https://actuary.org/article/writing-long-term-care-in-a-short-term-world.

[48] Scott A. White, Report Regarding Long-Term Care Insurance Premium Rate Cap Legislation, Va. State Corp. Comm’n (Nov. 1, 2023), https://rga.lis.virginia.gov/Published/2023/RD596/PDF.

[49] Long-Term Care Insurance Rate History, Cal. Dep’t of Ins., https://www.insurance.ca.gov/01-consumers/105-type/95-guides/05-health/01-ltc/rate-history.cfm (last visited Feb. 3, 2026).

[50] See Kate Bundorf, The Effects of Offering Health Plan Choice Within Employment-Based Purchasing Groups (NBER Working Paper No. w9996, Sept. 2003), https://papers.ssrn.com/sol3/papers.cfm?abstract_id=450897; see also Liran Einav & Jonathan Levin, Managed Competition in Health Insurance, 13 J. Eur. Econ. Ass’n 998, 998–1021 (Dec. 2015).

[51] See AHIP, supra note 13, at 3, 8; see also AHIP, supra note 3, at 3, 8.

[52] Hans Scheil, Long-Term Care Insurance: Traditional vs. Hybrid—What You Need to Know Before It’s Too Late, Cardinal Guide (Apr. 15, 2025), https://cardinalguide.com/posts/long-term-care-insurance-traditional-vs-hybrid-what-you-need-to-know-before-its-too-late.

[53] See Am. Acad. of Actuaries, supra note 7, at 1; see also Am. Ass’n for Long-Term Care Ins., 2022 Linked Benefit Long-Term Care Insurance Price Index Released (Apr. 4, 2022), https://www.aaltci.org/news/long-term-care-insurance-association-news/2022-linked-benefit-long-term-care-insurance-price-index-released.

[54] See Am. Ass’n for Long-Term Care Ins., Buyers Ages for LTC Insurance Reported (Nov. 6, 2020), https://www.aaltci.org/news/long-term-care-insurance-association-news/buyers-ages-for-ltc-insurance-reported; see also Am. Ass’n for Long-Term Care Ins., Most Long-Term Care Insurance Buyers Between 50 and 69 (Dec. 18, 2019), https://www.aaltci.org/news/long-term-care-insurance-news/most-long-term-care-insurance-buyers-between-50-and-69.

[55] See Am. Ass’n for Long-Term Care Ins., 2025 Long-Term Care Insurance Facts, Data, Prices and Statistics—2025 Reports, https://www.aaltci.org/long-term-care-insurance/learning-center/ltcfacts-2025.php (last visited Apr. 21, 2026); see also Am. Ass’n for Long-Term Care Ins., Nearly Half of Oldest Long-Term Care Insurance Applicants Declined (July 15, 2022), https://www.aaltci.org/news/long-term-care-insurance-association-news/applicants-declined.

[56] See Am. Acad. of Actuaries, supra note 7, at 1; see also Nat’l Ass’n of Ins. Comm’rs, Insurance Topics | Long-Term Care Insurance (updated Feb. 24, 2025), https://content.naic.org/insurance-topics/long-term-care-insurance.

[57] Nat’l Ass’n of Ins. Comm’rs, supra note 56.

[58] See Laura Skufca & Gerard “Chuck” Rainville, 2021 Caregiving Out-of-Pocket Costs Study, AARP Pub. Pol’y Inst. (June 29, 2021), https://www.aarp.org/pri/topics/ltss/family-caregiving/family-caregivers-cost-survey; see also Ari Houser, Selena Caldera, Brendan Flinn & Rita Choula, Valuing the Invaluable 2026: Family Caregivers’ Contribution Reaches $1 Trillion, AARP Pub. Pol’y Inst. (Mar. 26, 2026), https://doi.org/10.26419/ppi.00402.001.

[59] Cal. Dep’t of Ins., supra note 26.

[60] Cal. Dep’t of Ins., supra note 12.

[61] See Cal. Dep’t of Ins., Filing Fees for Documents Approved in Filings Submitted to Policy Regulation and Approval Bureau I, https://www.insurance.ca.gov/0250-insurers/0300-insurers/0100-applications/pab/FilingFeesPAB.cfm (last visited Apr. 21, 2026); see also 10 Cal. Code Regs. § 2202 (2026); see also Tex. Dep’t of Ins., supra note 42.

[62] Nat’l Ass’n of Ins. Comm’rs, Limited Long-Term Care Insurance Model Act §§ 1, 4, Model No. 642 (2018), https://content.naic.org/sites/default/files/model-law-642.pdf.

[63] Nat’l Ass’n of Ins. Comm’rs, Limited Long-Term Care Insurance Model Regulation §§ 8, 12–14, 22–23, Model No. 643 (2018), https://content.naic.org/sites/default/files/inline-files/MDL-643.pdf.

[64] See About, Interstate Ins. Prod. Regul. Comm’n, https://www.insurancecompact.org/about (last visited Apr. 21, 2026); see also Membership, Interstate Ins. Prod. Regul. Comm’n, https://www.insurancecompact.org/regulator-resources/membership (last visited Apr. 21, 2026); see also FAQ, Interstate Ins. Prod. Regul. Comm’n, https://www.insurancecompact.org/about/faq (last visited Apr. 21, 2026).

[65] See Core Standards for Individual Long-Term Care Insurance Policies, Interstate Ins. Prod. Regul. Comm’n, https://www.insurancecompact.org/standards/record-adopted-standards/core-standards-individual-long-term-care-insurance-policies (last visited Apr. 21, 2026); see also Individual Long-Term Care Insurance Application Standards, Interstate Ins. Prod. Regul. Comm’n, https://www.insurancecompact.org/standards/record-adopted-standards/individual-long-term-care-insurance-application-standards (last visited Apr. 21, 2026); see also Standards for Long-Term Care Insurance Benefit Features, Interstate Ins. Prod. Regul. Comm’n, https://www.insurancecompact.org/standards/record-adopted-standards/standards-long-term-care-insurance-benefit-features (last visited Apr. 21, 2026); see also Rate Filing Standards for Individual Long-Term Care Insurance—Modified Rate Schedules, Interstate Ins. Prod. Regul. Comm’n, https://www.insurancecompact.org/standards/record-adopted-standards/rate-filing-standards-individual-long-term-care-insurance (last visited Apr. 21, 2026); see also Rate Filing Standards for Individual Long-Term Care Insurance—Issue Age Rate Schedules, Interstate Ins. Prod. Regul. Comm’n, https://www.insurancecompact.org/standards/record-adopted-standards/rate-filing-standards-individual-long-term-care-insurance-issue (last visited Apr. 21, 2026); see also Standards for Filing Revisions to In-Force Rate Filing Schedules for Individual Long-Term Care Insurance, Interstate Ins. Prod. Regul. Comm’n, https://www.insurancecompact.org/standards/record-adopted-standards/standards-filing-revisions-force-rate-filing-schedules (last visited Apr. 21, 2026).

[66] See Interstate Ins. Prod. Regul. Comm’n, 2022 Annual Report 16–17, https://www.insurancecompact.org/sites/default/files/2023-03/2022-annual-report.pdf; see also Interstate Ins. Prod. Regul. Comm’n, 2021 Annual Report 14–15, https://www.insurancecompact.org/sites/default/files/2022-11/annual_report_2021.pdf; see also Interstate Ins. Prod. Regul. Comm’n, Turnaround Times (Feb. 20, 2025), https://www.insurancecompact.org/news-events/news/turnaround-times.

Executive Summary

California Assembly Bill 1776, the COMPETE Act, would expand the state’s Cartwright Act to reach single-firm conduct for the first time in the Act’s 119-year history. It would create state-law liability for monopolization, attempted monopolization, maintenance of monopoly power, and monopsonization, moving California well beyond federal Section 2 doctrine.

The bill lowers the core thresholds federal courts use to separate anticompetitive conduct from competition on the merits. It makes optional market-power thresholds, recoupment in predatory-pricing cases, cross-market balancing for multi-sided platforms, as-efficient-competitor analysis, and quantitative evidence of harm. It then directs courts to interpret the law liberally and maximize deterrence.

That framework invites over-enforcement. It would expose aggressive price competition to predatory-pricing claims, require one-sided analysis of multi-sided platforms, extend monopsony liability into unsettled labor-market doctrine, and replace the consumer welfare standard with a broad mandate to protect “all trade participants” and broader social interests.

The likely result is more strategic litigation, more settlements driven by defense costs, and more pressure on firms to raise prices, reduce service quality, limit product integration, or alter employment practices to avoid liability. Because many firms serving California operate nationally, AB 1776 would also set de facto national antitrust rules and raise unresolved dormant Commerce Clause concerns.

The Legislature should proceed cautiously before enacting the COMPETE Act as written. Any single-firm conduct reform should restore recoupment, permit cross-market balancing, require market-power thresholds, retain the consumer welfare standard, address extraterritorial effects, and exclude single-firm conduct from criminal liability.

I.         Introduction

California Assembly Bill 1776, the COMPETE Act (Competition and Opportunity in Markets for a Prosperous, Equitable, and Transparent Economy),[1] would significantly expand the Cartwright Act[2] —California’s primary antitrust statute—by extending it to reach single-firm conduct for the first time in the Act’s 119-year history. To date, the Cartwright Act has targeted concerted action, leaving unilateral conduct largely to federal antitrust law—the Sherman Act, the Clayton Act, and the Federal Trade Commission Act.[3] AB 1776 would upend that allocation by creating state-law liability for monopolization, attempted monopolization, maintenance of monopoly power, and monopsonization by a single firm.

Proponents frame this change as closing a “gap” in California law. But the absence of single-firm liability under the Cartwright Act reflects a longstanding division of labor between state and federal regimes, under which Section 2 of the Sherman Act governs unilateral conduct.[4] While some states maintain single-firm conduct provisions, many track federal law or include harmonization clauses. Those clauses typically direct courts to interpret state statutes consistently with federal law, even if they do not require strict conformity.[5] AB 1776 breaks from that approach. It expressly provides that federal precedent is not binding, inviting state courts to adjudicate claims outside established federal doctrine.

At the same time, the bill discards much of the analytical framework that gives content to single-firm liability. Section 16732 lists 10 requirements commonly applied by federal courts and provides that none is necessary to establish liability under California law. Section 16733 instructs courts to “liberally interpret” the statute while “maximizing” deterrence. Section 16730 further provides that federal precedent is relevant only to the extent it is consistent with California law. The bill thus expands liability while rejecting the doctrinal tools that have traditionally structured and limited its application—without offering a coherent alternative.

The bill rests on two contestable premises. First, it assumes that exclusionary single-firm conduct is widespread and harmful enough to justify a new state cause of action, but the statutory text does not specify that theory of harm, and the economics literature does not support it. Second, it rejects the doctrinal standards federal courts use to distinguish anticompetitive conduct from vigorous competition, yet offers no workable substitute.

TABLE 1: Federal Law vs. AB 1776

AB 1776’s remedies—criminal liability, treble damages, private rights of action with low pleading thresholds, and no market-power requirement—are disproportionate to the harms its proponents identify, even on their own terms. The bill also lowers core liability standards. It eliminates any recoupment requirement for predatory pricing, bars cross-market balancing for multi-sided platforms, and does not require proof of market power. As a result, plaintiffs—both public and private—could prevail on claims that federal courts would reject.

These features carry significant spillover effects. The firms most likely to face liability under AB 1776 operate nationally or globally. California courts would therefore set de facto national antitrust standards by applying a lower liability threshold.

The bill applies to any firm engaged in trade or commerce in California, regardless of where it is incorporated or headquartered. A manufacturer based in Ohio, a Delaware-incorporated retailer, or a Texas-based logistics firm all face Cartwright Act liability if their conduct affects California markets or participants. The statute imposes no industry or size limits. Any company that employs California workers, sells to California consumers, or competes in California markets falls within its scope.

That breadth has practical consequences. The firms most exposed to AB 1776 are not primarily California-based. They are national and multinational companies whose California operations cannot be separated from their broader business models. Firms will not maintain one set of practices for California and another elsewhere. Instead, they will adjust their conduct nationwide to comply. The bill’s extraterritorial effects are not incidental—they are the predictable result of applying a state-specific antitrust regime to firms that operate across integrated national and global markets.

II.      Federal–State Allocation of Antitrust Authority

The Cartwright Act, enacted in 1907, formed part of a broader wave of state competition laws aimed at curbing cartel behavior and supplementing common-law remedies for monopolies and price-fixing.[6] The statute targets “trusts,” defined as combinations of two or more persons to restrain trade or suppress competition. That focus on coordinated conduct has anchored enforcement against horizontal price-fixing, bid-rigging, and market-allocation agreements.

By contrast, Section 2 of the Sherman Act governs unilateral conduct. A single firm violates federal law when it acquires or maintains monopoly power through exclusionary practices. The Cartwright Act contains no comparable provision. The California Supreme Court confirmed in Cianci v. Superior Court that the Act is “broader in range and deeper in reach” than the Sherman Act within its domain,[7] but courts have consistently recognized that this domain does not extend to unilateral conduct.[8]

As a result, California law does not provide a direct cause of action against a single firm that refuses to deal with a rival on terms that would enable competition, acquires nascent competitors to eliminate emerging threats, or uses market power to impose exclusionary contracts. Federal law reaches these forms of conduct under Section 2, but federal litigation is costly, time-consuming, and imposes demanding evidentiary burdens.

The California Law Revision Commission (CLRC) argues that state antitrust law can complement federal enforcement by expanding enforcement capacity, enabling additional private plaintiffs, and tailoring remedies to California-specific conditions.[9]

III.    The CLRC Framework and AB 1776’s Design

Assembly Bill 1776 largely codifies—nearly verbatim—the recommendations of the California Law Revision Commission (CLRC), which spent several years evaluating whether to extend state antitrust law to single-firm conduct. The bill adopts the CLRC’s core design choices: an industry-neutral framework that applies across the economy, a directive to maximize deterrence, and a break from federal antitrust doctrine as a binding guide.

Understanding AB 1776 therefore requires examining both the CLRC’s process and its proposed statutory framework. The bill translates those recommendations into enforceable law by expanding liability to unilateral conduct, lowering or eliminating traditional evidentiary screens, and granting courts broad discretion to define the contours of liability without a clear alternative framework.

A.      The CLRC’s Break from Federal Antitrust

The CLRC spent several years reviewing the state’s antitrust laws, culminating in recommendations that underpin Assembly Bill 1776. The CLRC proposed a new single-firm conduct provision that applies across the entire California economy, rather than targeting specific sectors such as technology. AB 1776 adopts this industry-neutral approach.

At the same time, the CLRC’s recommendations—enacted largely without modification—depart sharply from established federal antitrust principles. The CLRC proposed instructing courts to maximize deterrence of antitrust violations, drawing on Clayworth v. Pfizer, Inc.[10] AB 1776 § 16733 codifies that directive verbatim. The CLRC also introduced “Judicial Guidance”—a list of 10 federal evidentiary screens that “may constitute evidence” of a violation but are not required to establish liability.[11] AB 1776 § 16732 adopts that list in full, making optional key elements of federal analysis, including market power (subdivision (i)), recoupment in predatory pricing (subdivision (g)), cross-market balancing for multi-sided platforms (subdivision (f)), and the as-efficient-competitor test (subdivision (h)). The CLRC further proposed barring courts from offsetting harms in one market with benefits in another; AB 1776 § 16731(b) adopts that rule. In each instance, the bill tracks the CLRC’s recommended language.

While AB 1776 closely follows the CLRC’s recommendations, ICLE scholars and others sharply criticized the Commission’s process and conclusions. In formal comments, they argue that the decision to untether California antitrust law from the federal error-cost framework and the consumer welfare standard is misguided.[12] By favoring the risk of over-enforcement (false positives) over under-enforcement (false negatives), the CLRC’s approach risks chilling procompetitive conduct, including price cutting and vertical integration.

These critics also contend that the CLRC’s framework effectively rejects key U.S. Supreme Court precedents, including Brooke Group, Trinko, and Amex, and instead aligns California law more closely with European Union competition policy.[13] They further argue that the CLRC assumes federal antitrust law has failed, yet offers little empirical evidence that California consumers or businesses suffer from reduced competition due to gaps in current enforcement.

In practice, the CLRC’s framework lowers the evidentiary burden for plaintiffs without defining a clear alternative standard. It directs courts to reduce reliance on established screens but leaves them to determine what suffices to establish liability. These changes benefit plaintiffs’ attorneys, who gain broader grounds for treble-damages litigation; state enforcers, who gain expanded authority; and competitors seeking new claims against dominant firms. While these groups have legitimate interests, a deterrence-maximizing approach departs from the prevailing law & economics consensus, which emphasizes protecting consumer welfare and avoiding over-deterrence that can harm the broader economy.

B.       AB 1776’s Statutory Framework

AB 1776 would add Sections 16730 through 16733 to the California Business and Professions Code. The bill’s enactment clause states: “An act to add Sections 16730, 16731, 16732, and 16733 to the Business and Professions Code, relating to business regulations.” It does not amend Section 16720, the Cartwright Act’s definition of “trust,” which still requires “two or more persons.” Instead, new Section 16731(a)(1) provides that “restraint of trade” includes conduct “cognizable under Section 16720, whether directed, caused, or performed by one or more persons.” This incorporation-by-reference approach extends Cartwright Act liability to single-firm conduct—covering unilateral pricing, exclusive dealing, refusals to deal, product design, and contracting practices across all industries—without revising the statute’s core definitions.

Section 16731 makes it unlawful for one or more persons to restrain trade or to “monopolize or monopsonize, to attempt to monopolize or monopsonize, to maintain a monopoly or monopsony, or to combine or conspire with another person to monopolize or monopsonize.” The inclusion of “monopsonize” marks a significant expansion. It introduces buyer-side liability into California antitrust law, extending coverage to labor markets and other input markets.

Section 16730(b) underscores that shift by identifying “workers’ freedom to choose employment” as a protected competitive interest. The bill thus explicitly incorporates labor-market competition into the statute’s core objectives.

Section 16731 requires courts to evaluate anticompetitive effects and procompetitive justifications “within the same relevant market,” prohibiting cross-market balancing.

Section 16732 lists factors that may inform liability but are not required. These include: termination of a prior course of dealing; differential treatment of rivals; below-cost pricing; conduct that “makes no economic sense” absent a harmful purpose; quantitative evidence of harm; harm on multiple sides of a platform or harm outweighing benefits on another side; a probability of recoupment in predatory pricing; comparison to an as-efficient competitor; market-share or market-power thresholds recognized under federal Section 2; and a defined relevant market where direct evidence of market power exists.

Section 16733 directs courts to interpret California antitrust law liberally and to remain “mindful that California favors ‘maximizing’ effective deterrence of antitrust violations.” Section 16730 further provides that federal precedent is not binding and may be considered only to the extent it is “consistent with California law.”

Enforcement follows existing Cartwright Act mechanisms. Private plaintiffs, the attorney general, district attorneys, and qualifying city attorneys may bring suit. Available remedies include injunctive relief, treble damages, and attorneys’ fees. The Act’s criminal provisions also apply to violations of AB 1776.

IV.    Economic Tradeoffs and Doctrinal Shifts

AB 1776 rests on two core assumptions: that exclusionary single-firm conduct is both widespread and harmful enough to justify a deterrence-maximizing regime, and that courts can identify such conduct without the analytical screens that structure federal antitrust law. Both assumptions warrant scrutiny.

Sections IV.A through IV.C examine the economic and legal implications of that shift. Together, they highlight three throughlines: the tradeoff between false positives and false negatives, the risk of protecting competitors rather than competition, and the move away from the consumer welfare standard toward a set of competing policy goals without a clear method of resolution. By lowering evidentiary thresholds and prioritizing deterrence, AB 1776 increases the likelihood of over-enforcement, weakens the distinction between harmful and beneficial conduct, and leaves courts to resolve competing interests without a principled framework.

A.      Error Costs and Overdeterrence

Antitrust enforcement produces two types of errors. A false positive condemns procompetitive conduct—penalizing behavior that benefits consumers. A false negative allows anticompetitive conduct to persist without remedy. Frank H. Easterbrook, writing before his appointment to the 7th U.S. Circuit Court of Appeals, identified the asymmetry between these errors that makes careful calibration essential: markets can partially self-correct false negatives, because above-competitive profits attract entry that erodes monopoly power over time.[14] False positives do not self-correct. An erroneous condemnation deters the challenged conduct across future periods through legal precedent, chilling similar conduct by other firms.^[15] ICLE scholars apply this framework to California’s reform proposals in comments to the CLRC and subsequent analyses.[16]

A maximally deterrent antitrust regime increases the risk of false positives by design. It captures more anticompetitive conduct, but only by also condemning more procompetitive conduct that less aggressive standards would permit.

The CLRC explicitly prioritizes maximizing deterrence over calibrating enforcement to specific harms. It recommends departing from federal standards—such as recoupment and below-cost pricing requirements—which it characterizes as “rigid rules” that can “unduly restrict” enforcement.[17] While the CLRC rejects an “abuse of dominance” standard due to vague thresholds, it adopts a broad “restraint of trade” framework to reach a wider range of conduct than federal law.[18] AB 1776’s directive to maximize deterrence reflects a policy choice: accept more false positives to reduce false negatives, and systematically favor plaintiffs across theories of unilateral harm, regardless of the strength of the underlying claims.

ICLE’s comments note that this approach mirrors the European Union’s precautionary approach to antitrust, which assumes markets may not self-correct effectively. That approach can yield benefits, but “comes, almost by definition, at the expense of short-term growth.[19] The comments warn that adopting such a framework “is a costly policy stance in those circumstances where it is not clearly warranted by underlying risk and uncertainty,” particularly for a state like California, whose economy depends on innovation and startup growth.

B.       Competition vs. Competitors

The Supreme Court has long held that antitrust law protects “competition, not competitors.”[20] That distinction determines whether antitrust law serves consumers—here, California citizens broadly—or instead protects individual rivals. A firm that wins customers through lower prices harms competitors but benefits consumers and should not face antitrust liability.^[21] By contrast, a firm that uses exclusionary threats to cut off rivals harms the competitive process itself and may warrant intervention. The analytical tools that AB 1776 makes optional—market-power analysis, recoupment, and multi-sided platform balancing—exist to distinguish between these scenarios.

AB 1776 replaces that filtering function with a directive to maximize deterrence. Courts instructed to resolve ambiguity broadly and favor deterrence will impose liability in more cases, including those where the conduct harms a rival but not competition. The result shifts the statute’s focus from protecting competition to protecting competitors.

The empirical case for this expansion is weak. In his August 2024 presentation to the CLRC, ICLE President Geoffrey Manne showed that claims of rising market concentration remain contested.[22] Some studies based on publicly traded firms and broad industry codes suggest increasing concentration, but more granular analyses point in the opposite direction. Gerard Hoberg and Gordon Phillips, accounting for multi-industry firms, find declining average Herfindahl-Hirschman Index (HHI) scores.[23] C. Lanier Benkard and co-authors report that “decreases in concentration over time are broad-based” and that this result “contradicts the prevailing popular opinion.”[24] As Manne emphasized, concentration data alone “says nothing about the amount of competition,” and therefore has no direct normative implication for antitrust policy.[25]

C.      The Consumer Welfare Standard and Its Proposed Replacement

Federal antitrust law has long organized liability around consumer welfare—harm to consumers through higher prices, reduced output, or diminished quality. The Supreme Court described the Sherman Act as “a consumer welfare prescription” in Reiter v. Sonotone Corp.[26] This standard limits liability to conduct that harms consumers, not merely rivals, and provides a principled boundary for antitrust enforcement.

AB 1776 replaces that framework with a broader and less defined set of objectives. The bill declares that California antitrust law protects (1) “free and fair competition” (2) for “all trade participants, including workers and consumers,” and (3) “an environment that is conducive to the preservation of our democratic, political, and social institutions.”[27] These goals are distinct and often in tension, yet the statute does not explain how courts should resolve those conflicts. ICLE scholars warned during the CLRC process that extending protection to “trading partners” as a class—rather than focusing on consumers and the competitive process—risks politicizing enforcement and shielding less-efficient firms at consumers’ expense.[28]

These tensions arise in routine cases. A firm that lowers prices benefits consumers but harms rivals. A firm that integrates complementary features improves quality for users but reduces demand for standalone products. A firm that grows through successful competition may increase concentration, raising political concerns even if its conduct benefits consumers. Under the consumer welfare standard, courts resolve these conflicts by asking whether the net effect on consumers is positive. Under AB 1776’s “all trade participants” standard—combined with a directive to maximize deterrence—no comparable principle applies. Courts must weigh competing interests case by case, producing outcomes that are difficult to predict and hard for firms to anticipate in structuring their conduct.

V.      Predatory Pricing and the Recoupment Requirement

Federal predatory-pricing doctrine rests on a simple economic insight: below-cost pricing harms consumers only if the firm can later recoup its losses through supracompetitive prices. AB 1776 discards that requirement without replacing it, exposing firms to liability for the very conduct—aggressive price competition—that antitrust law seeks to protect.

Sections V.A and V.B explain the implications of that shift. Together, they show that removing the Brooke Group recoupment screen eliminates a core safeguard against false positives, leaves courts without a coherent test to distinguish harmful from beneficial pricing, and predictably deters procompetitive price cutting. The result is a regime that increases liability risk while weakening the economic foundation of predatory-pricing law, to the detriment of consumers.

A.      Brooke Group and Recoupment

In Brooke Group Ltd. v. Brown & Williamson Tobacco Corp.,[29] the Supreme Court held that a predatory-pricing claim requires proof of two elements: (1) pricing below an appropriate measure of cost, and (2) a dangerous probability of recouping the losses through later supracompetitive pricing. The Court explained why both elements matter: “Without a dangerous probability of recoupment, it is highly unlikely that a firm would engage in predatory pricing,” and absent recoupment, “predatory pricing produces lower aggregate prices in the market, and consumer welfare is enhanced.”[30]

The logic of the recoupment requirement is straightforward. A firm that prices below cost without any realistic prospect of later monopoly pricing incurs losses with no path to recovery. Those low prices transfer value to consumers on every unit sold. The firm must eventually raise prices to competitive levels (inviting entry), exit the market, or continue absorbing losses—none of which reflects a strategy that harms consumers. Predatory pricing becomes plausible only when below-cost pricing represents a temporary sacrifice backed by an expectation of future monopoly pricing. Recoupment tests whether that expectation is economically credible.

Then-Judge Stephen Breyer underscored the error-cost stakes in Barry Wright Corp. v. ITT Grinnell Corp.: “the consequence of a mistake here is not simply to force a firm to forego legitimate business activity it wishes to pursue; rather, it is to penalize a procompetitive price cut, perhaps the most desirable activity (from an antitrust perspective) that can take place in a concentrated industry.”[31] ICLE’s comments to the CLRC and subsequent scholarship argue that eliminating the Brooke Group recoupment requirement would move California law toward the European model, which assesses pricing without regard to recoupment—a standard ICLE describes as having “no basis in economic theory or evidence.”[32]

B.       Eliminating Recoupment and Its Effects

Section 16732 of AB 1776 provides that a plaintiff need not establish “whether a defendant is likely to recoup losses from below-cost pricing” to prove a predatory-pricing violation. The bill offers no substitute analytical test. Instead, courts must evaluate such claims under a general mandate to interpret antitrust law liberally and maximize deterrence.

Under this framework, a firm that prices aggressively to gain market share, match a competitor’s promotion, or expand demand for a new product could face liability without any showing of market power, exclusionary intent, or a realistic prospect of recouping losses through supracompetitive pricing. A plaintiff need only persuade a court that prices fell below some measure of cost and that the conduct harmed a trade participant enough to justify liability.

Firms will respond predictably. The expected cost of a predatory-pricing claim equals the probability of liability multiplied by expected damages—treble damages plus attorneys’ fees. When firms cannot predict how courts will characterize their pricing—because no recoupment screen anchors the analysis and courts must maximize deterrence—they will price more conservatively. The conduct most likely to be deterred is precisely the aggressive price competition that antitrust law has traditionally protected.

The Supreme Court adopted the recoupment requirement in Brooke Group to address this problem. Before that decision, courts often condemned aggressive pricing based on intent or market structure, which led to frequent false positives.[33] Brooke Group added recoupment to reduce the risk of penalizing procompetitive price cuts. AB 1776 effectively returns California to the pre-Brooke Group regime—one federal courts abandoned because it deterred beneficial price competition.

Consumers bear the cost of that shift. When firms avoid aggressive pricing to limit litigation risk, prices rise above competitive levels. In concentrated markets, this means consumers pay more—not because firms exercise monopoly power, but because legal risk discourages price competition. Even familiar examples of sustained low pricing—such as Costco’s $1.50 hot dog and soda combo or its $4.99 rotisserie chicken—become harder to sustain under a regime that treats aggressive pricing as presumptively suspect.

VI.    Multi-Sided Markets and the Rejection of Amex

Many of California’s most significant firms—including Google, Meta, Apple, and Amazon—operate multi-sided platforms that fund services on one side of the market with revenue from another. The Supreme Court recognized in Amex that platform conduct must be evaluated across both sides to avoid systematically biased results. AB 1776 prohibits that approach, requiring courts to assess competitive harm one side at a time while disregarding offsetting benefits on the other.

Sections VI.A through VI.D show how this shift departs from the economics of platform competition. Multi-sided markets depend on cross-market interactions, and pricing decisions on one side cannot be understood in isolation. By rejecting cross-market balancing and making multi-sided harm optional, AB 1776 removes the tools needed to distinguish harmful conduct from business models that benefit consumers overall. The result is a framework that risks misidentifying harm, over-deterring efficient cross-subsidization, and exposing a wide range of industries—not just technology—to liability for standard competitive practices.

A.      Multi-Sided Platforms and Cross-Market Effects

A multi-sided platform serves multiple groups of users simultaneously, creating value by facilitating interactions among them. The economics of these platforms differ fundamentally from single-sided firms. A credit card network becomes more valuable to merchants as more consumers carry the card, and more valuable to consumers as more merchants accept it. These indirect network effects mean that pricing decisions on one side of the platform shape the size, composition, and behavior of the other.

Pricing therefore operates as an integrated system. A platform that charges merchants higher fees may use that revenue to fund consumer rewards, fraud protection, or broader acceptance, attracting more users and increasing transaction volume for merchants. The causal relationship runs in both directions. Evaluating a price increase on one side of the platform without accounting for effects on the other side produces an incomplete—and often misleading—assessment. That does not mean harms on one side can never outweigh benefits on the other. It means cross-market effects are central to competition among multi-sided platforms and cannot be ignored without risking harm to consumers.

Many of California’s most significant firms operate multi-sided platforms. Google funds free search and maps through advertising. Meta offers free social networking supported by advertising revenue. Apple’s App Store connects developers and users through a commission-based model. Amazon’s marketplace links third-party sellers with consumers while competing as a retailer. In each case, pricing decisions on one side of the platform are analytically inseparable from their effects on the other.

B.       Amex and Two-Sided Markets

The Supreme Court addressed the implications of platform economics for antitrust analysis in Ohio v. American Express Co.[34] The Court held that, because Amex operates a two-sided transaction platform, evidence of a price increase on the merchant side “cannot, by itself, demonstrate an anticompetitive exercise of market power.”[35] It emphasized that “the two-sided market for credit-card transactions should be analyzed as a whole,” warning that focusing on only one side “tends to distort the competition that actually exists” and risks “mistaken inferences” that could chill legitimate competition.”[36] The Court therefore requires plaintiffs in two-sided platform cases to show anticompetitive effects across both sides of the market before the burden shifts to defendants to offer procompetitive justifications.

This framework reflects the underlying economics of multi-sided platforms. Costs imposed on one side and benefits delivered on the other are linked through the platform’s business model. Evaluating only costs or only benefits produces a systematically biased assessment of welfare effects. Herbert Hovenkamp, a leading antitrust scholar at the University of Pennsylvania Law School, explains that market power on a multi-sided platform cannot be inferred from conditions on a single side, because apparent price increases or market power may be offset by services or subsidies on the other.[37]

ICLE applies this logic directly to the CLRC’s proposed rejection of Amex. Evidence of a price effect on one side of a two-sided platform may reflect neutral, procompetitive, or anticompetitive conduct. Distinguishing among those possibilities requires examining both sides of the platform, rather than isolating one dimension of the market.[38]

C.      Banning Cross-Market Balancing

Two provisions of AB 1776 displace the Amex framework. Section 16731(b) requires courts to evaluate anticompetitive effects and procompetitive justifications “within the same relevant market,” prohibiting cross-market balancing. Section 16732(f) further provides that plaintiffs need not show harm on more than one side of a multi-sided platform, or that harm on one side outweighs benefits on another.

Read together, these provisions require courts to assess platform conduct one side at a time and to disregard offsetting benefits on the other side. A platform that raises advertising prices to fund free consumer services would be judged solely on advertiser-side costs, without credit for consumer benefits. A payment network that charges merchants higher fees while providing cardholders with superior fraud protection and rewards could face liability based only on merchant-side effects.

The bill offers no economic justification for this approach. The CLRC dismisses Amex, asserting:

[Amex] created a confusing precedent as to the type and amount of evidence needed to show harm in cases involving two sided platforms. This case also used assumptions about the interconnectedness of the two sides that may not translate to market realities in other circumstances, and could allow firms to escape antitrust liability for causing harm on one side of a platform and masking it with benefits on the other side.[39]

ICLE scholars responded:

As in the Amex case itself, such an approach would confer benefits on certain platform-business users (in Amex, retailers) at the direct expense of consumers (in Amex, literal consumers of retail goods purchased by credit card).

Adopting such an approach in California—whose economy is significantly dependent on multisided digital-platform firms, including both incumbents and startups—would imperil the state’s economic prospects and exacerbate the incentives for such firms to take jobs, investments, and tax dollars elsewhere.[40]

By prohibiting cross-market balancing, AB 1776 adopts an analytical framework that will misidentify harm in multi-sided markets. A rule that counts costs on one side without crediting benefits on the other will find liability even when overall consumer welfare improves. Courts applying this rule cannot distinguish between conduct that harms consumers and conduct that harms only competitors. Any adverse effect on one group of users becomes actionable, regardless of offsetting benefits to others.

The predictable result is liability for business models the Supreme Court declined to condemn in Amex—models that often benefit consumers overall despite imposing costs on certain commercial users. Rather than address that distinction, AB 1776 forbids courts from considering it.

The same error extends to the bill’s treatment of vertical restraints. ICLE’s comments to the CLRC synthesize a large empirical literature showing that vertical integration and related practices produce predominantly procompetitive or neutral effects.[41] As former FTC Bureau of Economics Director Francine Lafontaine explains, when firms adopt vertical restraints, they typically improve product quality and service, benefiting consumers as well as producers.[42] Even more skeptical reviews acknowledge that few studies identify vertical practices that likely harm competition.[43] AB 1776’s decision to subject these practices to heightened scrutiny under a maximize-deterrence mandate conflicts with that empirical record.

D.      One-Sided Liability for Cross-Subsidies

The prohibition on cross-market balancing exposes any firm that subsidizes one group of users with revenue from another to antitrust liability. While technology platforms provide the most visible examples, this model appears across many industries.

Google offers free search, maps, and email to consumers while charging advertisers for access to those users. Its advertising prices cannot be evaluated in isolation from the consumer services they fund. Under AB 1776, a plaintiff could challenge those prices based solely on advertiser costs, with no credit for the consumer benefits they support.

The same structure applies to Meta, Apple, and Amazon. Meta funds free social networking through advertising. Apple charges App Store commissions while providing developers and users with software review, device integration, and payment infrastructure. Amazon charges third-party sellers marketplace fees while operating fulfillment and consumer-trust systems. A one-sided liability framework counts costs imposed on one group while ignoring the benefits delivered to another, even when both arise from the same transaction.

Credit card networks operate the same two-sided model the Supreme Court analyzed in Amex. Visa and Mastercard charge merchants interchange fees while providing consumers with fraud protection, rewards, and payment convenience. Under AB 1776, a challenge to those fees would proceed without accounting for those consumer benefits. Merchants may gain; consumers lose.

This structure extends beyond technology and finance. Newspapers and broadcasters sell advertising to fund content that consumers receive without direct payment. Grocery retailers use loyalty programs and data revenues to support lower prices. Health care systems cross-subsidize services, using profitable lines to fund emergency or community care. These models depend on integrated pricing across different user groups.

AB 1776’s one-sided analysis will identify harm where a complete assessment shows net consumer benefit. Courts will evaluate pricing and revenue models without the tools to measure their full effects. Firms that compete by delivering value across multiple sides of a market face the greatest exposure.

VII.  Monopsony and Labor-Market Uncertainty

AB 1776 extends antitrust liability to monopsonization—buyer-side market power—and identifies workers’ freedom to choose employment as a protected competition interest. Federal law offers limited precedent for applying single-firm antitrust doctrine to labor markets, and the economics literature does not provide a settled framework for defining relevant labor markets or measuring employer wage-setting power.

Sections VII.A through VII.C show how the bill codifies this uncertainty. AB 1776 applies the same expanded liability structure used for product markets—no market-power threshold, liberal interpretation, and maximum deterrence—to an area where both doctrine and measurement remain underdeveloped. The result is a regime that lacks clear standards for identifying harm, extends liability across a wide range of employers, and encourages firms to adjust hiring and compensation practices to manage legal risk rather than compete aggressively for talent.

A.      Monopsony and Unsettled Economics

As noted above, AB 1776 extends the Cartwright Act to monopsonization and identifies workers’ freedom to choose employment as a protected competition interest.[44] Federal law offers little guidance on applying Section 2 to unilateral labor-market conduct by a single employer. The analytical tools for identifying monopsony power remain far less developed than those used to assess monopoly power in product markets.

Federal antitrust enforcement in labor markets has focused on horizontal agreements among employers—no-poach agreements and wage-fixing arrangements—rather than single-employer monopsonization. AB 1776 expands liability into an area where both the doctrine and the underlying economics remain unsettled.

B.       Defining Labor Markets

ICLE scholars Geoffrey Manne, Brian Albrecht, and Dirk Auer argue that the economics literature lacks a clear consensus on how to assess labor-market power.[45] They note that standard tools used in product markets—geographic market definition, substitution analysis, and price-cost measurement—do not translate cleanly to labor markets. ICLE reiterates these concerns in its CLRC comments and subsequent analysis of the California proposals.[46]

In product markets, defining a relevant market turns on substitution. The “hypothetical monopolist” test asks whether a firm controlling all supply could profitably raise prices by 5%–10%. That inquiry maps onto observable behavior: where consumers go when prices rise. Economists can answer it using price and quantity data to estimate demand elasticities and cross-elasticities across products.

In labor markets, the parallel question is whether a single employer could profitably reduce wages. But the substitution analysis is far less tractable. Workers do not switch among employers as easily as consumers switch among products. Non-wage attributes—working conditions, career opportunities, organizational culture, job security, and proximity to family or professional networks—differentiate jobs in ways that are difficult to measure.

Individual circumstances further complicate the analysis. A software engineer in San Francisco may not view a position in Austin as a substitute, regardless of higher pay, due to family ties or housing investments—or she may, if the job offers remote work.[47] Whether two positions fall within the same labor market depends on worker-specific preferences and mobility constraints that aggregate data often cannot capture.

Geographic market definition has become especially uncertain with the rise of remote work. Pre-pandemic studies relied on commuting patterns. Today, many California workers are employed by firms based in other states, and many firms recruit across multiple regions. Research finds that employees may accept 5%–25% lower compensation for remote or hybrid work.[48] At the same time, California’s employment regulations—including expense-reimbursement requirements, daily overtime rules, and numerous local minimum-wage ordinances—have led some out-of-state employers to exclude California residents from remote hiring.[49] These factors make labor-market boundaries harder to define and monopsony power more difficult to measure.

C.      Codifying Unsettled Theory

AB 1776 applies the same expanded liability structure to monopsonization that it applies to monopolization: no market-power threshold, no required methodology for measuring labor-market power, a mandate for liberal interpretation, and an instruction to maximize deterrence. Courts applying this framework will lack settled tools to define relevant labor markets, measure wage-setting power, or distinguish aggressive competition for workers from anticompetitive conduct.

The exposure extends beyond technology firms. Any employer that hires a large share of workers with specialized skills in a given area—a hospital system employing specialized nurses, a logistics firm employing warehouse workers, or a media company employing niche editorial staff—could face monopsonization claims. Without a market-power threshold, firms cannot determine in advance whether their size or hiring patterns create liability risk. The open-ended standard, combined with a deterrence mandate, invites courts to resolve uncertainty in plaintiffs’ favor.

Firms will respond by adjusting employment practices. To reduce litigation risk, employers may avoid compensation strategies or hiring practices that could be characterized as evidence of wage-setting power, even when those practices reflect competition for talent. They may limit long-term commitments or rely more heavily on outsourcing to reduce direct labor-market exposure. These responses may narrow opportunities and reduce flexibility for the workers the statute aims to protect.

VIII.      The Loss of the Consumer Welfare Standard

The consumer welfare standard gives courts a single, measurable question: whether the challenged conduct leaves consumers better off or worse. That inquiry distinguishes conduct antitrust law should reach from conduct it should not and allows courts to screen out claims that allege harm to competitors but not to competition.

AB 1776 replaces that standard with a multi-objective declaration protecting “free and fair competition” for “all trade participants” and the preservation of “democratic, political, and social institutions,” without any mechanism for resolving the conflicts and tradeoffs those objectives create. Sections VIII.A and VIII.B show that these conflicts are inevitable and that removing a clear limiting principle leaves courts to balance competing interests case by case. An “all-and-sundry welfare” approach does not guide that analysis—it invites inconsistent outcomes and risks undermining the consumer benefits antitrust law is designed to protect.

A.      Consumer Welfare as a Limiting Principle

The consumer welfare standard does not privilege consumers over workers or other market participants as a normative matter. It provides courts with a coherent, measurable limit on what conduct antitrust law should reach. As the Supreme Court stated in Reiter v. Sonotone Corp., the Sherman Act is “a consumer welfare prescription.”[50] The Court reinforced in Brunswick Corp. v. Pueblo Bowl-O-Mat, Inc. that plaintiffs must show harm to competition, not merely harm to themselves as competitors.[51] Together, these principles establish that antitrust law does not shield market participants from losing to superior rivals.

The standard shapes how courts analyze liability. When a plaintiff challenges pricing, exclusive dealing, or product integration, a court asks whether the conduct raised prices, reduced output or quality, or limited consumer choice. If not—if the conduct harmed a competitor while benefiting consumers—the court can dismiss the claim without full merits analysis. This screening function reduces litigation costs and filters out claims that, even if factually accurate, do not describe antitrust violations.

B.       No Limiting Principle

AB 1776 replaces the consumer welfare standard with a declaration that California antitrust law protects “free and fair competition” for “all trade participants, including workers and consumers,” and “an environment that is conducive to the preservation of our democratic, political, and social institutions.”

This formulation does not function as a workable legal standard. It combines three distinct objectives—protecting competition, protecting trade participants as a class, and preserving political institutions—without providing a method to resolve conflicts among them. Those conflicts arise routinely. A firm that lowers prices benefits consumers but harms rivals; both are “trade participants.” A firm that vertically integrates to secure inputs may benefit workers and customers through stability while disadvantaging independent suppliers; all fall within the statute’s scope. The consumer welfare standard resolves these tensions by asking whether consumers are better or worse off. AB 1776 provides no comparable principle. Combined with a mandate to maximize deterrence, the statute leaves courts to prioritize among competing interests case by case, producing inconsistent and unpredictable outcomes.

Murat Mungan and John Yun identify a further consequence of overbroad liability standards. When antitrust violations no longer clearly signal consumer harm, the reputational penalty that reinforces legal sanctions weakens.[52] Firms found to have harmed consumers through anticompetitive conduct face social and market consequences that amplify deterrence. Firms found liable for conduct that the public views as competitive or beneficial do not. By extending liability to conduct whose anticompetitive character is not self-evident, AB 1776 risks diluting this reputational mechanism and, with it, the overall deterrent effect of antitrust law.

IX.   Lower Liability Standards and Extraterritorial Effects

AB 1776 systematically lowers the thresholds for antitrust liability below those federal courts have developed over decades. It makes core screening tools optional—market power, recoupment, and cross-market balancing—thereby expanding liability across a wide range of conduct that federal law would not reach.

Sections IX.A through IX.C show how these lower standards extend beyond California. Because federal law does not preempt state antitrust law, and firms operating in national or digital markets cannot realistically maintain California-specific business practices, the bill’s standards would shape conduct nationwide. California enforcement actions and settlements would set de facto national rules, while imposing compliance costs on firms across the country. That extraterritorial effect raises unresolved dormant Commerce Clause concerns about whether the burdens on interstate commerce are proportionate to California’s asserted interests.

A.      Lowering Liability Thresholds

AB 1776 establishes lower thresholds for antitrust liability than federal law. Section 16732 renders optional 10 factors that federal courts treat as requirements—not out of confusion about antitrust goals, but because experience shows each screens out claims that would condemn procompetitive conduct.

The market-power threshold requires a defendant to possess monopoly power—often inferred from a market share above 50% in a properly defined relevant market—before unilateral conduct can qualify as monopolization.[53] This requirement reflects a basic constraint: a firm without market power cannot harm the competitive process through unilateral action. Without the ability to control prices or exclude rivals, practices such as aggressive pricing, exclusive dealing, and refusals to deal remain disciplined by competition. AB 1776 removes this threshold, exposing firms of any size or position to monopolization claims.

Other federal screens serve similar functions. Recoupment distinguishes temporary price competition from predation; eliminating it collapses that distinction. The Amex framework requires cross-market analysis for two-sided platforms; prohibiting it produces systematically incomplete welfare assessments. Each screen targets a known source of error in antitrust adjudication.

The as-efficient-competitor test asks whether the challenged conduct would exclude a rival that matches the defendant’s efficiency, or only less efficient firms. That comparison identifies conduct that distorts competition rather than reflects it. AB 1776 makes this test optional, allowing liability where conduct harms only rivals that cannot match the defendant’s performance. Under this approach, even a low-price offering—such as Costco’s $1.50 hot dog and soda combo—could be characterized as unlawful predatory pricing by less efficient competitors.

B.       Nationwide Spillover Effects

These lower liability thresholds have national consequences. The California attorney general and private plaintiffs may bring antitrust cases in state courts against firms that operate nationwide. When those cases succeed under AB 1776’s standards—without market-power analysis, recoupment, or multi-sided market balancing—they produce precedents and settlements that shape firms’ conduct beyond California.

A firm that settles a predatory-pricing claim under AB 1776 will not maintain one pricing strategy in California and another elsewhere. For many businesses, especially digital platforms, geographic differentiation is costly or infeasible. Settlement terms adopted to resolve a California claim will apply across the firm’s operations. California courts, applying lower standards than federal courts, would effectively set national policy for the challenged conduct.

Federal antitrust law does not preempt state antitrust law under Parker v. Brown.[54] But when firms cannot segment their practices by state, California’s rules will govern their behavior nationwide. By enacting AB 1776, the California Legislature would influence not only in-state commerce but commercial conduct across the United States.

C.      Dormant Commerce Clause Risks

AB 1776’s extraterritorial reach raises concerns under the dormant Commerce Clause, which bars states from imposing burdens on interstate commerce that are “clearly excessive in relation to the putative local benefits.”[55] Courts have not resolved whether a state antitrust law that drives nationwide changes in firm behavior—particularly through digital markets—satisfies this standard. Shira Liu concludes that the constitutional question remains unsettled.[56]

What is clear is that AB 1776’s lower liability standards would impose compliance costs on firms across the country that may exceed any in-state benefits. A firm that adjusts its national pricing strategy to avoid California liability under a no-recoupment standard changes behavior for consumers in all 50 states based solely on California law. Whether such nationwide effects are proportionate to California’s regulatory interests remains unanswered by courts and unaddressed by the Legislature.

X.      Strategic Litigation and Competitive Distortion

AB 1776’s combination of broad liability, low analytical thresholds, and treble damages creates strong incentives for strategic litigation—suits aimed at extracting settlements through defense costs rather than remedying anticompetitive harm. By weakening screening mechanisms and prioritizing deterrence, the bill increases both the volume and leverage of weak claims, shifting competitive disputes from the market to the courtroom.

A.      Strategic Litigation Incentives

AB 1776’s combination of low pleading standards, broad liability categories, no market-power threshold, and treble damages creates strong incentives for strategic antitrust litigation—claims filed not to remedy anticompetitive harm, but to leverage the cost of defense into favorable settlements.

When a statute presumes liability across a wide range of conduct and removes the analytical screens that would allow early dismissal, defending even a weak claim becomes costly. Antitrust litigation routinely requires extensive discovery, expert analysis, and depositions, often costing millions of dollars. A defendant that expects to prevail on the merits may still choose to settle because the cost of defense exceeds the cost of settlement. The resulting terms—pricing concessions, supply commitments, data-sharing obligations, or limits on business practices—reflect litigation pressure, not a judicial finding of anticompetitive conduct.

AB 1776’s directive to interpret antitrust law liberally and maximize deterrence intensifies these incentives. Courts operating under that mandate will be less likely to dismiss claims at the pleading stage, more likely to treat ambiguous evidence as supporting liability, and more inclined to resolve legal uncertainty against defendants. Each of these effects increases the expected value of weak claims—and, in turn, the number of such claims filed.

B.       Litigation as Competition

In practice, AB 1776’s private right of action will primarily benefit the plaintiffs’ bar and rivals seeking advantages they cannot achieve through competition. A less-efficient firm competing with a larger rival may be unable to match price or quality. Under AB 1776, it can instead threaten litigation—alleging predatory pricing (no recoupment required), cross-market harm in a platform business (no balancing permitted), or monopsony in labor markets (no market-power threshold). Each theory forces the defendant to incur substantial defense costs and creates settlement pressure regardless of the claim’s merits.

This form of litigation does not protect consumers. It protects the litigating rival at consumers’ expense. Settlements may require defendants to raise prices to avoid further predation claims, reduce the quality or scope of platform services to limit cross-market effects, or constrain employment practices to mitigate monopsony allegations. Each outcome benefits the rival’s competitive position while harming consumers or workers.

XI.   Comparative Frameworks and Evidence

AB 1776 does not operate in isolation. Its departures from established doctrine are best assessed against three benchmarks: federal Section 2 law, which the bill systematically weakens across key elements; New York’s Twenty-First Century Antitrust Act, which has repeatedly stalled under scrutiny; and the European Union’s Digital Markets Act, whose early enforcement record illustrates the practical effects of a similar regulatory approach.

Sections XI.A through XI.C show a consistent pattern. Where AB 1776 diverges most sharply from established frameworks—by lowering liability thresholds, rejecting analytical safeguards, and prioritizing deterrence—it aligns with approaches that have either failed to gain legislative traction or produced mixed results in practice.

A.      More Plaintiff-Favorable Standards

Table 2 compares AB 1776 with federal Section 2 doctrine across the doctrinal elements most relevant to the bill’s effects.

TABLE 2: Federal Law vs. AB 1776

AB 1776 does not simply fill the Cartwright Act’s historical gap on single-firm conduct. It adopts more plaintiff-favorable standards on each contested element, allowing the California attorney general and private plaintiffs to prevail on claims that federal courts would reject.

B.       Lessons from New York

New York State has considered extending its antitrust law to single-firm conduct since at least the 2019–2020 legislative session, when state Sen. Michael Gianaris (D-Queens) introduced the Twenty-First Century Antitrust Act as S. 8700-A.[57] He has introduced similar versions in each subsequent session.[58] The most recent version, S. 335, was introduced in January 2025.

The proposal would amend the Donnelly Act—New York’s primary antitrust statute governing restraints of trade—in two key ways. First, it would prohibit monopolization, attempted monopolization, and conspiracy to monopolize, treating unilateral conduct similarly to agreements. Second, it would establish an “abuse of dominance” standard for firms with a dominant position in product or labor markets. That standard would allow courts to infer dominance from direct or indirect evidence, including the ability to impose supracompetitive prices or subcompetitive wages. The bill would thus lower the legal thresholds required to establish antitrust violations.

In October 2023, Sen. Gianaris presented the proposal to the CLRC’s Single-Firm Conduct Working Group, which rejected it. The group concluded that terms such as “dominant position” and “abuse of dominance” were too vague to provide workable standards.[59] Although the New York Senate has passed versions of the bill in each session since its introduction, the Assembly has not taken it up.[60] That repeated inaction reflects sustained resistance to an analytically weak framework.

AB 1776 avoids the “abuse of dominance” label but reaches a similar result through different means: lower thresholds, fewer evidentiary requirements, and a mandate to maximize deterrence. The New York experience suggests a broader pattern. Proposals that closely track federal doctrine tend to survive scrutiny; those that depart most sharply from it tend to stall.

C.      Lessons from Europe

Although AB 1776 is not modeled on the European Union’s Digital Markets Act (DMA),[61] the two share underlying assumptions that make the DMA’s experience relevant. The CLRC declined to adopt a formal European “abuse of dominance” standard,[62] but AB 1776 incorporates similar logic through a different route. It mirrors the European approach in three respects: it removes market-power thresholds as a prerequisite for liability, prohibits cross-market balancing, and departs from the effects-based, consumer-welfare framework that has long distinguished U.S. antitrust law. ICLE has cautioned against this “Europeanization” of California law, urging continued alignment with consumer welfare, effects-based analysis, and error-cost discipline.[63]

The DMA provides the most prominent example of this approach in practice. In force since March 2024, it authorizes the European Commission to designate firms as “gatekeepers” based on size, user base, and the provision of “core platform services.”[64] The Commission has designated firms such as Alphabet, Amazon, Apple, ByteDance, Meta, Microsoft, and Booking.com, and imposed obligations related to interoperability, data portability, and self-preferencing.[65]

These obligations apply without requiring case-specific proof of market power, consumer harm, or the absence of efficiencies.[66] The DMA’s self-preferencing rules rest on a contested premise—that integration by a platform is inherently suspect—rather than recognizing that integration often generates efficiencies.[67] AB 1776 adopts a similar shortcut by making market power, recoupment, cross-market balancing, and as-efficient-competitor analysis optional. Both regimes relax traditional screens and presume harm without requiring case-specific evidence.

Early evidence from the DMA’s implementation illustrates the risks of that approach. A 2025 survey by the European Centre for International Political Economy (ECIPE), a Brussels-based policy institute, found that while most consumers support intervention in digital markets, 39% report needing more steps to complete tasks that were previously simple, and roughly one-third report more fragmented and confusing digital experiences.[68] A separate analysis by Carmelo Cennamo and co-authors estimates that DMA provisions could reduce revenues across EU service sectors by up to €114 billion due to lost efficiencies.[69] Sector-specific losses range from €4.4 billion to €59 billion in retail and €14 billion to €21 billion in accommodation. Neither study finds evidence of a meaningful shift in market structure toward alternative providers. Interoperability and unbundling mandates have reduced product quality, while the competitive landscape remains largely unchanged.

These outcomes reflect a broader concern. As ICLE has documented, the stated goals of digital competition policy—greater “contestability” and “fairness”—often diverge from their practical effect: redistributing value from successful firms to less-efficient rivals.[70] The mechanisms differ—conduct mandates in the DMA, expanded tort liability in AB 1776—but the underlying approach is similar: skepticism of market-power thresholds, rejection of efficiency balancing, and a willingness to second-guess business decisions without proof of consumer harm. The DMA’s record offers the best available evidence of how that approach operates in practice. California legislators should weigh that experience before adopting a framework that has degraded digital services for consumers without achieving its structural goals.

ICLE’s comments also point to broader macroeconomic patterns.[71] As the comments note, ECIPE research shows that U.S. GDP per capita exceeded the EU’s by 47% in 2010 and by 82% in 2021, a widening gap under differing regulatory approaches. Even studies often cited to support concerns about concentration cut the other way on closer review. For example, David Autor and his co-authors observe:

An alternative perspective on the rise of [large firms and increased concentration] is that they reflect a diminution of competition, due to weaker U.S. antitrust enforcement. Our findings on the similarity of trends in the United States and Europe, where antitrust authorities have acted more aggressively on large firms, combined with the fact that the concentrating sectors appear to be growing more productive and innovative, suggests that this is unlikely to be the primary explanation, although it may be important in some industries.[72]

XII.  Criminal Liability and Due Process Concerns

AB 1776 raises potential concerns for criminal enforcement under the Cartwright Act. A core principle of American law is the presumption of innocence: prosecutors bear the burden of proving guilt beyond a reasonable doubt. As William Blackstone observed, “the law holds that it is better that ten guilty persons escape than that one innocent suffer.”[73] This principle reflects a broader tradeoff. Efforts to deter harmful conduct must be balanced against the risk of deterring lawful conduct, especially in areas of uncertainty.

Several doctrines operationalize that balance. The void-for-vagueness doctrine requires that criminal laws give clear notice of prohibited conduct. The rule of lenity requires courts to interpret ambiguous penal statutes narrowly in favor of defendants. Together, these constraints limit the scope of criminal liability and protect against overreach.

AB 1776’s interpretive mandates complicate this framework. Section 16733 directs courts to construe antitrust law liberally and to remain “mindful” that California favors “maximizing” deterrence. That instruction does not distinguish between civil and criminal cases. The Cartwright Act, however, carries criminal penalties—up to three years in county jail for individuals and fines up to $6 million for corporations[74] —and it is unclear whether those penalties extend to the new single-firm conduct provisions. This issue is not theoretical. In 2024, the California attorney general announced plans to resume criminal antitrust prosecutions for the first time in 25 years.[75]

In civil cases, broad construction and deterrence mandates may be aggressive but permissible. In criminal cases, they conflict with foundational constraints: proof beyond a reasonable doubt, the rule of lenity, and the requirement of fair notice. Clear notice is easier to establish when conduct is categorically unlawful. A cartel agreement among gas stations to fix prices presents little ambiguity. By contrast, whether below-cost pricing—such as Costco’s $1.50 hot dog and soda combo—violates antitrust law depends on a complex, fact-specific inquiry. Single-firm conduct typically falls under the rule of reason, not per se illegality, making criminal enforcement more difficult to justify.

Federal practice reflects this distinction. Criminal enforcement under the Sherman Act has focused on Section 1 collusion—price-fixing and bid-rigging—where conduct is per se unlawful.[76] Proposals to extend criminal liability to Section 2 monopolization have faced resistance, based on concerns about fair notice and the difficulty of proving specific intent to harm competition.[77]

AB 1776 heightens these concerns. Its instruction to construe the statute broadly operates in tension with lenity, while its open-ended structure—making federal standards optional without specifying what suffices for liability—raises vagueness issues. A prosecutor could argue that “liberal interpretation” defines the elements of the offense, with the reasonable-doubt standard applying only to proof of those elements. That approach would expand criminal liability in practice while formally preserving the burden of proof.

The result is uncertainty about what conduct is criminal. If California courts depart from federal antitrust doctrine without clear replacement standards, firms and individuals may lack fair notice of the law’s boundaries. AB 1776 does not address this tension, and it remains unclear whether its deterrence mandate can be reconciled with the constitutional limits that govern criminal punishment.

XIII.     Conclusion and Recommendations

Assembly Bill 1776 adopts the California Law Revision Commission’s recommendations with minimal change. Those recommendations abandon the error-cost framework and the consumer welfare standard that have disciplined antitrust enforcement for decades. The bill makes optional each major evidentiary screen federal courts use to distinguish procompetitive from anticompetitive conduct—market-power thresholds, recoupment in predatory pricing, cross-market balancing for multi-sided platforms, the as-efficient-competitor test, and quantitative evidence of harm. In their place, it directs courts to maximize deterrence and construe liability broadly, a combination that predictably increases costly false positives. Markets can partially self-correct false negatives as entry erodes supracompetitive profits. False positives do not self-correct, because erroneous condemnation deters beneficial conduct going forward.

The consequences follow directly from these choices. Eliminating recoupment returns California predatory-pricing law to the pre-Brooke Group regime that federal courts rejected for overdeterring price competition. Prohibiting cross-market balancing mandates one-sided analysis of multi-sided platforms—counting costs without crediting offsetting benefits—that will mismeasure welfare for firms such as Google, Meta, Apple, Amazon, and the payment networks at issue in Amex. Extending liability to labor-market monopsony codifies an unsettled theory under an open-ended, deterrence-maximizing standard. Replacing the consumer welfare standard with a multi-objective declaration protecting “all trade participants” and broader social interests deprives courts of a coherent limiting principle.

The bill also creates strong incentives for strategic litigation. Broad liability, low analytical thresholds, treble damages, and a private right of action invite suits aimed at extracting settlements through defense costs. The likely results—higher prices, reduced service quality, or constrained business practices—benefit litigating rivals while harming consumers.

AB 1776’s reach extends beyond California. Firms serving California through digital channels cannot feasibly maintain California-specific business practices. As a result, the bill’s lower standards would set de facto national rules. That extraterritorial effect raises unresolved dormant Commerce Clause questions the Legislature has not addressed.

The interaction of AB 1776’s interpretive mandates with the Cartwright Act’s criminal penalties presents additional due-process concerns. An instruction to construe liability broadly and maximize deterrence conflicts with the rule of lenity and heightens vagueness risks, particularly if applied to single-firm conduct. With the California attorney general signaling a return to criminal antitrust enforcement, these concerns are immediate.

Comparative evidence reinforces these risks. The European Union’s Digital Markets Act—built on similar premises—has, after roughly two years, reduced the seamlessness of digital services for many users without producing measurable increases in competition. New York’s repeated failure to enact a structurally similar bill suggests that proposals departing sharply from federal doctrine face sustained scrutiny.

If the Legislature proceeds with single-firm conduct reforms, it should adopt the following changes:

• Calibrate deterrence. Replace the mandate to maximize deterrence with a directive to calibrate enforcement to identified harms, consistent with the error-cost framework and due-process constraints.
• Restore recoupment. Require proof of recoupment in predatory-pricing cases, consistent with Brooke Group, to preserve the distinction between competition and predation.
• Permit cross-market balancing. Allow courts to assess harms and benefits across both sides of multi-sided platforms, consistent with Amex.
• Require market-power thresholds. Tie liability to demonstrable market power, especially for labor-market claims where measurement remains unsettled.
• Retain the consumer welfare standard. Use consumer welfare as the organizing principle to provide a clear limiting rule.
• Address extraterritorial effects. Evaluate dormant Commerce Clause implications and limit the statute to conduct with substantial effects in California.
• Commission independent economic analysis. Assess the bill’s macroeconomic effects on businesses, workers, and consumers, including nationwide spillovers.
• Exclude single-firm conduct from criminal liability. Reserve criminal enforcement for per se unlawful collusion that provides clear notice.

AB 1776 is not a marginal adjustment. It represents a comprehensive departure from established antitrust doctrine. Any reform should preserve the analytical tools that distinguish harmful conduct from competition on the merits and should align enforcement with both economic evidence and constitutional constraints.

[1] Cal. Assemb. B. 1776, 2025–2026 Reg. Sess. (Cal. 2025).

[2] Cal. Bus. & Prof. Code §§ 16700–16770.

[3] State enforcers may sue under all three statutes. Private plaintiffs may sue under the Sherman and Clayton Acts, see 15 U.S.C. §§ 15, 26, but not under the FTC Act, which only the Federal Trade Commission may enforce. State attorneys general may bring parens patriae actions for Sherman Act violations under § 4C of the Clayton Act, 15 U.S.C. § 15c. Although the Clayton Act often targets concerted conduct—most notably merger review under § 7—it also reaches certain unilateral conduct. The Robinson-Patman Act, codified as an amendment to the Clayton Act, 15 U.S.C. § 13, governs some forms of unilateral price discrimination. Whatever one thinks of Robinson-Patman as a policy matter, it remains relevant to assessing whether any gap exists in the law governing single-firm conduct that A.B. 1776 would fill.

[4] 15 U.S.C. § 2.

[5] See Note, Antitrust Federalism, Preemption, and Judge-Made Law, 133 Harv. L. Rev. 2557 (2020).

[6] Thomas Greene, Robert C. Fellmeth, Thomas A. Papageorge & Kathleen J. Tuttle, A Century of Government Antitrust Enforcement Under the Cartwright Act, 17 Antitrust & Unfair Comp. L. Sec. St. B. Cal. 173 (2008).

[7] Cianci v. Superior Court, 40 Cal. 3d 903, 920 (Cal. 1985).

[8] See Asahi Kasei Pharma Corp. v. Cotherix, Inc., 204 Cal. App. 4th 1, 8 (2012); Flagship Theaters of Palm Desert LLC v. Century Theaters, Inc., 198 Cal. App. 4th 1366, 1386 (2011); Freehand Corp. v. Adobe Sys. Inc., 852 F. Supp. 2d 1171, 1185 (N.D. Cal. 2012).

[9] See Cal. Law Revision Comm’n, Tentative Recommendation: Antitrust Law: Single Firm Conduct (Dec. 2025), https://clrc.ca.gov/pub/Misc-Report/TR-B750.pdf (hereinafter CLRC Recommendation), at 4 (“At a minimum, adopting state law would allow such matters to proceed under state law, even if litigated in federal court.”), 7 (“Vertical integration in major California industries, along with the scale of certain digital platforms, presents competitive challenges not anticipated by the original antitrust drafters.”), 15 n.122 (“Since enactment of the Class Action Fairness Act of 2005, most consumer Cartwright Act class actions have proceeded in federal court. This shift has produced several side effects, including a tendency among federal judges to conflate Cartwright claims with federal antitrust claims, treating them as effectively identical even when they are not.”).

[10] CLRC Recommendation, supra note 9, at 13.

[11] CLRC Recommendation, supra note 9, at 15.

[12] Geoffrey A. Manne, Dirk Auer, Brian Albrecht & Lazar Radic, Int’l Ctr. for L. & Econ., Comments on Memorandum 2025-21 on the Draft Language for Single-Firm Conduct Provision (May 23, 2025), https://laweconcenter.org/wp-content/uploads/2025/05/CLRC-Comments.pdf (hereinafter ICLE 2025 CLRC Comments).

[13] CLRC Recommendation, supra note 9.

[14] Frank H. Easterbrook, The Limits of Antitrust, 63 Tex. L. Rev. 1, 2–4 (1984).

^[15] False negatives may occasionally produce precedent, but they more often arise without enforcement and thus without comparable precedential effect.

[16] ICLE 2025 CLRC Comments, supra note 12; Geoffrey A. Manne & Dirk Auer, Int’l Ctr. for L. & Econ., Against the “Europeanization” of California’s Antitrust Law: Comments of the International Center for Law & Economics on the Single-Firm Conduct Expert Report (2024), https://laweconcenter.org/wp-content/uploads/2024/05/Comments-of-the-International-Center-for-Law-California-Law-Revision-Commission-Single-Firm-Conduct.pdf; Geoffrey A. Manne, Dirk Auer & Brian Albrecht, California’s Ill-Advised Turn Toward Europeanized Theories of Harm for Single-Firm Conduct, CPI Antitrust Chron. (June 2025), https://www.pymnts.com/cpi-posts/californias-ill-advised-turn-toward-europeanized-theories-of-harm-for-single-firm-conduct.

[17] CLRC Recommendation, supra note 9, at 19 (“Indicators of anticompetitive intent vary by circumstance, and rigid rules that demand specific fact patterns can unduly restrict enforcement.”).

[18] CLRC Recommendation, supra note 9, at 9, 11.

[19] ICLE 2024 CLRC Comments, supra note 16.

[20] Brown Shoe Co. v. United States, 370 U.S. 294, 320 (1962) (emphasis in original).

^[21] Price reductions—always valuable to consumers—often harm less efficient competitors, which may be unable to match the better deal.

[22] Geoffrey A. Manne, Int’l Ctr. for L. & Econ., Understanding Concentration and Competition: Implications for California’s Antitrust Policy, Presentation Before the Cal. Law Revision Comm’n (Aug. 15, 2024), https://laweconcenter.org/wp-content/uploads/2025/05/Manne-CLRC-presentation-2024-08-15.pdf.

[23] Gerard Hoberg & Gordon M. Phillips, Scope, Scale and Concentration: The 21st Century Firm (Nat’l Bureau of Econ. Rsch., Working Paper No. 30672, 2022), https://www.nber.org/system/files/working_papers/w30672/w30672.pdf.

[24] C. Lanier Benkard, Ali Yurukoglu & Anthony Lee Zhang, Concentration in Product Markets (Nat’l Bureau of Econ. Rsch., Working Paper No. 28745, 2021), https://www.nber.org/system/files/working_papers/w28745/w28745.pdf.

[25] Manne, supra note 22.

[26] Reiter v. Sonotone Corp., 442 U.S. 330, 343 (1979).

[27] Cal. Assemb. B. 1776, § 16730 (Cal. 2025).

[28] ICLE 2025 CLRC Comments, supra note 12; Manne, Auer & Albrecht, supra note 16.

[29] Brooke Group Ltd. v. Brown & Williamson Tobacco Corp., 509 U.S. 209 (1993).

[30] Id. at 224.

[31] Barry Wright Corp. v. ITT Grinnell Corp., 724 F.2d 227, 234 (1st Cir. 1983).

[32] ICLE 2025 CLRC Comments, supra note 12; Manne, Auer & Albrecht, supra note 16.

[33] U.S. Dep’t of Justice, Competition and Monopoly: Single-Firm Conduct Under Section 2 of the Sherman Act 51 (2008), https://www.justice.gov/sites/default/files/atr/legacy/2009/05/11/236681.pdf.

[34] Ohio v. Am. Express Co., 585 U.S. 529 (2018).

[35] Id. at 2278.

[36] Id. at 2278, 2287.

[37] Herbert Hovenkamp, Antitrust and Platform Monopoly, 130 Yale L.J. 1952 (2021).

[38] ICLE 2025 CLRC Comments, supra note 12; Manne, Auer & Albrecht, supra note 16.

[39] CLRC Recommendation, supra note 9, at 20.

[40] ICLE 2025 CLRC Comments, supra note 12.

[41] ICLE 2024 CLRC Comments, supra note 16.

[42] Francine Lafontaine & Margaret Slade, Exclusive Contracts and Vertical Restraints: Empirical Evidence and Public Policy, in Handbook of Antitrust Economics 391 (Paolo Buccirossi ed., 2008).

[43] ICLE 2024 CLRC Comments, supra note 16, citing James C. Cooper, Luke M. Froeb, Dan O’Brien & Michael G. Vita, Vertical Antitrust Policy as a Problem of Inference, 23 Int’l J. Indus. Org. 639 (2005) (surveying the empirical literature and concluding that, although “some studies find evidence consistent with both pro- and anticompetitive effects . . . virtually no studies can claim to have identified instances where vertical practices were likely to have harmed competition”); James Cooper, Luke Froeb, Daniel O’Brien & Michael Vita, Vertical Restrictions and Antitrust Policy: What About the Evidence?, Comp. Pol’y Int’l 45 (2005) (finding that “some studies find evidence consistent with both pro- and anticompetitive effects . . . virtually no studies can claim to have identified instances where vertical practices were likely to have harmed competition”).

[44] Cal. Assemb. B. 1776, §§ 16730(b), 16731(a)(2) (Cal. 2025).

[45] Geoffrey A. Manne, Brian C. Albrecht & Dirk Auer, Labor Monopsony and Antitrust Enforcement: A Distorting Mirror, 74 DePaul L. Rev. 1119 (2025).

[46] See ICLE 2025 CLRC Comments, supra note 12; Manne, Auer & Albrecht, supra note 16.

[47] See ICLE 2025 CLRC Comments, supra note 12; Manne, Auer & Albrecht, supra note 16.

[48] Zoë Cullen, Bobak Pakzad-Hurson & Ricardo Perez-Truglia, Home Sweet Home: How Much Do Employees Value Remote Work?, 115 AEA Papers & Proc. 276 (2025).

[49] See Michael J. Nader, Managing a California Remote Work Policy: Determining Which Laws Apply, Ogletree Deakins: Insights (Apr. 1, 2022), https://ogletree.com/insights-resources/blog-posts/managing-a-california-remote-work-policy-determining-which-laws-apply; Justworks, HR Compliance for Remote-First Companies: Managing HR Compliance Across Remote Teams (2026), https://assets.ctfassets.net/mnc2gcng0j8q/4cMqUT8G5XIzKjNVqXaDEb/6b07b29650b0dcda8f47e8e7f9967bd4/HR_Compliance_for_Remote-First_Companies_-_Guide.pdf (noting that “[s]ome states may create excessive compliance burdens or tax exposure,” leading “companies to exclude them from approved remote work locations”).

[50] Reiter, supra note 26.

[51] Brunswick Corp. v. Pueblo Bowl-O-Mat, Inc., 429 U.S. 477, 489 (1977) (“Plaintiffs must prove antitrust injury—i.e., injury of the type the antitrust laws were intended to prevent and that flows from what makes the defendant’s conduct unlawful. The injury must reflect the anticompetitive effect of the violation or of conduct the violation made possible.”) (emphasis in original).

[52] Murat C. Mungan & John M. Yun, A Reputational View of Antitrust’s Consumer Welfare Standard, 61 Hous. L. Rev. 569 (2024), https://scholarship.law.tamu.edu/cgi/viewcontent.cgi?article=2962&context=facscholar.

[53] United States v. Grinnell Corp., 384 U.S. 563, 571 (1966) (“We defined monopoly power as ‘the power to control prices or exclude competition.’ Such power may ordinarily be inferred from a predominant market share.”).

[54] Parker v. Brown, 317 U.S. 341 (1943); see also California v. ARC Am. Corp., 490 U.S. 93 (1989).

[55] Pike v. Bruce Church, Inc., 397 U.S. 137, 142 (1970).

[56] Shira Liu, Dormant Commerce Clause: A Potential Brake on State Antitrust Legislation, 33 Competition: J. Antitrust, UCL & Privacy Section 1 (2023) (observing that constitutional uncertainty persists because courts must apply the “notoriously unclear” Pike balancing test to assess whether the nationwide burdens of expansive state antitrust laws—such as those governing mergers or diverging from federal standards—clearly exceed their local benefits).

[57] S. 8700-A, 2019–2020 Reg. Sess. (N.Y. 2019), https://www.nysenate.gov/legislation/bills/2019/S8700.

[58] See, e.g., S. 933-C, 2021–2022 Reg. Sess. (N.Y. 2021), https://www.nysenate.gov/legislation/bills/2021/S933; S. 6748, 2023–2024 Reg. Sess. (N.Y. 2023), https://www.nysenate.gov/legislation/bills/2023/S6748; S. 335, 2025–2026 Reg. Sess. (N.Y. 2025), https://www.nysenate.gov/legislation/bills/2025/S335.

[59] The Bus. Council, S. 6748-B (Gianaris) / A. 10323 (Peoples-Stokes) (May 20, 2024), https://www.bcnys.org/memo/s6748-b-gianaris-a10323-peoples-stokes (reporting that the CLRC’s Single-Firm Conduct Working Group “outright rejected the Twenty First Century Anti-Trust Act and its proposed adoption of vague and undefined legal standards”).

[60] Jared P. Nagley & Helen Cho Eckert, Amending New York’s Donnelly Act: If at First You Don’t Succeed, Try, Try, and Try Again, Nat’l L. Rev. (Jan. 31, 2025), https://natlawreview.com/article/amending-new-yorks-donnelly-act-if-first-you-dont-succeed-try-try-and-try-again.

[61] Regulation (EU) 2022/1925 of the European Parliament and of the Council of 14 September 2022 on Contestable and Fair Markets in the Digital Sector and Amending Directives (EU) 2019/1937 and (EU) 2020/1828, 2022 O.J. (L 265) 1, https://eur-lex.europa.eu/legal-content/en/TXT/?uri=CELEX%3A32022R1925 (hereinafter Digital Markets Act).

[62] CLRC Recommendation, supra note 9, at 9 (“The Commission considered creating a distinct single-firm conduct framework for firms with significant market power, drawing on EU law prohibiting ‘any abuse by one or more undertakings of a dominant position within the internal market or a substantial part of it.’ The Commission ultimately declined to adopt such an approach, citing concerns about the vagueness and arbitrariness of defining thresholds for substantial market power, the use of differing conduct standards, and the failure of similar efforts in the United States.”).

[63] ICLE 2025 CLRC Comments, supra note 12, at 2–4; Manne, Auer & Albrecht, supra note 16.

[64] Digital Markets Act, supra note 61, arts. 2(1), 3(1)–(2).

[65] Eur. Comm’n, Gatekeepers Under the Digital Markets Act, https://digital-markets-act.ec.europa.eu/gatekeepers_en (last visited Apr. 22, 2026).

[66] Digital Markets Act, supra note 61, arts. 5–7 (establishing ex ante obligations for designated gatekeepers without requiring case-by-case effects analysis).

[67] Giuseppe Colangelo, Antitrust Unchained: The EU’s Case Against Self-Preferencing, 72 GRUR Int’l 538, 542–46 (2023); Lazar Radic, Opening the Walled Garden: Global Regulation and the Unbundling of Apple’s Ecosystem, Truth on the Mkt. (Mar. 19, 2026), https://truthonthemarket.com/2026/03/19/opening-the-walled-garden-global-regulation-and-the-unbundling-of-apples-ecosystem (“The MSCA—like the EU’s Digital Markets Act (DMA)—takes a different approach. It restricts forms of integration precisely where platforms often integrate for efficiency . . . These interventions assume, rather than demonstrate, that platform control is more likely to suppress than enhance competition . . . That assumption sits uneasily with the empirical literature on digital platforms”).

[68] Eur. Ctr. for Int’l Pol. Econ., What About Us? Consumer Response to the Digital Markets Act, Occasional Paper No. 10/2025, at 3 (2025), https://ecipe.org/publications/consumer-response-to-the-digital-markets-act.

[69] Carmelo Cennamo et al., Economic Impact of the Digital Markets Act on European Businesses and the European Economy 5 (CCIA Eur., June 2025), https://www.dmcforum.net/publications/economic-impact-of-the-digital-markets-act-on-european-businesses-and-the-european-economy.

[70] Lazar Radic, Geoffrey A. Manne & Dirk Auer, Regulate for What? A Closer Look at the Rationale and Goals of Digital Competition Regulations, 22 Berkeley Bus. L.J. 201 (2025), https://lawcat.berkeley.edu/record/1312409?v=pdf.

[71] ICLE 2024 CLRC Comments, supra note 16.

[72] David Autor, David Dorn, Lawrence F. Katz, Christina Patterson & John Van Reenen, The Fall of the Labor Share and the Rise of Superstar Firms, 135 Q.J. Econ. 645, 651 (2020) (citations omitted; emphasis added).

[73] 2 William Blackstone, Commentaries on the Laws of England 358 (George Sharswood ed., 1893).

[74] See S. 763, 2025–2026 Reg. Sess. (Cal. 2025).

[75] See, e.g., Niall E. Lynch & Sydney Kirlan-Stout, Criminal Antitrust Enforcement by the California Attorney General: What Can We Expect?, CPI Columns Cartel (Aug. 2024), https://www.pymnts.com/wp-content/uploads/2024/08/Cartel-Column-August-2024-Full.pdf.

[76] See Joseph Matelis & Daniel Richardson, Criminal Enforcement of Section 2 of the Sherman Act, 36 Antitrust 61, 65 (2022) (“Significantly, no grand jury appears to have returned a criminal Section 2 indictment in more than four decades, and nearly 50 years have passed since a grand jury returned an indictment based solely on a Section 2 charge.”); ABA Section of Antitrust L., Comments in Response to the Antitrust Modernization Commission’s Request for Public Comment on Criminal Penalties 5 (Nov. 14, 2005), https://govinfo.library.unt.edu/amc/public_studies_fr28902/criminal_pdf/051114_ABA_Criminal_Remedies.pdf (“For generations, the Antitrust Division has limited criminal enforcement to hard-core cartel conduct. While not an absolute guarantee against prosecuting other conduct, this practice reflects a long-standing, near-universal consensus that only such conduct warrants criminal prosecution.”).

[77] See Matelis & Richardson, supra note 76, at 66–68.

Executive Summary

American freight railroads are safer today than at any point in their history. Accident rates, hazardous-material releases, and employee fatalities have all declined over the past two decades, driven largely by private investment and innovation.

The Railway Safety Act, introduced after the 2023 East Palestine derailment, would impose prescriptive mandates, including crew-size requirements, federal detector standards, expanded inspections, and accelerated tank-car replacement timelines. These measures assume additional regulation will improve safety.

This issue brief argues the Act is unlikely to do so. It mandates sweeping rulemakings without cost-benefit analysis, relies on rigid requirements that risk locking in current technology, and includes provisions—especially the two-person crew mandate—that lack a clear evidentiary basis. It also expands regulation in ways that dilute focus on the highest-risk operations and may exceed the industry’s capacity to comply.

The costs are substantial. Compliance could reach billions of dollars, while the scale of accident reduction needed to justify those costs is implausible given current trends. Law & economics research further shows that regulatory accumulation in freight transportation raises costs, reduces output, and weakens innovation-driven growth.

A better approach is performance-based regulation: set measurable safety targets and allow firms flexibility in how to achieve them. Policymakers should preserve voluntary industry initiatives and support smaller carriers through targeted funding.

Freight rail safety is improving. The priority should be to sustain that progress, not impose mandates that risk undermining it.

I.          Introduction: A Safer Industry Meets Rising Regulatory Pressure

American freight railroads are safer today than at any point in their history. Over the past two decades, the train accident rate per million train-miles has fallen by more than 38%,[1] h hazardous-materials train accident rates have declined by at least 61%,[2] and employee on-duty fatalities reached an all-time low of seven in 2023.[3] Bureau of Labor Statistics injury-rate data place rail transportation squarely in the middle of the American industrial landscape—comparable to general freight trucking and retail trade, and well below air transportation, warehousing, and urban transit.[4] Freight rail remains one of the safest modes of surface freight transport in the United States, and the gap continues to widen.

Political demand for sweeping new railroad safety mandates has nevertheless surged. The February 2023 derailment of Norfolk Southern Train 32N in East Palestine, Ohio—caused by an overheated wheel bearing[5] —pushed rail safety to the center of public debate. Congress responded by introducing and reintroducing the Railway Safety Act across two successive sessions.[6]

This issue brief argues that the proposed Railway Safety Act, despite its rhetorical appeal, suffers from three fundamental deficiencies. First, it mandates sweeping federal rulemakings without requiring cost-benefit analysis, in tension with Executive Order 14192’s deregulatory framework.[7] Second, its prescriptive mandates risk locking in current technology and discouraging the innovation that has driven the industry’s safety gains. Third, several prominent provisions—particularly the two-person crew mandate—appear driven less by safety evidence than by political economy concerns. A better approach would recognize the industry’s substantial safety gains and avoid layering on regulations that do not address identifiable problems. Short of that, policymakers should adopt performance-based regulation that sets measurable safety targets while preserving flexibility to innovate.

The Railway Safety Act also offers a useful case study in how political-economy dynamics can produce net-harmful regulation. A high-profile incident generated intense public demand for action. Concentrated interest groups then shaped the legislative response, while policymakers largely overlooked a factual record of improving safety and ongoing innovation. The result is a bill that would likely impose billions in compliance costs without demonstrated safety benefits, even as the industry’s baseline trajectory remains positive.

This pattern is not unique to rail policy. It reflects a broader regulatory tendency: policymakers confronting complex, technical systems often focus on prescriptive solutions crafted under political pressure, rather than evaluating how those mandates may displace more effective, innovation-preserving alternatives. Error-cost analysis and regulatory forbearance provide a more reliable framework for achieving public-interest goals without foreclosing beneficial market adaptation.

Recent academic work underscores the stakes. Bentley Coffey, Patrick McLaughlin, and Pietro Peretto show in a multi-industry endogenous-growth model that per-unit transportation costs constrain innovation-driven growth; policies that raise those costs—including cumulative regulation—can dampen the translation of manufacturing productivity gains into aggregate output.[8] Their companion empirical paper finds that a 5% increase in federal regulatory restrictions on a freight transportation mode raises unit costs by 0.8–2.3% and reduces quantities shipped by 1.4–4.1%, with effects that persist and compound over time.[9] In this context, the Railway Safety Act is likely to do more harm than good.

II.          The Railway Safety Act: Structure, Mandates, and Tensions

The Railway Safety Act first appeared as S. 576 in March 2023, introduced by Sen. Sherrod Brown (D-Ohio) in the immediate aftermath of the East Palestine derailment. The Senate Commerce Committee approved the bill in May 2023, but it never reached the floor during the 118th Congress.[10] Lawmakers reintroduced it in the House in February 2025 (H.R. 928) and again in the Senate on Feb. 24, 2026 (S. 3903), led by Sen. Jon Husted (R-Ohio) and cosponsored by Sens. Maria Cantwell (D-Wash.), Roger Marshall (R-Kan.), Eric Schmitt (R-Mo.), Tammy Baldwin (D-Wis.), Amy Klobuchar (D-Minn.), Bernie Moreno (R-Ohio), and John Fetterman (D-Pa.).[11]

The bill’s core provisions include a mandatory two-person crew for all Class I freight trains; a federal framework governing the spacing, maintenance, sensitivity, and data protocols for wayside defect-detection systems; an expanded definition of “high-hazard trains” (HHT) using a five-car threshold for flammable gases; accelerated tank-car retrofit and phaseout timelines; expanded pre-departure and locomotive inspection requirements; state notification requirements for hazardous-material movements; and higher civil penalties.[12]

These mandates sit uneasily alongside the current administration’s deregulatory framework. Executive Order 14192, issued Jan. 31, 2025, directs agencies to identify at least 10 existing regulations for repeal for every new rule they propose.[13] For fiscal year 2025, the order further requires that the total incremental cost of all new regulations be “significantly less than zero.”[14]

An executive order does not bind Congress, but it highlights a clear policy divide. The administration has articulated a pro-growth, deregulatory agenda, while some congressional constituencies continue to favor prescriptive mandates. That tension raises a central question: whether additional command-and-control safety rules make sense when, as discussed infra, the industry’s safety record continues to improve under the current framework.

The Railway Safety Act would require the U.S. Department of Transportation (DOT) and the Federal Railroad Administration (FRA) to issue a series of new rules with significant compliance costs. This is precisely the type of regulatory expansion Executive Order 14192 seeks to prevent. The interaction between the two is not merely theoretical. If the FRA attempted to comply with both simultaneously, it would need to repeal a substantial number of existing rules and demonstrate a net reduction in regulatory costs—an implausible outcome given the scope of the Act’s mandates.

Much of the political momentum behind the Railway Safety Act rests on the perception that train derailments are a growing crisis. A commonly cited comparison notes more than 1,300 U.S. derailments in 2019, compared with just 73 in the European Union, even though EU rail traffic spans more than five times as many rail-kilometers.[15] That comparison obscures more than it reveals.

First, roughly 60% of U.S. derailments occur in rail yards at very low speeds—the functional equivalent of parking-lot fender benders.[16] Mainline derailments, which pose the greatest risk to public safety, remain at historic lows. Second, the EU uses different reporting methodologies, emphasizing rates per billion train-kilometers and often excluding comparable low-severity yard incidents. Third, the underlying systems differ fundamentally. The U.S. network is freight-dominant and moves far more tonnage per rail-kilometer, while European systems are overwhelmingly passenger-oriented. Direct comparisons between the two are not meaningful.

The long-term trend is clear. Federal Railroad Administration data show equipment-caused train accidents have declined 38% since 2005,[17] and axle- and bearing-related accidents have fallen roughly 81% since the early 1980s. The agency attributes much of this progress to the widespread adoption of wayside hot-bearing detectors and related monitoring technologies.[18] These gains have occurred under the existing regulatory framework. Any policy change should therefore account for how new mandates might disrupt the innovation that produced them.

III.          Private Incentives and Prescriptive Safety Regulation

Prescriptive safety regulation typically rests on a familiar market-failure story: firms underinvest in safety because they do not fully internalize the social costs of accidents, and government mandates must close the gap. The Railway Safety Act adopts that logic. The freight rail industry’s response after East Palestine, however, complicates the narrative.

Within months of the derailment, Class I railroads deployed roughly 1,000 additional wayside detectors, lowered bearing-temperature alert thresholds, adopted industry-wide predictive analytics, and expanded first-responder access to real-time railcar data—investments that outpaced any plausible federal rulemaking.[19] These actions reflect strong private incentives: liability exposure, reputational capital, network reliability, and the economics of asset preservation. In this environment, the case for additional prescriptive regulation weakens. The proposed mandates would largely codify existing practices, but at higher cost and with less flexibility.

The industry has also committed to developing a shared standard for bearing-trending analysis—an algorithmic approach that identifies degradation patterns before absolute temperature thresholds trigger alerts.[20] This trajectory suggests that the Act does not correct a market failure. It instead layers compliance costs onto an already-functioning incentive structure, with effects that extend beyond the industry’s balance sheet.

The Minnesota Wayside Detector System Study, a legislatively mandated assessment completed in March 2026, reinforces this point.[21] The study surveys roughly five wayside-detection technologies deployed on Minnesota’s rail system and 11 additional detector types in use across North America, including hot-bearing detectors, wheel-impact load detectors, acoustic bearing detectors, dragging-equipment detectors, and machine-vision inspection portals. It also examines emerging systems such as distributed fiber-optic sensing and light detection and ranging (LiDAR)-based infrastructure monitoring.[22]

The study’s cost-benefit analysis proves especially revealing. It models three detector-spacing scenarios—10, 15, and 20 miles—applied statewide to nearly 1,000 miles of Class II and Class III track. In each case, recurring operating costs exceed quantified safety benefits, producing a negative cumulative net impact over a 10-year horizon.[23] Because a state legislature commissioned the study, not the industry, its findings carry particular weight. They undermine the premise that federally mandated detector standards would generate net benefits, especially for smaller carriers.

Federal Railroad Administration inspections point in the same direction. In January 2024, the FRA published its High-Hazard Flammable Train Route Assessment, summarizing inspections of more than 2,600 wayside detectors across 28 railroads. Inspectors found that railroads generally monitored detector performance closely, though they identified issues at roughly 120 sites, including calibration drift and inverted transducers.[24] The broader lesson is straightforward: voluntary industry standards often precede federal regulation, not the reverse, because firms can adapt more quickly than regulators.

IV.          Structural and Analytical Defects in the Railway Safety Act

Even setting aside the political-economy dynamics discussed above, the Railway Safety Act suffers from a set of analytical and structural defects that warrant separate attention. Its core provisions impose prescriptive, uniform mandates rather than risk-weighted, performance-based approaches that an error-cost framework would favor. The bill also omits the cost-benefit discipline that typically constrains major rulemakings. Several key terms—from the five-car threshold in the expanded high-hazard-train definition to fixed detector-spacing and two-person-crew requirements—appear to reflect negotiated compromises rather than empirical evidence about where risk actually concentrates.

These design choices have concrete consequences. They direct safety capital toward specifications likely to become obsolete before regulators finalize implementing rules. They shield incumbent detection technologies from competition by next-generation alternatives. They also impose fixed compliance costs on smaller carriers that do not track marginal risk on their networks. The result is a regulatory regime that treats a dynamic engineering problem as a static checklist—one that predictably produces less safety, less innovation, and more concentrated harm among the least well-resourced participants than the performance-based alternatives discussed above.

A.             The Absence of Cost-Benefit Constraints

The most fundamental objection to the Act is structural. It mandates sweeping DOT rulemakings without requiring the agency to show that each rule’s benefits justify its costs. This approach departs from the longstanding cost-benefit framework set out in OMB Circular A-4—reinstated in its 2003 form by Executive Order 14192—which requires rigorous analysis for economically significant regulations.[25] More importantly, the Act’s prescriptive mandates would limit the Federal Railroad Administration’s ability to make evidence-based regulatory decisions. Even if the agency’s own analysis shows that a requirement’s costs exceed its safety benefits, the statute would require the agency to proceed.

Available estimates suggest that the compliance burden would run into the billions over a decade. One analysis places the cost of the detector-spacing provision alone between $1.1 billion and $2.2 billion.[26] The 2008 Positive Train Control mandate—a single, comparable requirement—cost U.S. freight railroads roughly $10 billion to $15 billion over its implementation period.[27] The Act would layer multiple additional mandates on top of that baseline, with cumulative costs that could rival or exceed the PTC experience.

To justify expenditures of that magnitude, proponents would need to show that the Act prevents a comparable volume of accidents. Before adopting such costly measures, Congress should at least direct the Congressional Research Service or the FRA to estimate the likely compliance burden.

Empirical benchmarks highlight the gap. An analysis of FRA accident data commissioned by the North Carolina Department of Transportation finds that the average property-damage cost of a rail incident is about $122,000 in 2020 dollars.[28] Even high-severity derailments—those involving roughly 40 or more loaded freight cars—reach only about $2.6 million.[29]

A simple break-even calculation illustrates the mismatch. Assume total compliance costs comparable to Positive Train Control—about $10 billion over a decade. Using the $122,000 average property-damage figure, the Act would need to prevent roughly 80,000 incidents to cover its costs. Even valuing each prevented event at $2.6 million, the Act would still need to prevent nearly 4,000 severe derailments over the same period. Neither scenario is plausible. Mainline derailments on Class I railroads number in the low hundreds each year, and the long-term trend continues to decline. No realistic set of prescriptive mandates could reduce incidents at the scale required to satisfy this cost-benefit threshold.

The empirical literature reinforces these concerns. Recent work finds that regulatory accumulation in freight transportation suppresses investment in productivity-enhancing innovation: a 5% increase in regulatory restrictions raises unit shipping costs by 0.8–2.3% and reduces freight volumes by 1.4–4.1%, with effects that compound over time.[30] These are not one-time adjustment costs; they represent persistent constraints on economic growth. The broader literature reaches similar conclusions. One estimate finds that federal regulation added since 1949 reduced annual U.S. output growth by about one percentage point on average, resulting in a 28% reduction in output by 2005.[31]

B.             Misaligned Risk Targeting in Core Provisions

Expanding the HHT definition to include flammable gas may respond to East Palestine. The Act’s five-car threshold, however—far below the existing standard of 20 or more continuous tank cars or 35 total—rests on no clear risk assessment and would sweep a far larger set of trains into heightened regulation.[32] Existing voluntary “Key Train” practices, codified in Circular OT-55-R, already impose 50-mph speed limits and siding requirements on the highest-risk movements, including those carrying toxic-inhalation-hazard materials, large volumes of hazmat, or spent nuclear fuel.[33] Lowering the threshold risks shifting attention away from genuinely high-risk operations toward routine freight traffic, imposing costs without corresponding safety gains.

The Act’s expanded locomotive and pre-departure inspection requirements reflect a prescriptive, process-oriented approach. They dictate how railroads must inspect equipment rather than what safety outcomes they must achieve. Nothing in the record suggests that more manual inspections would have prevented the East Palestine derailment, which resulted from a bearing failure that developed between automated detector readings, not from an inspection lapse.[34]

C.             Prescriptive Design and Innovation Suppression

The regulatory-design literature has long identified the limits of this approach. Cary Coglianese and David Lazer distinguish among technology-based regulation, which specifies means; performance-based regulation, which specifies ends; and management-based regulation, which requires firms to design their own safety systems to a specified vision.[35] Performance-based and management-based approaches generally outperform prescriptive mandates because they preserve flexibility and encourage efficient solutions. Railroads have already deployed acoustic bearing detectors, machine-vision portals, and trending-analysis algorithms. But the Federal Railroad Administration’s existing prescriptive framework has not adapted to allow these technologies to satisfy regulatory requirements, effectively reducing the return on safety innovation and likely discouraging further investment.

The Act’s provisions establishing federal standards for detector thresholds, spacing, and data protocols pose a particular risk to innovation. The relevant technology remains in flux. The Minnesota study identifies five wayside-detector types currently deployed in the state, 11 additional types in use elsewhere in North America, and a set of emerging technologies—including distributed fiber-optic sensing for landslide and rockfall detection, light detection and ranging (LiDAR)-based infrastructure monitoring, and IoT-enabled remote-condition systems—many still in pilot-project phases.[36] A federal rule written today would codify 2026-era technology. Given the Federal Railroad Administration’s multi-year rulemaking timelines, the resulting standards could become obsolete before they take effect.

Positive Train Control offers a cautionary example. Congress mandated PTC in 2008, locking in a technological specification before the system had fully matured. Implementation ultimately cost the industry roughly $10 billion to $15 billion and required repeated statutory deadline extensions.[37] The mandate imposed substantial costs and delays while foreclosing potentially superior alternatives.[38] The Act’s detector and tank-car provisions risk repeating that pattern.

Coffey, McLaughlin, and Peretto’s growth model explains why this matters beyond railroads. In their framework, transportation is an integral component of production, and per-unit transportation costs set a lower bound on market prices and returns to innovation. When prescriptive mandates raise those costs, they do more than reallocate resources within the transportation sector; they weaken the innovation-growth feedback loop across the broader economy.[39] Technology-freezing regulation in freight rail thus carries macroeconomic consequences.

D.            Tank Car Mandates and Capacity Constraints

Section 110 illustrates similar design flaws. It accelerates the phaseout of older-generation tank cars in certain Class 3 flammable-liquid service, setting a primary deadline of Dec. 31, 2027, with a one-year fallback to Dec. 31, 2028 if the secretary determines that manufacturing or retrofit capacity is insufficient. Section 110(b) directs the secretary to remove or revise any conflicting deadlines immediately, and Section 110(d) requires a Government Accountability Office (GAO) review of manufacturing and retrofit capacity within 18 months of enactment.

This acceleration largely duplicates a transition already in progress. Bureau of Transportation Statistics data show that compliance with DOT-117 standards rose from 56% in 2021 to 59% in 2022, with about 6,914 additional cars built or retrofitted in 2023—a trajectory that would converge on the current statutory deadline of May 1, 2029 without further intervention.[40] At most, the Act advances compliance by roughly 17 months. Section 110(b)’s directive to revise conflicting deadlines also risks creating a regulatory gap by voiding existing schedules before new ones take effect, introducing uncertainty into fleet planning and capital investment.

Capacity constraints compound the problem. The Bureau of Transportation Statistics reports that certified facilities produced or retrofitted about 6,914 tank cars in 2022, reflecting the system’s practical annual throughput.[41] Compressing the deadline to December 2027 would require the industry to process the remaining noncompliant fleet at a pace well above demonstrated capacity. If that compression leads to equipment shortages or service disruptions, it would impose costs on shippers and consumers without clear incremental safety benefits, given that safer tank-car designs are already entering the fleet through market-driven replacement.[42]

E.              The Two-Person Crew Mandate

The Act’s two-person crew mandate presents a similar evidentiary gap. The Federal Railroad Administration concluded in 2019 that it “cannot provide reliable or conclusive statistical data” on whether one-person crews are safer or less safe than multiple-person crews.[43] The agency reversed course in its 2024 final rule without introducing new statistical evidence, relying instead on qualitative judgments about redundancy.[44] The mandate would override collective-bargaining agreements and eliminate the operational flexibility railroads use to tailor staffing to route conditions, traffic density, and available technology. Its primary effect would be to preserve employment levels—a legitimate policy goal in some contexts, but one better addressed through collective bargaining than through safety regulation.

V.          A Performance-Based Alternative

If prescriptive mandates are the wrong tool, what should replace them? The regulatory-design literature and the railroad industry’s recent experience point to a performance-based approach.

First, the Federal Railroad Administration should adopt performance-based regulation. Instead of prescribing detector spacing, crew size, and inspection procedures, the agency should set measurable safety targets—derailments per million train-miles, hazmat release rates, bearing-failure rates—and allow railroads to determine how best to achieve them. Marc Scribner finds the FRA has been “highly prescriptive” and “slow to adopt performance-based alternatives,” and calls for a systematic shift toward outcome-oriented regulation.[45]

First, the Federal Railroad Administration should adopt performance-based regulation. Instead of prescribing detector spacing, crew size, and inspection procedures, the agency should set measurable safety targets—derailments per million train-miles, hazmat release rates, bearing-failure rates—and allow railroads to determine how best to achieve them. Marc Scribner finds the FRA has been “highly prescriptive” and “slow to adopt performance-based alternatives,” and calls for a systematic shift toward outcome-oriented regulation.

Third, smaller carriers need targeted support, not unfunded mandates. The Minnesota study shows that detector mandates would impose net losses on rural short-line railroads. Existing programs—the Consolidated Rail Infrastructure and Safety Improvements (CRISI) Program, Infrastructure for Rebuilding America (INFRA) grants, and state initiatives such as Minnesota’s Short Line Infrastructure Maintenance Tax Credit—can help smaller carriers adopt safety technologies at a pace consistent with their financial capacity.[46]

Finally, the macroeconomic evidence supports regulatory restraint. Coffey, McLaughlin, and Peretto find that regulatory accumulation in freight transportation reduces labor, fuel, and capital productivity across all major freight modes. Their counterfactual simulations show that these costs do not remain within the regulated firms; they propagate through the supply chain, raising prices, reducing shipment volumes, and weakening the innovation-driven growth process. Performance-based regulation that limits prescriptive rule accumulation, preserves investment incentives, and allows railroads to allocate resources toward the highest-return safety technologies would advance both safety and economic growth.

VI.          Conclusion

The Railway Safety Act responds to a real tragedy. But sound regulatory design requires more than urgency and good intentions. The Act imposes sweeping prescriptive mandates without cost-benefit justification, in tension with Executive Order 14192’s deregulatory framework. It targets risks imprecisely, substitutes uniform rules for evidence-based prioritization, and would require the Federal Railroad Administration to proceed with rulemakings even where costs exceed benefits.

These design flaws carry predictable consequences. The Act risks locking in current-generation technology, discouraging the innovation that has driven decades of safety gains, and imposing fixed compliance costs that fall hardest on smaller carriers. It would layer billions of dollars in new mandates onto an industry where accident rates, hazmat releases, and equipment-related failures have all declined over time under the existing framework. Its most visible provision—the two-person crew mandate—lacks a clear evidentiary basis and reflects political compromise more than demonstrated safety need.

The broader lesson extends beyond rail policy. Prescriptive, command-and-control regulation often treats dynamic engineering and operational problems as static compliance exercises. In doing so, it can displace more effective, innovation-preserving approaches while introducing economy-wide costs. As the law & economics literature shows, regulatory accumulation in freight transportation raises per-unit costs, suppresses productivity, and weakens the innovation-growth process that underpins long-run economic performance.

A better approach is available. Performance-based regulation would set clear, measurable safety targets—such as derailments per million train-miles or hazmat release rates—while allowing railroads to determine how best to meet them. Policymakers should preserve the voluntary, industry-led initiatives that have already accelerated safety improvements, continue to rely on the FRA’s collaborative advisory process, and provide targeted support to smaller carriers through existing funding programs. This framework aligns incentives, encourages technological progress, and directs resources toward the highest-return safety investments.

Freight rail safety is improving. The central policy challenge is not to react reflexively to high-profile incidents, but to sustain and accelerate that progress. The Railway Safety Act, as currently structured, risks doing the opposite—imposing large, certain costs in pursuit of uncertain and unquantified benefits.

[1] Fed. R.R. Admin., U.S. Dep’t of Transp., Train Accident (Not at Highway-Rail Crossings) Summary, https://data.transportation.gov/stories/s/dsuf-xcni (last visited May 1, 2026) (showing incidents fell from 4.139 per million miles in 2005 to 2.554 in 2025).

[2] Fed. R.R. Admin., U.S. Dep’t of Transp., Rail Safety Overview Report (1.12), https://data.transportation.gov/stories/s/Rail-Safety-Overview-Report-1-12-/dsuf-xcni (last visited May 1, 2026) (reporting 39 hazmat releases in 2005 and 15 in 2025); see also Ass’n of Am. R.Rs., Press Release, FRA 2023 Data Affirms Rail’s Strong, Sustained Safety Record (Mar. 4, 2024), https://www.aar.org/news/fra-2023-data-affirms-rails-strong-sustained-safety-record (claiming a roughly 75% reduction based on FRA data not currently available).

[3] Fed. R.R. Admin., U.S. Dep’t of Transp., Employee on Duty Fatalities, Injuries, and Illnesses, https://data.transportation.gov/stories/s/Employee-on-Duty-Fatalities-Injuries-and-Illnesses/khzx-vxu4 (last visited Apr. 24, 2026) (showing a 68% decline since 2005).

[4] Bureau of Lab. Stats., U.S. Dep’t of Lab., Table 1: Incidence Rates of Nonfatal Occupational Injuries and Illnesses by Industry and Case Type, 2024, https://www.bls.gov/iif/nonfatal-injuries-and-illnesses-tables/table-1-injury-and-illness-rates-by-industry-2024-national.htm (last visited May 1, 2026) (reporting incidence rates per 100 full-time workers: rail transportation (3.4), general freight trucking (3.1), retail trade (3.0), construction (2.2), grocery stores (4.1), air transportation (6.5), urban transit systems (6.1), and warehousing and storage (4.8)).

[5] Nat’l Transp. Safety Bd., Norfolk Southern Railway Derailment with Subsequent Hazardous Material Release and Fires, East Palestine, Ohio (Feb. 3, 2023), NTSB/RIR-24/05, https://www.ntsb.gov/investigations/AccidentReports/Reports/RIR2405%20CORRECTED.pdf.

[6] Railway Safety Act of 2023, S. 576, 118th Cong. (2023); Railway Safety Act of 2025, H.R. 928, 119th Cong. (2025); Railway Safety Act of 2026, S. 3903, 119th Cong. (2026); see also Press Release, Sen. Maria Cantwell, Cantwell, Husted, Colleagues Reintroduce Bipartisan Railway Safety Act (Feb. 24, 2026), https://www.commerce.senate.gov/2026/2/cantwell-husted-colleagues-reintroduce-bipartisan-railway-safety-act.

[7] Unleashing Prosperity Through Deregulation, Exec. Order No. 14,192, 90 Fed. Reg. 9,065 (Feb. 6, 2025).

[8] Bentley Coffey, Patrick A. McLaughlin & Pietro F. Peretto, Transportation, Innovation and Growth (Working Paper, Apr. 9, 2026), https://papers.ssrn.com/sol3/papers.cfm?abstract_id=6444398.

[9] Bentley Coffey, Patrick A. McLaughlin & Pietro F. Peretto, Regulation and the Cost of Moving Goods (Hoover Inst. Econ. Working Paper, Mar. 20, 2026), https://www.hoover.org/sites/default/files/research/docs/26108-McLaughlin-Coffey-Peretto.pdf.

[10] S. 576, 118th Cong. (2023); see also Cong. Budget Off., Cost Estimate for S. 576 (Sept. 2023), https://www.cbo.gov/publication/59947.

[11] H.R. 928, 119th Cong. (2025); S. 3903, 119th Cong. (2026).

[12] S. 3903, 119th Cong. §§ 102–109 (2026).

[13] Exec. Order No. 14,192, § 3, 90 Fed. Reg. at 9,065, supra note 7.

[14] Id. § 3(b).

[15] See Alex N. Press, As Rail Executives Grow Richer, Train Derailments Have Become Commonplace, Jacobin (Feb. 2023), https://jacobin.com/2023/02/train-derailments-east-palestine-norfolk-southern-profits; see also Gregory Labelle, Letter: Does the U.S. Even Care About Rail Safety? The Numbers Suggest It Doesn’t, LehighValleyLive.com (Apr. 11, 2023), https://www.lehighvalleylive.com/opinion/2023/04/does-the-us-even-care-about-rail-safety-the-numbers-suggest-it-doesnt-letter.html. U.S. figures derive from FRA Form 6180.54 accident/incident data; see Fed. R.R. Admin., Off. of Safety Analysis, Train Accidents and Rates, https://safetydata.fra.dot.gov/officeofsafety/publicsite/query/TrainAccidentsFYCYWithRates.aspx. EU figures derive from Eurostat; see Eurostat, Railway Safety Statistics in the EU, https://ec.europa.eu/eurostat/statistics-explained/index.php?title=Railway_safety_statistics_in_the_EU.

[16] Fed. R.R. Admin., U.S. Dep’t of Transp., Accident/Incident Data (2019), https://data.transportation.gov/stories/s/2ju5-8zxb (showing 784 yard derailments out of 1,344 total derailments in 2019).

[17] Fed. R.R. Admin., supra note 2.

[18] Fed. R.R. Admin., U.S. Dep’t of Transp., Off. of Rsch., Dev. & Tech., An Implementation Guide for Wayside Detector Systems (2019), https://railroads.dot.gov/elibrary/implementation-guide-wayside-detector-systems.

[19] See Ass’n of Am. R.Rs., Freight Railroads Announce Key Safety Measures in Drive to Zero Accidents (Mar. 2023), https://www.aar.org/news/freight-railroads-announce-key-safety-measures-in-drive-to-zero-accidents (announcing approximately 1,000 additional hot-bearing detectors, a 170°F alert threshold, and expanded trending analysis); Ian Jefferies, Ass’n of Am. R.Rs., Statement for the Record Before the Subcomm. on R.Rs., Pipelines & Hazardous Materials, H. Comm. on Transp. & Infrastructure 3 (July 23, 2024), https://www.congress.gov/event/118th-congress/house-event/LC73449/text [hereinafter Jefferies Statement] (citing FRA and AAR data and describing implementation of voluntary commitments, including expansion of AskRail to more than 2.3 million first responders); Fed. R.R. Admin., U.S. Dep’t of Transp., Safety Advisory 2023-01: Evaluation of Policies and Procedures Related to the Use and Maintenance of Hot Bearing Wayside Detectors, 88 Fed. Reg. 13,376 (Mar. 3, 2023); see also Cong. Rsch. Serv., East Palestine, OH, Train Derailment and Hazardous Materials Shipment by Rail: Frequently Asked Questions, R47435 (2024).

[20] CPCS Transcom for Minn. Dep’t of Transp., 2025 Wayside Detector System Study (2026), https://www.lrl.mn.gov/docs/2026/mandated/260603.pdf [hereinafter MnDOT CPCS Report]; see also Ass’n of Am. R.Rs., Freight Railroads Announce Key Safety Measures in Drive to Zero Accidents (Mar. 2023), https://www.aar.org/news/freight-railroads-announce-key-safety-measures-in-drive-to-zero-accidents.

[21] MnDOT CPCS Report, supra note 20.

[22] Id. §§ 2.1, 3.1 (cataloguing 16 distinct wayside detector technologies); id. at 34 (describing distributed fiber-optic sensing); id. (describing light detection and ranging (LiDAR)).

[23] MnDOT CPCS Report, supra note 20, ch. 6, Economic and Industry Impacts, at 53–58 (modeling 10-year cost-benefit scenarios at 10-, 15-, and 20-mile detector spacing on Class II and III track and finding negative cumulative net impacts across all scenarios).

[24] Fed. R.R. Admin., U.S. Dep’t of Transp., High-Hazard Flammable Train Route Assessment & Legacy Tank Car Focused Inspection Program: Summary Report (Jan. 22, 2024), https://railroads.fra.dot.gov/sites/fra.dot.gov/files/2024-01/HRA%20Final%20Report_01.22.24.pdf.

[25] Unleashing Prosperity Through Deregulation, supra note 7; Off. of Mgmt. & Budget, Circular A-4 (2003).

[26] Michael F. Gorman, Rail Safety Policy After East Palestine, Regulation (Summer 2023), https://www.cato.org/regulation/summer-2023/rail-safety-policy-after-east-palestine.

[27] Cong. Rsch. Serv., Positive Train Control (PTC): Overview and Policy Issues, R42637 (2018), https://www.congress.gov/crs-product/R42637; see also Ass’n of Am. R.Rs., Positive Train Control, https://www.aar.org/issue/positive-train-control (estimating industry PTC investment at approximately $15 billion).

[28] Steven Bert et al., The Comprehensive Cost of Rail Incidents in North Carolina, Report No. FHWA/NC/2020-44, at 7 & app. fig. 45, at A-19 (Inst. for Transp. Rsch. & Educ., N.C. State Univ., Dec. 2020).

[29] Id. at 7 fig. 6 & app. figs. 45 & 47, at A-19, A-21.

[30] Coffey, McLaughlin & Peretto, Regulation and the Cost of Moving Goods, supra note 9.

[31] John W. Dawson & John J. Seater, Federal Regulation and Aggregate Economic Growth, 18 J. Econ. Growth 137 (2013), https://papers.ssrn.com/sol3/papers.cfm?abstract_id=2223315.

[32] Compare 49 C.F.R. § 171.8 (current HHFT definition), with S. 576 § 102 (proposed HHT definition).

[33] Ass’n of Am. R.Rs., Recommended Railroad Operating Practices for Transportation of Hazardous Materials, Circular OT-55-R (eff. July 1, 2022), https://www.aar.org/wp-content/uploads/2022/07/2022-07-01-OT-55-R-CPC-KBD.pdf.

[34] Nat’l Transp. Safety Bd., supra note 5.

[35] Cary Coglianese & David Lazer, Management-Based Regulation: Prescribing Private Management to Achieve Public Goals, 37 Law & Soc’y Rev. 691, 693–96 (2003), https://papers.ssrn.com/sol3/papers.cfm?abstract_id=297162.

[36] MnDOT CPCS Report, supra note 20, figs. 2 & 10, at 5–6 & 16–17, §§ 3.2.4–3.2.5, at 32, 34.

[37] Jerry Ellig & Patrick A. McLaughlin, Preventing a Regulatory Train Wreck: Mandated Regulation and the Cautionary Tale of Positive Train Control (Mercatus Ctr., Working Paper, June 2016), https://www.mercatus.org/media/57556/download.

[38] Id.

[39] Coffey, McLaughlin & Peretto, Transportation, Innovation and Growth, supra note 8.

[40] Railway Safety Act of 2026, S. 3903, 119th Cong. § 110(a), (c), (d) (2026); Bureau of Transp. Stat., U.S. Dep’t of Transp., Fleet Composition of Rail Tank Cars Carrying Flammable Liquids: 2023 Report to Congress (Sept. 2023), https://www.bts.gov/sites/bts.dot.gov/files/2023-09/BTS_Tank_Car_Report_To_Congress_9_13_2023.pdf (reporting monthly production rates); Cong. Rsch. Serv., Freight Rail Safety Issues in the 119th Congress, R47911 (2025) (summarizing industry concerns about accelerated retrofit timelines in current Railway Safety Act proposals).

[41] Bureau of Transp. Stat., U.S. Dep’t of Transp., Fleet Composition of Rail Tank Cars Carrying Flammable Liquids: 2023 Report to Congress v–vi (Sept. 2023), https://www.bts.gov/sites/bts.dot.gov/files/2023-09/BTS_Tank_Car_Report_To_Congress_9_13_2023.pdf.

[42] Id.

[43] Train Crew Staffing, 84 Fed. Reg. 24,735, 24,741 (May 29, 2019), https://www.govinfo.gov/content/pkg/FR-2019-05-29/pdf/2019-11088.pdf.

[44] Train Crew Size Safety Requirements, 89 Fed. Reg. 25,502, 25,508 (Apr. 9, 2024), https://www.federalregister.gov/documents/2024/04/09/2024-06625/train-crew-size-safety-requirements.

[45] Marc Scribner, Toward Performance-Based Transportation Safety Regulation, Competitive Enter. Inst. (Mar. 2017), https://cei.org/studies/toward-performance-based-transportation-safety-regulation.

[46] MnDOT CPCS Report, supra note 20, ch. 6.

Executive Summary

Global health governance has shifted in ways that weaken democratic accountability while concentrating policy influence. The World Health Organization (the WHO), once a technical coordinating body, now operates within a broader ecosystem of Democratically Deficient Organizations (DoDOs)—intergovernmental institutions, philanthropic foundations, NGOs, and private actors that shape policy without meaningful public oversight. The move from assessed contributions to donor-driven, earmarked funding has severed the link between WHO priorities and member-state control. In its place, a small group of funders—most notably the Bill & Melinda Gates Foundation and Bloomberg Philanthropies—now exert outsized influence over global health agendas.

This brief shows how that system operates in practice. In tobacco control and pandemic governance, foundation funding, WHO authority, NGO advocacy, and academic research reinforce one another to produce policy consensus insulated from scrutiny. Law & economics frameworks help explain the result: incentives favor persistence over performance. The WHO’s Framework Convention on Tobacco Control (FCTC) has not accelerated global declines in smoking, while discouraging harm-reduction approaches that have succeeded in countries such as Sweden. Proposals to expand WHO authority in pandemic preparedness risk replicating the same institutional failures revealed during COVID-19.

The problem is not insufficient resources, but weak accountability. The brief proposes reforms to restore it: rebalance WHO funding toward assessed contributions, strengthen transparency and conflict-of-interest rules, open governance processes, embrace harm reduction, and return policymaking authority to domestic democratic institutions. Without such changes, the continued expansion of the DoDO model will deepen existing failures—with consequences measured in human lives.

“The urge to save humanity is almost always a false front for the urge to rule.”

— H.L. Mencken

I.        Introduction

The World Health Organization was founded in 1948 with a mandate to coordinate international efforts to combat disease and improve public health. In its early decades, it built considerable prestige through tightly targeted, outcome-oriented vertical programs: mass vaccination campaigns, technical assistance to developing nations, and focused assaults on diseases such as smallpox, malaria, yaws, and tuberculosis.[1] The eradication of smallpox in 1980 remains the WHO’s signature achievement—and arguably the greatest accomplishment of any international organization.[2]

The seeds of institutional decay, however, were already taking root. In 1978, the Alma-Ata Declaration announced the goal of “Health for All by the Year 2000,” shifting the WHO’s strategy from vertical disease control to a comprehensive, horizontal model centered on primary health care and the social determinants of health.[3] This was not a marginal adjustment. It marked a shift in institutional character. Health policy became as political as scientific. In that arena, the advantage rarely goes to those with the best evidence; it goes to those with the loudest voices and the deepest pockets.[4]

This strategic pivot opened the door to broader structural changes in global health governance—changes that fit squarely within the framework of Democratically Deficient Organizations (DoDOs).[5] As explained in the introductory paper in this series, the DoDO concept describes how intergovernmental bodies, NGOs, foundations, and corporations exercise substantial policy-shaping authority while remaining largely insulated from democratic accountability. The result is a self-reinforcing ecosystem: failure justifies expansion, dissent is recast as moral deviance, and those most affected by policy decisions have little meaningful voice in shaping them.

This paper applies the DoDO framework to global health. The global health DoDO ecosystem centers on the WHO, alongside a constellation of philanthropic foundations (led by the Bill & Melinda Gates Foundation and Bloomberg Philanthropies), advocacy NGOs (often funded by those same foundations), pharmaceutical companies, and a revolving cadre of public health academics whose careers depend on the regulatory apparatus they help design.

Bruce Yandle’s “Baptists and bootleggers” analogy provides a useful complement. Baptists supply the moral justification for regulation, while bootleggers quietly profit from it.[6] In global health governance, the “Baptists” include public health campaigners, WHO officials, and celebrity philanthropists who frame every policy question in terms of lives saved or children protected. The bootleggers include pharmaceutical companies that benefit from regulatory choices, consulting firms that administer programs, NGOs whose funding depends on sustained crisis, and academics whose grants and publications flow from institutional consensus. No conspiracy is required. Incentives do the work.

II.        The Political Economy of WHO Financing

To understand how the WHO became a DoDO, follow the money. The organization’s original design relied on assessed contributions from member states—mandatory dues based on national income. The World Health Assembly (WHA) retained collective control over how those funds were spent.[7] The system was imperfect, but it preserved at least a loose connection between the WHO’s priorities and the collective will of its members. That connection has now been severed.

By the 2023–2024 biennium, roughly 84% of the WHO’s budget came from voluntary contributions, most of them earmarked by donors rather than the WHA.[8] In that same period, the United States was the largest contributor, followed by the Bill & Melinda Gates Foundation and Germany. Each of those three donors individually provided more funding than most member states combined.[9] Lawrence Gostin, himself a WHO collaborating center director, noted in 2015 that the organization fully controlled only about 20% of its budget.[10] Donors effectively dictate the rest.

This is not just a shift in financing. It is a quiet transformation in governance. When a private foundation contributes more to a UN agency than nearly every sovereign state—and specifies how the money must be spent—it acquires policymaking authority without any of the constraints that bind even the most dysfunctional governments. The Gates Foundation faces no elections, no parliamentary scrutiny, no freedom-of-information requests, and no judicial review. It answers only to its founders.[11]

Devi Sridhar and Ngaire Woods describe this arrangement as “trojan multilateralism”: the use of multilateral institutions to advance bilateral and private donor agendas.[12] The pattern is familiar. A foundation selects a priority—polio eradication, pandemic preparedness, vaccine delivery. It funds WHO programs aligned with that priority, often building parallel structures such as GAVI (the Global Alliance for Vaccines and Immunization) and CEPI (the Coalition for Epidemic Preparedness Innovations) that sit alongside, and often sidestep, the WHO’s formal governance. It also funds academic research to validate the approach, NGOs to advocate for it, and media coverage to amplify it. The appearance of multilateralism remains. The substance does not. Accountability to member states—let alone voters—evaporates.[13]

The concern here is not philanthropy as such. Foundations can and often do support valuable health interventions. A foundation that builds clinics or expands vaccine access raises few objections. The problem arises when funding comes with policy conditions. At that point, financial support becomes a vehicle for influence. In intergovernmental bodies and executive agencies—already distant from democratic oversight—those conditions further dilute accountability. In the case of the WHO, earmarked funding has compounded weak WHA oversight and accelerated its evolution into a DoDO.

III.        Gates and the Philanthropic Policy Model

The Bill & Melinda Gates Foundation warrants particular scrutiny—not because its intentions are necessarily malign, but because its influence is so extensive that it reshapes the structure of global health governance. With annual spending on health-related activities averaging roughly $4 billion in recent years, much of it directed to NGOs and academic researchers, the foundation is not merely a major donor. It operates as a de facto health policy institution, shaping spending priorities and regulatory agendas, especially at the WHO and in lower-income countries.[14]

Anne-Emanuelle Birn traces the origins of this model to the Rockefeller Foundation’s dominance of international health in the early 20th century.[15] Linsey McGoey’s analysis of the Gates Foundation shows how its grant-making creates a dense web of financial dependencies that discourages dissent.[16] Universities, think tanks, NGOs, and media organizations that receive Gates funding have strong incentives to endorse its approach and equally strong incentives to avoid scrutinizing it. McGoey’s conclusion is blunt: there is “no such thing as a free gift.” Every grant carries strings, whether acknowledged or not.

The foundation’s influence on the WHO raises sharper concerns. It funds not only the academics and NGOs whose research and advocacy inform WHO decision making, but also the WHO itself—both directly and indirectly. For example, it funds GAVI, which the WHO helps govern.[17] More importantly, as David McCoy and his co-authors document in The Lancet, the foundation participates directly in policymaking across the system:[18]

The foundation actively engages in policy making and agenda setting activities; it has representatives that sit on the governing structures of many global health partnerships; it is part of a self-appointed group of global health leaders known as the H8 (together with WHO, the World Bank, GAVI Alliance, the Global Fund, UNICEF, the United Nations Population Fund [UNFPA], and UNAIDS); and has been involved in setting the health agenda for the G8. The Gates Foundation is also involved in setting the research agenda of several public health priorities, a role that was controversially criticised by the former head of WHO’s malaria programme, who complained that the dominance of the Gates Foundation in malaria research risked stifling the diversity of views among scientists.[19]

This is the DoDO ecosystem in full view. A large philanthropic foundation—accountable primarily to its founders and board—funds research, advocacy, and media, while also participating directly in policymaking and agenda setting. The foundation and its institutional allies operate largely outside democratic oversight, yet shape regulatory and policy decisions that affect billions.

IV.        Bloomberg and the Financing of Tobacco Control

If the Gates Foundation shows how one large DoDO can shape health policy broadly, Bloomberg Philanthropies shows how a single DoDO can dominate a narrow regulatory domain. Since 2005, Michael Bloomberg has committed more than $1.6 billion to global tobacco-control initiatives through the Bloomberg Initiative to Reduce Tobacco Use.[20]

The initiative operates through a network of implementing partners, including the Campaign for Tobacco-Free Kids (CTFK), Vital Strategies, the Johns Hopkins Bloomberg School of Public Health, the CDC Foundation, and the WHO.[21] These organizations provide technical assistance, fund advocacy campaigns, support research, and help governments implement tobacco-control policies in low- and middle-income countries.

Through this structure, philanthropic funding supports both the development of global norms—particularly those associated with the WHO’s Framework Convention on Tobacco Control (FCTC)—and the domestic advocacy needed to entrench those norms in national regulatory systems.

Adopted in 2003, the FCTC was the first treaty negotiated under WHO auspices and remains the organization’s most ambitious exercise of normative authority. More than 180 countries are parties. Its objectives were straightforward: reduce tobacco use, limit exposure to secondhand smoke, and address the global burden of tobacco-related disease.[22]

The results have fallen short. When the FCTC entered into force in 2005, the world had roughly 1.1 billion smokers. Nearly two decades later, that number remains largely unchanged.[23] Population growth of about 20% has produced a corresponding decline in prevalence, but that decline largely reflects trends already underway before the treaty.

Steven Hoffman and his co-authors confirm this point in a rigorous quasi-experimental analysis published in the BMJ, finding “no evidence to indicate that global progress in reducing cigarette consumption has been accelerated by the FCTC treaty mechanism.”[24]

Some countries have achieved sharper declines in smoking prevalence—notably Sweden, the United Kingdom, and Japan. These outcomes largely reflect the availability of lower-risk nicotine alternatives, including snus, e-cigarettes, and heated tobacco products, which enable substitution away from combustible cigarettes.

Sweden provides the clearest population-level example. Since the introduction of snus in the 1970s, Swedish consumers have steadily shifted toward this lower-risk form of smokeless tobacco. Most EU countries banned snus in 1992, preventing similar substitution. Sweden secured an exemption when it joined the EU in 1995.

The divergence is stark. Adult male smoking prevalence in Sweden has fallen below 5%, far below the EU average of more than 20%.[25] Smoking-related mortality among Swedish men is correspondingly the lowest in the EU, a pattern Lars Ramström and Tom Wikmans link to substitution from cigarettes to snus.[26]

The EU ban on snus created a decades-long natural experiment across an entire customs union. The outcome is clear: harm reduction works.[27] Continuing the experiment looks less like caution than inertia bordering on negligence. (Proposed as a formal trial, it would likely fail any ethics review.)

The FCTC and the WHO have nonetheless rejected snus as a harm-reduction tool. A 2003 WHO report states: “There is no evidence to recommend that any smokeless tobacco product should be used as part of a harm reduction strategy.”[28] More recent guidance reiterates the position,[29] emphasizing risks while discouraging comparative claims about relative harm.[30]

The WHO applies an even more precautionary approach to e-cigarettes—rebranded as “electronic nicotine delivery systems,” or ENDS—despite evidence that they are far less harmful than cigarettes and more effective than conventional nicotine-replacement therapies at supporting cessation.[31]

The FCTC sits at the center of Bloomberg’s campaign. The institutional architecture is a textbook DoDO. Bloomberg funds academic research that provides intellectual support. It funds NGOs such as CTFK and members of the Global Alliance for Tobacco Control (GATC), which lobby governments to adopt FCTC-aligned policies. It also funds the WHO, which supplies the normative framework those NGOs deploy.

Bloomberg Philanthropies is not the only funder. The Gates Foundation, the Wellcome Trust, the Robert Wood Johnson Foundation, and various governments also contribute.[32] In some cases, government agencies fund NGOs that advocate for the agencies’ own policies, creating the appearance of grassroots support and short-circuiting democratic scrutiny.[33]

Pharmaceutical companies add another layer. Firms that manufacture nicotine-replacement therapies and cessation drugs benefit from regulatory limits on competing nicotine products. Some have funded advocacy organizations that lobby for those restrictions.[34] The logic of Yandle’s model becomes visible: bootleggers underwriting Baptists.

Taken together, this ecosystem reflects a top-down agenda shaped by a small group of funders, with Bloomberg at its center. Money flows through academics and NGOs that generate and promote policy arguments,[35] creating the appearance of broad consensus while narrowing the range of acceptable views.

This manufactured consensus is reinforced by expansive interpretations of FCTC Article 5.3. Intended to shield policy from tobacco-industry influence, it has been applied so broadly that it restricts engagement with entities only tangentially connected to tobacco.

The result borders on the absurd. The FCTC Conference of the Parties has met behind closed doors since 2010, excluding journalists and the public,[36] while NGOs aligned with the prevailing framework participate and publicly shame dissenting countries through “Dirty Ashtray Awards.”[37]

Interpol was denied observer status at COP6 in 2014 because it had accepted funding from Philip Morris International for anti-smuggling efforts.[38] The WHO’s Tobacco Cessation Consortium, launched in 2022, excludes tobacco and non-medical nicotine firms under Article 5.3 while including pharmaceutical and technology companies as partners.[39]

The contrast is stark. Companies that profit from medically sanctioned cessation products are welcomed. Companies that produce lower-risk alternatives—often more effective at helping smokers quit—are excluded.

The consequences can be perverse. In 2022, the WHO refused Emergency Use Listing for Covifenz, a COVID-19 vaccine developed by the Canadian company Medicago, because Philip Morris International held roughly a one-third stake in the firm. Health Canada had already authorized the vaccine, with efficacy exceeding 70%.[40] The WHO’s objection was not safety, but institutional policy under the FCTC.

The result was that low- and middle-income countries lost access to a viable vaccine during a pandemic because of a tobacco company’s investment. It is difficult to imagine a clearer case of policy overriding public health.

Some insiders have begun to question the system. Former WHO department directors Robert Beaglehole and Ruth Bonita conclude in The Lancet that the FCTC “is no longer fit for purpose, especially for low-income countries,” and call for harm reduction to be integrated into the framework.[41]

The treaty now functions less as a tool for harm reduction than as an institution oriented toward its own preservation.[42] The incentives embedded in the DoDO ecosystem have, so far, impeded reform.

V.        Cascades, Consensus, and Policy Entrenchment

The persistence of failed policies in the face of contrary evidence demands explanation. Timur Kuran and Cass Sunstein’s concept of availability cascades offers a useful framework.[43] An availability cascade occurs when repetition, not evidence, drives belief. The more often a claim circulates in public discourse, the more cognitively “available” it becomes—and the more likely people are to accept it as true.[44]

In the global health DoDO ecosystem, these cascades are driven by availability entrepreneurs. The mechanism is straightforward. A foundation funds research at a prestigious university. The research supports the foundation’s preferred policy. The foundation funds NGOs to promote the findings. Those NGOs lobby the WHO. The WHO issues guidance citing the research. Governments implement the guidance. The foundation then funds further research showing that implementation falls short, justifying additional funding and renewed intervention.

This circuit operates with minimal external constraint. At no point does meaningful democratic accountability enter the process. Research circulates within a closed funding network. NGOs answer to donors, not the populations they claim to represent. The WHO’s governance structures are dominated by health ministry officials rather than elected representatives. Governments often implement WHO guidance through administrative action rather than legislation, further insulating the process from political scrutiny.

Christopher Snowdon has documented a related phenomenon in the form of “sock puppet” NGOs—organizations that present as independent but rely on funding from the same entities whose policies they promote.[45] In that setting, a government agency uses an NGO to advance policies that would struggle under direct political scrutiny.

In global health, the pattern is broader and less transparent. Funding often originates from private foundations rather than governments, obscuring the underlying relationships.

The Gates Foundation’s simultaneous funding of the WHO, GAVI, academic research centers, and advocacy NGOs creates what can fairly be described as full-spectrum dominance of the policy space. Nearly every voice in the conversation is, directly or indirectly, funded by the same source.

VI.        Institutional Expansion Without Reform

The FCTC model is now spreading across global health governance with notable speed. The same DoDO architecture—WHO normative authority, foundation funding, NGO advocacy, and closed-door decision making—has expanded beyond tobacco to alcohol, sugar, ultra-processed foods, and pandemic governance.[46]

The proposed WHO Pandemic Agreement (formerly the “Pandemic Treaty”) marks the most ambitious extension of this model to date.[47] Drawing on the FCTC precedent, it would expand the WHO’s role in pandemic preparedness and response. Proposed provisions include surveillance obligations, benefit-sharing requirements for pathogen data, and “sustainable financing” mechanisms that would further entrench reliance on voluntary contributions—and with it, donor influence over priorities.

The COVID-19 pandemic exposed the institutional weaknesses the DoDO framework predicts. The WHO’s delayed response to the initial outbreak, its acceptance of Chinese government assurances, and its reluctance to investigate the laboratory-leak hypothesis reflected structural constraints, not isolated errors. An organization dependent on the goodwill of powerful member states and donors will predictably hesitate to challenge them.[48]

The Independent Panel for Pandemic Preparedness and Response reached a similar conclusion: the WHO lacked both the authority and the independence to respond effectively. The proposed remedy—more authority, more funding, more institutional machinery—misses the point. Expanding the powers of an institution already shaped by donor influence does not resolve the underlying problem.[49] It reinforces it.

The pattern extends to pharmaceutical regulation. The WHO’s prequalification program, funded in significant part by the Gates Foundation through UNITAID, plays a central role in determining which drugs and vaccines qualify for purchase with international donor funds.[50]

Despite decades of WHO involvement, substandard and falsified medicines remain widespread in low- and middle-income countries. Institutional incentives emphasize process compliance—reports, meetings, and grant disbursement—over measurable outcomes.[51]

The system excels at coordination and documentation. It performs less well at ensuring that patients in Lagos or Dhaka receive effective medicines.[52]

VII.        Incentives, Insulation, and Persistence

The preceding analysis reveals a consistent structural logic that defines health DoDOs. Several reinforcing mechanisms sustain the system and explain its persistence.[53]

Incentive misalignment. Every actor in the health DoDO ecosystem benefits from the continuation of the problem rather than its resolution. The WHO’s budget expands with each new health crisis. Foundations justify their existence by identifying new threats. NGOs rely on ongoing emergencies to sustain funding. Pharmaceutical companies profit from both disease and treatment. Academics build careers around an expanding regulatory apparatus. This is not conspiracy but ecology. Each actor pursues its own incentives, and those incentives align in ways that perpetuate the system.[54]

Democratic insulation. The system operates at a remove from democratic accountability. International agreements bind successor governments. Voluntary contributions bypass legislative appropriations. Expert committees, rather than elected officials, set technical standards. The populations most affected—often in developing countries—have little or no voice in the institutions that shape the policies governing their health.

Knowledge monopoly. Foundations influence not only policy, but the production of policy-relevant knowledge. By funding research, training researchers, and shaping institutional prestige, they exercise significant agenda-setting power. Dissent rarely faces direct censorship. It is filtered out through funding decisions. Researchers who challenge prevailing assumptions risk losing grants, publication opportunities, and career advancement. The result is a self-reinforcing consensus.

Failure as justification. Failure does not trigger reform; it triggers expansion. When the FCTC falls short of reducing global tobacco use, the response is to call for stronger implementation and additional funding. When the WHO underperforms during crises such as Ebola or COVID-19, the response is to expand its authority and resources. Mancur Olson’s logic of collective action explains why.[55] The beneficiaries of the system—bureaucrats, consultants, NGOs—are concentrated, organized, and motivated. The costs fall on diffuse populations that face high barriers to collective action. The system therefore persists, even when its results do not justify its expansion.

VIII.        Accountability Before Expansion

The health DoDO complex may be the most consequential of all DoDO ecosystems because its failures are measured in human lives. Eight million tobacco deaths annually. Millions more from malaria, tuberculosis, and childhood diseases that persist not for lack of effective interventions, but because the institutional apparatus prioritizes process over outcomes. A pandemic response that too often deferred to institutional constraints. And now a proposed Pandemic Agreement that would entrench the same structures that failed.

The path forward requires more than marginal reform. It requires restructuring international health cooperation around accountability and subsidiarity.

Funding reform. The WHO’s financing model must restore the primacy of assessed contributions. As long as the organization depends on voluntary, earmarked funding from a small group of donors, it will reflect their priorities rather than those of its members. If governments are unwilling to fund the WHO through assessed contributions, that reluctance is itself informative. It suggests they do not value its activities enough to support them through democratic fiscal processes.

Transparency and conflicts of interest. The revolving door among foundations, the WHO, and the NGO sector should be transparent and subject to conflict-of-interest rules at least as stringent as those applied to the tobacco industry under Article 5.3. If influence from Philip Morris is unacceptable, influence from major foundations should face comparable scrutiny. At a minimum, equivalent standards of transparency and accountability should apply.

Open governance. The FCTC’s decision-making process must be opened to public scrutiny. Closed-door negotiations on policies affecting billions are incompatible with democratic governance. Excluding journalists, independent researchers, and dissenting voices protects institutional incumbents, not the public.

Harm reduction. Harm reduction should be treated as a core principle of health policy, not a deviation from orthodoxy. Evidence from Sweden, the United Kingdom, Japan, and New Zealand shows that risk-proportionate regulation saves lives. The FCTC’s refusal to incorporate this evidence reflects institutional incentives, not scientific caution.

Democratic control. Health policies that affect domestic populations should be determined by domestic democratic institutions, not delegated to international bodies with attenuated accountability. International cooperation has a role in coordination, information sharing, and technical assistance. It is not a substitute for democratic self-governance.

The architects of the postwar international order understood that legitimacy flows from accountability. They designed the WHO to serve its member states, not to direct them. That relationship has reversed. Restoring it will require confronting not only entrenched institutional interests, but also the assumption that technical expertise and good intentions can substitute for democratic consent.

They cannot. They never have. In global health, where the stakes are measured in millions of lives, the cost of that assumption is too high to ignore.

[1] World Health Org., The Third Ten Years of the World Health Organization: 1968–1977, 45–60 (1978).

[2] Frank Fenner et al., Smallpox and Its Eradication 1345–48 (World Health Org. 1988).

[3] World Health Org., Declaration of Alma-Ata, Int’l Conf. on Primary Health Care (Sept. 12, 1978), https://www.who.int/teams/social-determinants-of-health/declaration-of-alma-ata.

[4] Marcos Cueto, The Origins of Primary Health Care and Selective Primary Health Care, 94 Am. J. Pub. Health 1864, 1864–74 (2004).

[5] Julian Morris, Defending Democracy from the DoDOs: How Power Escapes Democratic Control, Int’l Ctr. for L. & Econ. 4–6 (Mar. 12, 2026), https://laweconcenter.org/wp-content/uploads/2026/03/Defending-Democracy-from-the-DoDOs.pdf.

[6] Bruce Yandle, Bootleggers and Baptists—The Education of a Regulatory Economist, 7 Regulation 12, 12–16 (1983); Adam Smith & Bruce Yandle, Bootleggers and Baptists: How Economic Forces and Moral Persuasion Interact to Shape Regulatory Politics 15–40 (Cato Inst. 2014).

[7] Devi Sridhar & Ngaire Woods, Trojan Multilateralism: Global Cooperation in Health, 4 Global Pol’y 325, 325–35 (2013).

[8] World Health Org., Programme Budget 2022–2023: Financial Overview and Contributors (2023), https://open.who.int (last visited Mar. 10, 2026); see also Flexibly Funding WHO? An Analysis of Its Donors’ Voluntary Contributions, 8 BMJ Glob. Health e011232 (2023).

[9] Id.

[10] Lawrence O. Gostin et al., The Normative Authority of the World Health Organization, 129 Pub. Health 854, 856–58 (2015).

[11] Sophie Harman, The Bill and Melinda Gates Foundation and Legitimacy in Global Health Governance, 22 Global Governance 349, 349–68 (2016).

[12] Sridhar & Woods, supra note 7.

[13] Devi Sridhar, Who Sets the Global Health Research Agenda? The Challenge of Multi-Bi Financing, 9 PLoS Med. e1001312 (2012).

[14] Calculations by authors based on Bill & Melinda Gates Found., 2023 Annual Report (2024), and Bill & Melinda Gates Found., 2024 Annual Report (2025) (finding roughly $2 billion in annual spending classified as “global health,” with comparable additional spending on closely related categories such as vaccine development, malaria, and tuberculosis).

[15] Anne-Emanuelle Birn, Philanthrocapitalism, Past and Present: The Rockefeller Foundation, the Gates Foundation, and the Setting(s) of the International/Global Health Agenda, 12 Hypothesis e8 (2014).

[16] Linsey McGoey, No Such Thing as a Free Gift: The Gates Foundation and the Price of Philanthropy (Verso 2015).

[17] David McCoy et al., The Bill & Melinda Gates Foundation’s Grant-Making Programme for Global Health, 373 Lancet 1645, 1645–53 (2009).

[18] Judith Richter, Public-Private Partnerships and International Health Policy-Making: How Can Public Interests Be Safeguarded? 31–50 (World Health Org. 2004).

[19] McCoy et al., supra note 17, at 1650.

[20] Bloomberg Philanthropies, Bloomberg Initiative to Reduce Tobacco Use, https://www.bloomberg.org/public-health/reducing-tobacco-use (last visited Apr. 2, 2026); Philanthropy N.Y., Bloomberg Philanthropies Launches $420 Million Initiative to Reduce Tobacco Use Globally (Feb. 2, 2023), https://philanthropynewyork.org/news/bloomberg-philanthropies-commits-additional-420-million-reduce-tobacco-use-globally (last visited Apr. 2, 2026).

[21] Campaign for Tobacco-Free Kids, Bloomberg Initiative to Reduce Tobacco Use, https://www.tobaccofreekids.org/what-we-do/global/bloomberg (last visited Apr. 2, 2026).

[22] WHO Framework Convention on Tobacco Control art. 5.3, May 21, 2003, 2302 U.N.T.S. 166.

[23] GBD 2019 Tobacco Collaborators, Spatial, Temporal, and Demographic Patterns in Prevalence of Smoking Tobacco Use and Attributable Disease Burden in 204 Countries and Territories, 1990–2019, 397 Lancet 2337 (2021); World Health Org., Global Report on Trends in Prevalence of Tobacco Use 2000–2030 (4th ed. 2021).

[24] Steven J. Hoffman et al., Impact of the WHO Framework Convention on Tobacco Control on Global Cigarette Consumption: Quasi-Experimental Evaluations Using Interrupted Time Series Analysis and In-Sample Forecast Event Modelling, 365 BMJ l2287 (2019); see also Steven J. Hoffman & Cigdem Tan, Overview of Systematic Reviews on the Health-Related Effects of Government Tobacco Control Policies, 15 BMC Pub. Health 744 (2015).

[25] Eur. Comm’n, Special Eurobarometer 539: Attitudes of Europeans Towards Tobacco and Related Products (2023).

[26] Lars Ramström & Tom Wikmans, Mortality Attributable to Tobacco Among Men in Sweden and Other European Countries: An Analysis of Data in a WHO Report, 12 Tobacco Induc. Dis. 14 (2014).

[27] Id.; see also Karl Fagerström & Elsy-Britt Schildt, Should the European Union Lift the Ban on Snus? Evidence from the Swedish Experience, 98 Addiction 1191, 1191–95 (2003).

[28] World Health Org., Scientific Advisory Committee on Tobacco Product Regulation (SACTob), Recommendation on Smokeless Tobacco Products 4 (2003).

[29] World Health Org. Framework Convention on Tobacco Control Secretariat, Smokeless Tobacco Products: Guide to Drafting Regulation to Implement the WHO FCTC 4, 8 (2024).

[30] Id. at 22.

[31] Julian Morris, The World Health Organization’s Opposition to Tobacco Harm Reduction: A Threat to Public Health?, Reason Fdn. (2016); Ann McNeill et al., E-Cigarettes: An Evidence Update—A Report Commissioned by Public Health England (Aug. 2015) (finding vaping approximately 95% less harmful than combustible cigarettes); Royal Coll. of Physicians, Nicotine Without Smoke: Tobacco Harm Reduction 56–75 (2016) (endorsing e-cigarettes as a cessation tool); Peter Hajek et al., A Randomized Trial of E-Cigarettes Versus Nicotine-Replacement Therapy, 380 New Eng. J. Med. 629, 629–37 (2019) (finding e-cigarettes nearly twice as effective as nicotine-replacement therapy for smoking cessation); Angela Difeng Wu et al., Electronic Cigarettes for Smoking Cessation: An Overview of Systematic Reviews and Evidence and Gap Map, Addiction (Mar. 26, 2026) (reporting higher ≥6-month cessation rates with nicotine e-cigarettes across multiple comparisons); Renée O’Leary et al., Examining E-Cigarettes as a Smoking Cessation Treatment: A Critical Umbrella Review Analysis, 266 Drug Alcohol Depend. 112520 (2025) (finding e-cigarettes more effective than other cessation treatments).

[32] See, e.g., Bill & Melinda Gates Found., Michael Bloomberg and Bill Gates Join to Combat Global Tobacco Epidemic, https://www.gatesfoundation.org/ideas/media-center/press-releases/2008/07/michael-bloomberg-and-bill-gates-join-to-combat-global-tobacco-epidemic (last visited Apr. 2, 2026); Wellcome Tr., Commercial Determinants of Health Programme; Robert Wood Johnson Found., Tobacco Policy Research and Advocacy Programs, https://www.rwjf.org (last visited Mar. 10, 2026). Philanthropic funding for global tobacco-control advocacy often flows through a small set of intermediary organizations—such as the Campaign for Tobacco-Free Kids, Vital Strategies, and the International Union Against Tuberculosis and Lung Disease—that distribute grants, coordinate advocacy campaigns, and provide technical assistance to governments. This structure creates a tightly integrated policy network aligning funding, research, and advocacy. See Sridhar & Woods, supra note 8.

[33] Christopher Snowdon, Sock Puppets: How the Government Lobbies Itself and Why 8–18 (Inst. of Econ. Aff. 2012).

[34] See, e.g., Pfizer Inc., U.S. Medical, Scientific, Patient and Civic Organization Funding Report: Fourth Quarter 2014 (2015) (listing grants to the Campaign for Tobacco-Free Kids, including $25,000 for the “2014 Youth Advocates of the Year Event”), https://cdn.pfizer.com/pfizercom/responsibility/grants_contributions/Yes_Report-2014-08.29.2017.pdf (last visited Apr. 2, 2026).

[35] Bloomberg Philanthropies, E-cigarettes, https://www.bloomberg.org/public-health/reducing-tobacco-use/e-cigarettes (last visited Apr. 4, 2026).

[36] See, e.g., World Health Org. Framework Convention on Tobacco Control, Conf. of the Parties, Fifth Sess., Summary Record of the Second Committee B Meeting, ¶ 2, FCTC/COP/5/REC/2 (Nov. 17, 2012), https://apps.who.int/gb/fctc/PDF/cop5/FCTC_COP5_REC2.pdf (last visited Apr. 4, 2026); World Health Org. Framework Convention on Tobacco Control, Conf. of the Parties, Rules of Procedure, Rule 5 (providing that sessions may be closed to the public).

[37] See Framework Convention Alliance (now Global Alliance for Tobacco Control), Dirty Ashtray Awards, COP advocacy materials (issued at successive Conferences of the Parties).

[38] WHO Framework Convention on Tobacco Control, Guidelines for Implementation of Article 5.3 (2008) (recommending that governments limit interactions with the tobacco industry and avoid partnerships with tobacco-related entities); see also WHO Framework Convention on Tobacco Control, Conf. of the Parties, Sixth Sess., Interpol’s Application for Observer Status, FCTC/COP/6/4 (July 24, 2014); see also Interpol, Interpol Global Complex for Innovation and Cooperation on Illicit Trade (discussing cooperation on illicit tobacco trade).

[39] World Health Org., Pharma, Tech and Social Media Companies Join Forces with WHO to Launch the Tobacco Cessation Consortium During World Health Summit in Berlin, Germany (Oct. 20, 2022), https://www.who.int/news/item/20-10-2022-pharma–tech-and-social-media-companies-join-forces-with-who-to-launch-the-tobacco-cessation-consortium-during-world-health-summit-in-berlin–germany.

[40] Tony Kirby, Philip Morris Ejected from Canadian Vaccine Collaboration, 11 Lancet Respiratory Med. e40 (2023); see also WHO Refuses Emergency Use Listing for Medicago COVID-19 Vaccine Due to Philip Morris International Ties, Tobacco Asia (Mar. 18, 2022).

[41] Robert Beaglehole & Ruth Bonita, Tobacco Control: Getting to the Finish Line, 399 Lancet 1865 (2022).

[42] Christopher Snowdon, Killjoys: A Critique of Paternalism 85–96 (Inst. of Econ. Aff. 2017) (discussing the “closed shop” dynamics of FCTC COP meetings).

[43] Timur Kuran & Cass R. Sunstein, Availability Cascades and Risk Regulation, 51 Stan. L. Rev. 683, 683–768 (1999).

[44] Id. at 720–35.

[45] Snowdon, supra note 33, at 15–18 (describing how government-funded NGOs can create an “astroturf” effect that simulates grassroots support for policies originating within the bureaucracy).

[46] Ilona Kickbusch et al., The Commercial Determinants of Health, 401 Lancet 1223 (2023); Jeff Collin, Kelley Lee & Karen Bissell, The Framework Convention on Tobacco Control: The Politics of Global Health Governance, 23 Third World Q. 265 (2002); Christopher Snowdon, The Art of Suppression: Pleasure, Panic and Prohibition Since 1800 185–220 (Little Dice 2011).

[47] World Health Org., WHO Pandemic Agreement: Zero Draft, WHO Doc. A/INB/4/3 (Feb. 1, 2023).

[48] Indep. Panel for Pandemic Preparedness & Response, COVID-19: Make It the Last Pandemic 22–40 (2021); Lawrence O. Gostin, Suerie Moon & Benjamin M. Meier, Reimagining Global Health Governance in the Age of COVID-19, 110 Am. J. Pub. Health 1615 (2020); World Health Org., WHO-Convened Global Study of Origins of SARS-CoV-2: China Part (2021) (describing the laboratory-incident hypothesis as “extremely unlikely,” a conclusion later questioned by governments and scientific commentators); David Relman, Opinion: To Stop the Next Pandemic, We Need to Unravel the Origins of COVID-19, 375 Science 1262 (2021).

[49] Lawrence O. Gostin, The Future of the World Health Organization: Lessons Learned from Ebola, 93 Milbank Q. 475, 480–82 (2015).

[50] Roger Bate, Phake: The Deadly World of Falsified and Substandard Medicines 145–70 (AEI Press 2014).

[51] World Health Org., Substandard and Falsified Medical Products (Fact Sheet) (updated 2023), https://www.who.int/news-room/fact-sheets/detail/substandard-and-falsified-medical-products; Shinobu Ozawa et al., Prevalence and Estimated Economic Burden of Substandard and Falsified Medicines in Low- and Middle-Income Countries, 1 JAMA Network Open e181662 (2018); Roger Bate et al., Substandard and Falsified Anti-Tuberculosis Drugs: A Preliminary Field Analysis, 17 Int’l J. Tuberculosis & Lung Disease 308, 308–11 (2013).

[52] Amir Attaran et al., How to Achieve International Action on Falsified and Substandard Medicines, 345 BMJ e7381 (2012).

[53] Morris, supra note 5, at 15–18.

[54] Yandle, supra note 6, at 14–15; Morris, supra note 5, at 4–7.

[55] Mancur Olson, The Logic of Collective Action: Public Goods and the Theory of Groups 53–65 (Harvard Univ. Press 1965).