Regulatory Comments

ICLE Response to First Review of the Digital Markets Act

List of Core Platform Services and Designation of Gatekeepers

Do you have any comments or observations on the current list of core platform services?

We would like to take this opportunity to focus on the question of whether artificial intelligence (AI) should be designated as a core platform service (CPS). Our view is that such a designation would be inappropriate and counterproductive. The reason is simple: AI is not a single, unitary technology or service. Treating it as though it would produce flawed regulatory outcomes, confuse enforcement, and risk stifling innovation.

AI is not a monolith but a heterogeneous collection of technologies, methods, and applications (Radic & Stout, 2024). There is no such thing as a clearly defined “AI market”. Instead, AI is better understood as a loosely connected bundle of technologies, each with its own characteristics, players, and business models. This becomes clear when looking at the different layers of what is sometimes called the “AI stack”. At the foundational level are semiconductors, computing hardware, and cloud or XaaS providers that make raw processing power available. On top of this comes data collection and preparation—an entire industry devoted to cleaning, labelling, and managing data. Model training follows, with quite different approaches—such as supervised, unsupervised, reinforcement, and transfer learning—each serving distinct purposes. Finally, trained models are deployed in different contexts: some in the cloud, others at the “edge” on local devices, others still on-premises—each tied to different firms and business strategies.

This technological diversity is mirrored in the sheer variety of AI applications. It makes little analytical sense to treat large language models (LLMs) for text generation as if they were part of the same service as computer vision systems for medical imaging, or to lump autonomous drones together with self-driving cars simply because both rely on AI. Radiology tools that analyse X-rays do not compete with protein-folding models used in medical research. These technologies serve different users, address different needs, and are not substitutable. Collapsing them all under the label “AI” is no more helpful than speaking of “food markets” or “technology markets”—a generality that obscures more than it clarifies.

It would be similarly misguided—and premature—to include a narrower group of generative-AI services (essentially seeking to capture offerings like ChatGPT, Claude, and Gemini) in the list of core platform services. The generative-AI space remains in its infancy and does not appear to have reached what Teece (1986) refers to as the paradigmatic stage of development. In other words, it is still unclear what the primary types of AI-based services will be in the years ahead. By arbitrarily capturing  today’s most successful generative-AI services—and potentially applying the DMA’s rigid rules to them—policymakers risk significantly impeding the development of these services by preventing their developers from exploring different features and platform architectures.

Designating AI as a single core platform service would therefore be a mistake. It risks generating inaccurate assessments of market power by either underestimating or overestimating concentration. If the relevant market is artificially broadened to include non-substitutable products, a firm’s dominance in a particular niche may be underestimated. Conversely, the narrative that AI is simultaneously vast and yet dangerously concentrated often rests more on technological anxiety and enforcement bias than on sound economic analysis.

Worse still, lumping diverse AI systems into one regulatory framework could distort innovation by implicitly favouring some technologies over others. Rules that are easier to apply to deterministic, “controllable” systems might inadvertently disadvantage more open-ended and creative forms of AI. And because this monolithic view obscures the real competitive dynamics of these nascent markets, it risks making authorities blind to emerging threats, new entry, or potential consumer benefits.

It must also be noted that competition, new entry, and technological advance are the hallmarks of every level of the AI stack. And the market leaders, such as they are, can hardly be characterized as “gatekeepers”. Moreover, it is startups, rather than incumbents, that have taken an early lead in generative AI (and in Web 2.0 before it).

A better approach would be to avoid blanket designations altogether and instead pursue a principled, case-by-case analysis of competition in the AI ecosystem through EU competition law. This means consistently asking basic but indispensable questions: who are the consumers?; what exactly is the product or service?; and to what extent is the AI in question substitutable either for human intelligence or for non-AI technologies? Only by working through these questions can enforcers arrive at meaningful market definitions and avoid abstract generalizations that mislead more than they help.

Designating “AI” as a core platform service would amount to an erroneous abstraction. AI is not one thing but a constantly evolving bundle of technologies with many uses and many implications. A more granular, evidence-based approach is essential if the Commission wishes to safeguard competition, protect consumers, and promote innovation without causing unnecessary harm.

Do you have any comments or observations on the designation process?

The DMA’s designation process exhibits procedural problems that may raise concerns under EU administrative-law principles, potentially including: an asymmetric application of evidentiary standards, inconsistencies in the application of qualitative criteria across different designation decisions, and insufficient procedural safeguards.

First, Art. 3(5) provides that undertakings may present “sufficiently substantiated arguments” to demonstrate that a core platform service should not be designated despite meeting quantitative thresholds. But it also appears to impose a heightened standard on evaluation of these arguments, allowing the Commission to reject them as insufficiently substantiated “because they do not manifestly call into question the presumptions under Article 3(2)”. This standard appears to create different evidentiary burdens for deviating from the presumptions in 3(2).

For Commission-initiated investigations under Article 3(8):

• The Commission may consider qualitative factors such as potential network effects, switching costs, and ecosystem integration;
• The standard appears to be whether there is sufficient indication of gatekeeper characteristics, regardless of the 3(2) criteria; and
• Recital 23 suggests that a forward-looking assessment is appropriate.

Yet for rebuttals under Article 3(5):

• Companies must meet the “manifestly call into question” threshold;
• The same qualitative factors that support Commission designation may be insufficient for rebuttal; and
• It is unclear whether forward-looking assessments could ever “manifestly call into question” the presumptions, which are based on backward-looking determinations.

The Commission’s approach to iPadOS and iMessage illustrates the potential inconsistencies in applying such qualitative criteria. In the iPadOS designation decision, despite falling below certain quantitative thresholds, the Commission initiated proceedings to assess whether iPadOS should be designated based on qualitative factors, including ecosystem effects and business-user importance. But in iMessage, the Commission decided not to designate the service, considering factors such as the relative number of users compared to competing services and the limited importance to business users. These contrasting outcomes raise questions regarding when qualitative factors override quantitative thresholds, what constitutes sufficient evidence to “manifestly call into question” presumptions, and whether similar analytical frameworks apply to Commission investigations and company rebuttals.

Unpredictable designation criteria create several inefficiencies. To start, they create compliance uncertainty. Companies must prepare for potential designation without clear guidance on likelihood, leading to potentially wasteful overcompliance or risky underpreparation. Second, they could harm innovation, with companies seeking to avoid features or growth that might trigger unclear thresholds. Finally, arbitrary designation decisions may advantage some competitors over others without any competitive justification, creating undesirable market distortions.

Perhaps more importantly, the principle of legal certainty, established in cases such as Case C-776/23 P Spain v. Commission, “requires that rules of law be clear and precise and predictable in their effect, so that interested parties can ascertain their position in situations and legal relationships governed by EU law and take steps accordingly”. The current interpretation of Article 3(5) raises concerns regarding the precision and predictability requirements of the Court of Justice’s case law on legal certainty. The “manifestly call into question” standard lacks detailed guidance; the interaction between qualitative and quantitative criteria remains unclear; and companies are often unable to predict with reasonable certainty how their arguments will be evaluated.

Furthermore, the DMA establishes a presumption-based system in which quantitative thresholds create rebuttable presumptions of gatekeeper status. But the application of these reveals tensions. In particular, the same factors (network effects, switching costs, ecosystem integration) appear to be sufficient for the Commission to initiate designation proceedings, yet insufficient for companies to rebut designation presumptions. This asymmetry may violate the general EU principle of equal treatment, which requires that comparable situations not be treated differently and different situations not be treated the same, unless objectively justified. It may also violate the principle of equality of arms and due process laid down in Art. 6 of the European Convention of Human Rights, which requires that each party has a reasonable opportunity to present its case and influence the outcome of a decision without a substantial disadvantage.

The processes by which these determinations are made may also violate the requirements under Art. 41 of the Charter of Fundamental Rights. Art. 41 CFR guarantees the right to good administration, including the right to be heard before adverse measures, the right of access to one’s file, and the obligation of administration to give reasons for decisions. Current DMA procedures raise potential concerns about each element. On the right to be heard, response windows can be too short for complex economic work, opportunities for oral presentation are limited, and there is no clear ability to reply to third-party submissions. On access to the file, the scope of access to the Commission’s economic analysis is uncertain, confidentiality claims can constrain meaningful review, and there is no established procedure comparable to merger control. Finally, as to the duty to give reasons, the detail provided in designation decisions varies, replies to company arguments can be cursory, and the underlying economic reasoning is not always fully articulated.

By contrast, Regulation 139/2004 on the control of concentrations provides for pre-notification meetings (Art. 4), multiple state-of-play meetings (Art. 18), access to file procedures (Art. 18), and the right to an oral hearing (Art. 14 of the Implementing Regulation). Likewise, antitrust proceedings (under Regulation 1/2003) include a detailed statement of objections, full access to the file, minus confidential information (Art. 27), oral hearing before an independent hearing officer, and peer-review procedures (Art. 14).

Despite entailing more severe economic consequences, the DMA’s procedures are, at the very least, less developed than these established frameworks, and may fall short on several dimensions. This is particularly true given the DMA’s penalties of up to 20% of worldwide turnover and potential breakups. Under CJEU jurisprudence (e.g., Intel v Commission, Case C-413/14 P), sanctions of this magnitude arguably require procedural safeguards proportionate to their severity. Likewise, the European Court of Human Rights’ Engel criteria suggest that administrative proceedings with such severe penalties may require criminal-law-level protections.

Obligations

Do you have any comments or observations on the current list of obligations that gatekeepers have to respect?

The DMA is now the cornerstone of the EU’s regulatory framework for digital platforms, imposing a detailed list of obligations on designated gatekeepers under Articles 5 to 7, 11, 14, and 15. These obligations were promoted as clear, self-executing “dos and don’ts”, capable of ensuring fairness and contestability in a fast-moving digital economy. But two structural problems immediately stand out. First, the obligations are badly drafted: their language is often vague and cumbersome. This produces legal uncertainty, raises compliance costs, and undermines the very promise of swift and effective enforcement. Second, the obligations are uncarefully drafted: rather than seamlessly complementing the EU’s broader regulatory architecture, they risk clashing with existing instruments, such as data protection, fundamental rights, and competition law. The drafting problems relate to the overall architecture of DMA obligations, as well as to individual duties and responsibilities.

General

The problem of poor drafting becomes immediately visible in Recital 65 DMA, which stresses that the obligations’ effectiveness depends on them being clearly defined and circumscribed “whilst fully complying with applicable law”. That said, the recital then limits this clarity requirement to obligations that are “susceptible of being further specified” and, in the event of circumvention, to all obligations. In practice, this means that only the obligations under Article 6 are open to specification by the Commission, while those under Article 5 and Article 7 are assumed to be sufficiently clear and self-executing.

This distinction is highly questionable. The obligations under both Articles 5 and 6 are drafted in cumbersome language that raises complex compliance questions. The relative clarity of Article 5 obligations cannot plausibly be explained by the text of the law. Nor can it be explained by enforcement experience. To be sure, some Article 5 obligations have close antecedents in competition law (e.g., Article 5(4) on most-favoured-nation clauses). Others, such as Article 5(2) on the cross-use of data, are informed by a more nuanced and evolving body of case law, with the CJEU taking a cautious approach to data sharing across services. Still others, such as Article 5(7) on disintermediation in payment systems and browsers, have little precedent. By contrast, some Article 6 obligations, notably Article 6(5) on self-preferencing, are more grounded in enforcement practice, the merits of such practice notwithstanding.

Against this background, there is no coherent reason to allow the further specification of Article 6 obligations while treating Article 5 obligations as self-executing. In fact, this asymmetry creates perverse incentives. Because Article 5 obligations may only be clarified in the event of circumvention, a gatekeeper uncertain of its compliance obligations may be tempted to “test the limits” of the law, effectively triggering Commission clarification through violation. This is wasteful from the perspective of administrative efficiency and detrimental to legal certainty. A more consistent approach would be to recognise that both Articles 5 and 6 obligations may need further specification, and to require the Commission to provide clarity—within the confines of the law—proactively rather than reactively.

Individual obligations

A central problem with the DMA’s obligations is that they have been drafted without sufficient regard for the tradeoffs they may create. One striking illustration is Article 6(7), which requires gatekeepers to provide business users with interoperability for the software and hardware features of their core platform services. Crucially, such access must be granted on an equal footing with the interoperability that exists internally among the gatekeeper’s own services.

Interoperability has long been promoted as a “super tool” to promote contestability in markets prone to tipping (Scott Morton et al., 2023). But it is also well known that interoperability is not costless. It can entail a tradeoff between competition and noncompetition values—most notably system integrity and security. In the digital space, opening access for one actor often means opening access for all actors; selective interoperability is practically impossible. Thus, obligations to interoperate may generate vulnerabilities that extend far beyond the narrow competition question at hand (Kerber & Schweitzer, 2017).

Recent enforcement practice illustrates the magnitude of this problem. In a public communication, the Commission revealed that business users have invoked Article 6(7) to request access to Apple’s Just-In-Time Compiler (JIT) engine in iOS, which is a core component of all major browser engines. Allowing third parties to inject and execute arbitrary code in such a sensitive area has been described by the cybersecurity community as a “major security vulnerability” (Kohlenberger, 2025). Such concerns cannot be dismissed as self-serving rhetoric from gatekeepers. They raise real risks of exploitation at scale, with implications for consumer protection, data security, and even national security.

The problem is not that interoperability is uniformly undesirable; rather, it is that the DMA’s obligation is drafted without recognising the tradeoffs involved. A rule that purports to promote contestability may, if enforced without regard for security, generate systemic harm. This is a textbook example of uncareful drafting: an obligation that may be well meaning in competition terms, but that fails to account for the broader legal and policy ecosystem into which it must fit.

In fact, some of the DMA’s interoperability obligations may fail even from a competition perspective. A case in point is Article 7, where gatekeeper providers of “number-independent interpersonal communications services” must interoperate their products with those of non-gatekeepers. This obligation may enable consumers using gatekeeper services to communicate with consumers using non-gatekeeper services. In that case, it is questionable why consumers of gatekeeper services would want to actually switch to other services, considering that a basic level of communication is enabled (Hovenkamp, 2023).

Interoperability obligations are but one example of the DMA’s uncareful drafting. Under Article 14, the DMA introduces an extended merger-reporting obligation for gatekeepers. This obligation appears to require notification where the “merging entities or the target of concentration provide core platform services or any other services in the digital sector or enable the collection of data”. Does the term “merging parties” not already incorporate the target of an acquisition, as the merging parties comprise the buyer and the target? Furthermore, why separate “core platform services” from “any other services in the digital sector”—considering it is virtually impossible for a core platform service to be unrelated to the digital sector (the regulation is called the “Digital” Markets Act, after all)? Also, how should one interpret whether a target “enables the collection of data”? Given that nearly all businesses in the digital economy operate with data, this provision risks becoming excessively broad.

This matters because badly written rules not only impair legal certainty, but also waste valuable enforcement resources.

Clashes with EU law

The DMA’s obligations have been drafted without sufficient attention to their interaction with the wider body of EU law. Formally, the regulation insists that it applies “without prejudice” to other instruments. In practice, however, this clause is something of a myth (Bania, 2023). The DMA is likely to generate frictions across multiple domains: potential double (or even triple) jeopardy under competition law (Robertson, 2024), tensions with guarantees of fundamental rights (Barczentewicz, 2022), and broader inconsistencies with parallel EU legislation. Even when head-on conflict does not arise, unintended consequences abound.

An example is the interaction between the DMA and the Data Act, both of which contain data-sharing rules. Read together, these measures contribute to what might be described as a policy of data immobility (Unekbas, 2023). The Data Act explicitly excludes gatekeepers from benefitting as recipients of data sharing. It is thus unclear whether a gatekeeper could ever benefit from data portability under Article 6 DMA. The Commission, exercising its power of specification, might conclude that transfers of data among large gatekeepers are inconsistent with the legislation’s objectives, particularly since the Data Act explicitly states that gatekeepers do not need further data access. Even where transfers are permitted, the Data Act imposes a requirement that they be made on fair, reasonable, and non-discriminatory terms, including a prohibition on favourable treatment of affiliated enterprises. This reduces the attractiveness of intrafirm data transfers, effectively constraining data flows even within the same corporate group.

Furthermore, the DMA itself restricts data use through Article 5(2), which prohibits gatekeepers from combining personal data across services without GDPR-compliant consent. Yet as gatekeepers are typically dominant undertakings, questions arise as to whether such consent can ever be regarded as “genuine”, given the CJEU’s view on the imbalance of power between users and gatekeepers. Recent developments suggest that even contractual relationships may no longer constitute a reliable lawful basis for data processing in such contexts.

Taken together, these overlapping rules mean that large technology firms face increasingly narrow and unpredictable avenues for lawful data acquisition. They cannot rely with certainty on data sharing, intrafirm data relocation, contractual ties, or even user consent. The DMA was meant to provide clarity and certainty; instead, it risks producing the opposite by entangling gatekeepers in a web of conflicting obligations that may undermine legal certainty across the EU’s digital policy framework.

Do you have any other comments in relation to the DMA obligations?

DMA obligations warrant two additional points of consideration that relate to enforcement and flexibility.

Enforcing the DMA obligations

The DMA was conceived as a pragmatic response to the perceived failures of traditional competition enforcement in digital markets. The rationale was clear: competition law was too slow, too resource intensive, and too ineffective to address entrenched gatekeeper power (Monti, 2021). The DMA was presented as a superior alternative. It consisted of a set of ex-ante obligations that would cut through procedural delays and deliver quick, effective remedies.

In practice, however, this promise has not materialised. The Commission’s first enforcement actions reveal lengthy proceedings, protracted compliance discussions, and frequent recourse to parallel competition-law investigations. Rather than largely replacing competition enforcement in the digital arena, the DMA currently functions as a weak complement, far from the “sector-specific competition law” it was supposed to embody (Petit, 2021).

A central reason lies in the non-self-executing nature of the obligations. Contrary to the legislative narrative, Articles 5–7, 11, 14, and 15 do not provide clearcut, automatic rules. Their language is often difficult to interpret. The recitals sometimes clarify, but at other times inject additional uncertainty. This undermines the very clarity and speed that justified the DMA in the first place. A legislative simplification of the obligations would therefore be welcome (Boscheck, 2024).

The challenge is compounded by the fact that many obligations require continuous specification in light of dynamic market realities. Digital markets evolve rapidly, and static “dos and don’ts” are inevitably ambiguous in practice. Specification proceedings, such as those launched against Apple regarding its App Store rules, demonstrate the value of an iterative process (Commission, 2024). These procedures not only help gatekeepers understand what is required of them but also enable the Commission to tailor obligations to real-world business models. Their systematic use should be encouraged.

Finally, enforcement is hampered by the DMA’s uneasy relationship with economic analysis. Legislators deliberately tried to sideline economics in the DMA’s design, excluding efficiency defences and relying on crude quantitative thresholds for designation. But as Fletcher and others remind us, economic insights remain indispensable (Fletcher et al., 2024). The obligations may look formalistic, but their effects are deeply economic: they reshape incentives, alter business models, and interact with network effects and data feedback loops. Ignoring these realities risks blunt, even counterproductive, enforcement. Economic analysis can and should play a constructive role, especially in diagnosing outcomes, monitoring effects, and informing specification proceedings. It is only by reintegrating economics into enforcement that the DMA can deliver on its initial promise of effective intervention.

Flexibility in the DMA obligations

A second promise of the DMA obligations was their supposed flexibility. Many emphasised that—unlike competition law, whose enforcement often arrived after the market had already “moved on”—the DMA would be nimble, capable of addressing fast-moving dynamics in digital markets. The regulation was presented as a “future-proof” tool, a law that could adapt quickly to new forms of gatekeeper conduct.

It is unclear whether the DMA truly meets the ambition. The fundamental dilemma of all technology regulation applies here: the law must aspire to regulate the future, but it can only be drafted with reference to the past. The DMA’s obligations draw heavily from prior antitrust cases (Vezzoso, 2024). In practice, the law “runs forward by looking backward”.

This is problematic in at least two respects. First, conduct prohibited by reference to past case law is not necessarily economically unsound today (exclusive dealing comes to mind). Gatekeepers may need to engage in similar practices to maintain ecosystem value or to support complementary innovation. Second, digital markets evolve rapidly. A practice that appeared abusive yesterday may be competitively irrelevant today, while novel sources of market power may not be captured at all. In this sense, the DMA risks being rigid and, over time, obsolete.

The regulation does contain mechanisms designed to provide flexibility (Witt, 2023). Article 12 allows the Commission to expand or update obligations via delegated acts following a market investigation. Article 13 introduces an “anti-circumvention” rule to prevent gatekeepers from exploiting loopholes. The Commission can also recommend new core platform services for designation. These are important tools, but they are not costless.

Market investigations preceding delegated acts require substantial resources and time, undermining the promised speed of intervention. Anti-circumvention risks degenerating into a “whack-a-mole challenge”: every time one gatekeeper strategy is prohibited, another may emerge, especially considering the sheer volume of complaints from business users (Uwe-Franck & Peitz, 2024). For an already-overstretched (some say “grossly understaffed”) authority charged with enforcing the DMA alongside antitrust and foreign-subsidy rules, such demands are daunting (Comte, 2025).

The real test is whether the Commission can deploy these tools pragmatically and prioritise interventions that safeguard contestability without stifling innovation. Flexibility is not just about having legal mechanisms to adapt; it is about using them judiciously, recognising when certain practices may enhance rather than hinder welfare, and ensuring that scarce enforcement resources are targeted where they matter most. Only then can the DMA avoid the fate of becoming yet another slow, rigid, backward-looking instrument in a forward-moving digital economy.

Enforcement

Do you have any comments or observations on the tools available to the Commission for enforcing the DMA?

The DMA was presented as a harmonising regulation grounded on Article 114 TFEU. Both the co-legislators and the Commission stressed that the alternative to a uniform EU framework was not “no regulation”, but rather the proliferation of 27 divergent national regimes. The DMA’s value proposition thus rests on its promise of legal certainty and a level regulatory playing field across the European Union.

Whether the DMA can deliver on this promise is open to question. The harmonisation clause in Article 1(6) DMA makes clear that the regulation does not pre-empt all national measures. It allows member states to apply national competition rules to prohibit unilateral conduct insofar as they are applied to undertakings other than gatekeepers, or to impose further obligations on gatekeepers within the meaning of the DMA.

This formulation creates two clear avenues for fragmentation. A narrow reading suggests that, so long as national rules are formally directed at undertakings other than designated gatekeepers, they remain untouched by the DMA. A broader reading permits member states to impose additional obligations on gatekeepers, so long as they do not directly contradict the DMA’s own obligations (Lamadrid & Fernandez, 2021). The implications are profound. The effect may be to preserve a wide margin for national authorities to regulate conduct that overlaps substantially with the DMA. In practice, this risks differential treatment of similar behaviour across the EU, undermining the DMA’s central harmonising function (van den Boom, 2023).

The German Act Against Restraints on Competition, with its amendments targeting undertakings of “paramount significance across markets”, is a case in point. Both the amendment’s aims and its list of prohibited practices are functionally equivalent to the DMA. The primary differences are that the German rule regards the list of prohibited practices as exhaustive, and the rule does not consider the practices at stake to be per-se prohibited; rather, it introduces a reversal of the burden of proof, allowing firms to provide objective justifications for their conduct, which is not possible under the DMA. But fundamentally, the German regime targets gatekeepers in all but name, raising the prospect of parallel enforcement with different standards (Colangelo, 2022).

Practice confirms that this is not a hypothetical concern. The Bundeskartellamt has initiated proceedings against Google and Meta concerning data collection and processing, and against Apple for alleged self-preferencing. These are issues already regulated under the DMA. The coexistence of national and EU enforcement in such cases risks precisely the fragmentation that the DMA was supposed to prevent (Barczentewicz, 2023).

A further concern is that careless enforcement of the DMA may elevate the risk of double or even triple jeopardy, in violation of the ne bis in idem principle (Cappai & Colangelo, 2021). The DMA is, in essence, a form of sector-specific competition law, targeted at large digital platforms. As such, it overlaps with traditional antitrust enforcement and, in some contexts, with parallel regulatory regimes. Unless carefully coordinated, this overlap can lead to multiple proceedings and multiple fines for the same conduct.

The Court of Justice has recently clarified the scope of ne bis in idem in bpost and Nordzucker. The Court held that the principle is a fundamental right that must be respected, although not in absolute terms. Restrictions may be permissible where they meet the requirements of legality, proportionality, and effectiveness: they must be grounded in law, not go beyond what is necessary, and be sufficiently coordinated to achieve legitimate regulatory aims. In practice, this means that the same conduct may fall under different regimes, but only if enforcement is properly structured and coordinated.

The DMA and the ECN+ Regulation contain provisions for cooperation between the Commission and national authorities. But whether these mechanisms are sufficiently effective in practice remains an empirical question. If coordination fails, the risk is that gatekeepers will be punished multiple times for the same underlying conduct, raising not just concerns of fairness but potential breaches of fundamental rights—thus elevating the problem into a matter of constitutional significance under EU law.

Do you have any comments in relation to the enforcement to the DMA?

This question is beset by a broader conceptual problem; it is unclear what the success metrics of DMA enforcement even are. In theory, the DMA is not outcome-driven: it does not prescribe a specific market structure or design. Instead, it claims to create opportunities for competitors and complementors on gatekeeper platforms by making markets “fair and contestable”.

This, however, raises the question of how fairness and contestability should be measured, or what they even mean. When is a core platform service truly “fair and contestable”? If the answer is simply that it is so once a gatekeeper complies with the DMA, then fairness and contestability collapse into tautologies. They become self-referential concepts, offering no real benchmarks. This is problematic because (i) the DMA’s obligations are often ambiguous, (ii) a gatekeeper might satisfy the letter of the law while sidestepping its intended “spirit”, and (iii) without clear goals and measurable benchmarks for success, it may be impossible—or, at least, exceedingly difficult—for either the Commission or the gatekeepers to know whether DMA compliance has been achieved. The result is likely to be confusion, costly litigation, and diminished effectiveness of the law.

Clear and objective metrics are therefore indispensable. They are needed not only to give concrete meaning to the DMA’s “spirit”, but also to determine whether the regulation meets its stated objectives. The Commission initially sought to place this burden on gatekeepers, requiring them to explain how and why changes to their core platform services complied with the DMA (Colangelo & Ribera, 2025). The volume of infringement decisions issued since the law’s entry into force suggests that this approach has failed. A more thoughtful and detailed framework is evidently needed.

The first step must be to unpack what “fairness and contestability” actually mean in the DMA. This should help to clarify both the objectives they are meant to serve and the criteria by which success or failure of enforcement can be assessed.

“Contestability” in the DMA seems to be defined as promoting “potential competition” by eliminating barriers to entry and expansion, thereby promoting existing rivals and fostering new market entries. For example, Article 6(4) DMA seeks to boost potential entries at the downstream and upstream level regarding app distribution by compelling gatekeepers to allow and technically enable third-party app stores and apps to interoperate with their operating systems.

“Fairness” is primarily linked to addressing “conflicts of interest” between gatekeepers and business users, particularly concerning value appropriation and conditions of access and competition. The fairness objective is likely redistributive in nature: Because gatekeepers have “unfairly” appropriated monopoly rents from the value that business users create on their platforms, the DMA aims to correct this by redistributing those rents to business users or competitors (Colangelo and Ribera, 2025). The assumption is that a severe power imbalance exists between gatekeepers, on the one hand, and businesses, competitors, and consumers, on the other, which has misaligned each actor’s contribution to the core platform service with the benefits they derive from it (Radic, Manne, & Auer, 2025).

The problem is that, even if we accept that “fairness” and “contestability” can be defined in theory, that does not tell us whether they have been advanced in practice. As vague and malleable standards, they provide no clear benchmark against which success can be measured.

For example, Art. 6(4)—on interoperability with and the installation of third-party applications—appears aimed at promoting potential competition. In theory, compliance with the provision could be measured by assessing the persistence of barriers to entry and expansion. But assessing compliance in this way would be problematic, as some barriers to entry—such as the screening of apps and app stores—protect the platform’s integrity and overall quality (Barczentewicz, 2022, July), and could therefore undermine other goals pursued by the DMA. In other words: success by one yardstick could mean failure by another.

Moreover, such an assessment lacks limiting principles. If there is no cost associated with seeking entry into the gatekeepers’ platform, self-interested business users—i.e., all business users—can always claim that entry is not sufficiently free, easy, or that there is not enough potential competition. Consumers’ revealed preferences—e.g., continued use of curated app stores with strong safeguards—risk being dismissed as evidence of platform obstruction or consumer inertia.

This makes Art. 6(4), and similar provisions aimed at boosting potential competition—such as Art. 6(9) on data portability or Art. 6(7) on interoperability with gatekeepers’ hardware and software—particularly ill-suited for interpretation by reference to their overarching objective. The mere fact that third-party app stores or sideloaded apps exist does not reveal whether they exert meaningful competitive pressure, nor whether their absence reflects gatekeeper obstruction, or simply a lack of consumer demand.

By the same token, if data portability or interoperability fail to boost rivals’ market position, the Commission or disappointed competitors can always claim that the gatekeeper did not provide “effective” access (Arts 6(7) and 6(9) require effective interoperability and data portability, respectively). But what does “effectiveness” mean in this context? Is interoperability “effective” if it is merely possible, or only if it actually produces increased entry? The former could be dismissed as falling short of the spirit of the law; the latter would require the Commission to continually redefine that spirit in line with its shifting expectations of what constitutes an acceptable level of potential competition. In either case, compliance becomes a moving target: the gatekeeper can never demonstrate success conclusively, while rivals can always claim that the regulatory experiment has not gone far enough.

Then there is the question of assessing the counterfactual. How is the Commission to determine whether competition is more vigorous than it would have been absent these provisions, or whether observed changes in competitive pressure are the product of gatekeepers’ conduct, rather than occurring despite it? As ICLE scholars have written elsewhere, regulations that pursue “potential competition” (or “contestability”) as an end in itself risk collapsing into tautology: every disappointed entrant can point to unrealized opportunities and blame the platform, while regulators are left to adjudicate inherently subjective claims (Radic, Manne, & Auer, 2025).

If third-party app stores fail to gain traction, gatekeepers may be accused of subtle obstruction; if they succeed, critics can argue that the DMA merely ratified an inevitability or that the gatekeeper could have done more. The same could be said of other provisions aimed at boosting contestability or potential competition. Either way, policymakers and firms are deprived of a clear benchmark for successful compliance. Worse still, the rule may impose real costs—such as by forcing platforms to dilute security and curation practices that consumers value—without producing any offsetting, measurable gains in competition.

Similar concerns arise with regard to DMA provisions aimed at addressing conflicts of interest. In the cases of, e.g., Arts. 5(4), 5(5), 5(6), 5(9), 5(10), 6(2), 6(10), & 6(13), success could  theoretically be measured by the extent to which a conflict of interest has been assuaged or eliminated. But here, again, determining what constitutes successful enforcement is elusive.

First, if a competitor or the Commission concludes that too few users have been steered to a rival site or payment system under, e.g., Art. 5(4), it can always allege that the gatekeeper obstructed steering or failed to promote it sufficiently. The Epic v. Apple litigation in the United States illustrates how such disputes can devolve into endless quarrels over degrees of access and facilitation (see Epic Games, Inc. v. Apple Inc., 2025).

Second, there is no objective standard for what counts as a “satisfactory” increase in traffic to competitors. Suppose a hypothetical gatekeeper provided flawless channels of communication between business users and customers, yet no users concluded contracts outside the platform. Would this constitute compliance with Art. 5(4)? On a formal reading, yes—but it is difficult to imagine the Commission accepting such an outcome as successful enforcement.

The deeper problem is that, despite the claim that the DMA is not outcome-driven, “perfect” compliance can rarely be judged without reference to outcomes. And because the law supplies no limiting principles or alternative benchmarks, the definition of compliance risks collapsing into whatever regulators—and, indirectly, rivals—want it to be. Simply put, conflicts of interest will be deemed resolved only when competitors are satisfied, an equilibrium that is inherently short lived, destined to be relitigated, and disconnected from consumers’ interests.

Third, and more generally, so-called “conflicts of interest” (i.e., when a company both operates and participates in a platform) are not inherently anticompetitive; in multisided markets, vertical integration and self-preferencing often reduce transaction costs, increase the incentives for investment, and improve the user experience (Manne & Radic, 2022; Manne, 2020; Manne & Bowman, 2021). Measuring success in terms of the “elimination” of such conflicts risks assuming the problem, rather than demonstrating it. Arguably, in the case of the DMA, this ship has already sailed. Thus, the fact that developers make greater use of steering or promotional links does not show that consumers are better off, nor that competition has intensified: it may just indicate that developers are shifting marketing costs onto gatekeepers’ ecosystems.

Moreover, “conflict of interest” provisions invite a one-way ratchet: if developers exercise their new rights aggressively, regulators will count this as success; if they do not, it can be portrayed as evidence that gatekeepers continue to undermine them in some more subtle way. As with Article 6(4), there is no falsifiable benchmark of success. Any outcome can be interpreted as proof that the DMA is necessary (and thus a “success” in the broadest terms), while the costs imposed on platforms and consumers—ranging from degraded user experience to higher prices—are downplayed or ignored. In this sense, the DMA risks becoming less a framework for evidence-based competition policy than a perpetual grievance mechanism for rivals who would prefer to compete through regulation, rather than in the marketplace.

Although we tentatively divide the DMA’s provisions into those aimed at fostering potential competition and those aimed at eliminating conflicts of interest, this taxonomy is more theoretical than practical. In reality, most provisions simultaneously affect both existing and potential competition, while also addressing conflicts of interest. This reflects the Commission’s view that contestability and fairness are “intertwined”, such that a single provision can be said to advance both objectives (Colangelo & Ribera, 2025).

The broader point is that, apart from the most straightforward provisions (e.g., Article 6(8)’s obligation to share advertising performance data or Article 6(2)’s prohibition on using third-party data to compete against business users), the metrics for judging the success of enforcement are ultimately circular. Because of the DMA’s self-referential nature—i.e., the absence of clear external benchmarks for assessing “fairness” and “contestability” (Radic, Manne & Auer, 2025)—the question will always remain whether more could have been done to eliminate conflicts of interest and lower barriers to entry. When are barriers sufficiently low, or conflicts of interest adequately mitigated? When have gatekeepers’ advantages been sufficiently dissipated? And what counts as an acceptable balance of bargaining power between gatekeepers and business users?

In this sense, both compliance and success become moving targets. Any of the tens of thousands of business users and competitors that are (by the DMA’s own admission) meant to benefit from it can always argue that they do not, or do not benefit enough. After all, the DMA’s built-in assumption of permanent imbalances of power between gatekeepers and all other stakeholders (Radic, Manne, & Auer, 2025) virtually guarantees an endless supply of such claims, precisely because there is little or no cost to making them—and because this, and other presumptions underlying the DMA, are essentially irrefutable.

A second step is therefore needed to determine whether enforcement of the DMA is succeeding and, ironically, it requires looking beyond the letter of the DMA. The Commission must clearly and unambiguously articulate the implicit objectives that underpin the regulation, which have thus far been obscured behind the rhetorical veil of “fairness and contestability”. The DMA must spell out its true aims, because meaningful assessment of compliance requires more than circular references to abstract principles. Without identifiable long-term policy goals embedded in each provision, enforcement risks devolving into an exercise in discretion without due process—where success is measured not against objective standards, but against regulatory expectations and the demands of the loudest rivals. This would produce an unstable and perpetually contested regime.

Scholars have attempted to distil these tacit objectives of the DMA in the broader pursuit of legal certainty. Colangelo & Ribera (2025) interpret the regulation as implicitly seeking to shape market structures by promoting consumer choice, platform openness, transparency, and the neutralization of competitive advantages. Radic, Manne, & Auer (2025) likewise underscore its redistributive thrust, emphasizing the levelling down of gatekeepers, the transfer of rents to complementors and competitors, and the facilitation of rivals.

The issue is that these implied metrics of “success” are likewise misguided and risk producing enforcement that merely levels down gatekeepers, distorts product design, and ultimately leaves consumers worse off. To be clear, it’s not that structure or process never matter, but the DMA elevates them into ends in themselves. “Openness”, “neutralization”, and “design transparency” are treated as proxies for regulatory success, even when they undermine scale and scope economies, degrade relevance and security (e.g., through weaker default protections or forced interoperability), or substitute one set of frictions (fragmentation, higher developer costs, lower ad effectiveness) for another.

When compliance is evidenced by more app stores, more toggles, or less personalization—rather than by better outcomes for users—enforcement drifts into undefined industrial policy. It equalizes rivals and standardizes designs, while leaving prices, quality, innovation, privacy, and security as afterthoughts. The effect resembles a “Harrison Bergeron” world of forced symmetry, where advantages are deliberately suppressed, rather than harnessed. That tilt increases false positives (blocking or reshaping efficiency-enhancing conduct) and chills investment in features that most users actually value.

In this way, the DMA’s tacit measures of success fall into the same trap as its explicit ones, chasing circular, abstract goals that risk harming consumers and distorting markets.

First, the DMA favours certain business models and product designs. Where multiple compliance options exist, it effectively prescribes outcomes, narrowing gatekeepers’ ability to choose solutions that might satisfy the rule at lower cost or with better consumer performance. For example, Article 5(2) prohibits combining data across CPSs, implicitly disfavouring behavioural advertising. In response to Meta’s program, the Commission indicated that gatekeepers must offer a less-personalized alternative (e.g., contextual ads). If that interpretation stands, the Commission should explain why this particular design mandate improves consumer welfare, rather than merely reshaping business models.

Second, “openness” is too often tallied by counting access points, rather than benefits to users. Article 6(4) aims to enable alternative app stores and apps on gatekeeper OSs. True success, however, should entail entry that yields lower prices, higher quality, or greater innovation for end users without degrading security or privacy. Treating “more app stores” as a sufficient metric mistakes means for ends.

Third, “neutralizing competitive advantages” risks penalizing competition on the merits. Article 6(2) bars gatekeepers from using non-public business-user data to compete with those users—even where the resulting products benefit consumers. Article 6(5) prohibits self-preferencing. The relevant question should, instead, be whether the conduct harms consumers (through higher prices, lower quality, or reduced innovation), not whether it disadvantages rivals. Success should therefore require evidence that intervention prevents consumer harm, not merely that it levels down a gatekeeper.

These structural/process metrics also suffer from deeper methodological flaws. Causality is hard to establish: more business users or new entries could reflect unrelated dynamics or simply the weakening of gatekeeper offerings, rather than welfare-enhancing competition (Radic, Manne & Auer, 2025). Conversely, if entry lags, a gatekeeper expands into adjacent markets, or maintains market share, it does not follow that enforcement failed. Unless the Commission equates “foul play” with growth or expansion—an interpretation it has not claimed—rival-focused indicators will routinely misfire.

Further, enforcement will often rely on gatekeeper-provided data (e.g., for personalized advertising), thereby compounding asymmetry. The DMA’s one-size-fits-all approach papers over differences across CPSs: switching costs, baseline contestability, and security/privacy tradeoffs vary widely (e.g., browsers vs. online social networking). Any serious assessment must account for these heterogeneities—and, ultimately, for consumer outcomes—rather than whether a checklist of structural changes has been ticked.

Regulation—no matter how ambitious—cannot remake underlying economic realities. To the extent the DMA’s implicit metrics prioritize market reconfiguration over actual consumer outcomes, they risk costly, distortionary, and self-defeating enforcement. In short, interventions should be justified by net benefits to consumers, not by rival-centric or design-prescriptive targets.

Effectiveness and Impact on Business Users and End Users of the DMA

Do you have any comments or observations on how the gatekeepers are demonstrating their effective compliance with the DMA?

There are several observations to be made based on gatekeeper’s compliance reports, dedicated websites, and workshops.

Compliance costs

First, the DMA involves high compliance costs, as well as resources diverted from other, potentially more productive endeavours. These costs cannot be ignored, as sound policymaking requires weighing the costs and the benefits of regulation, including the costs of administering the system and potential error costs (i.e., the costs of false positives and false negatives).

As pointed out by Barczentewicz (2025a, 2025b), the costs of DMA compliance for gatekeepers are substantial. The Commission initially projected compliance costs of roughly €10 million annually for all gatekeepers combined. In its compliance workshop earlier this year, however, Amazon reported its costs are “multiple orders of magnitude beyond that predicted amount”, while a Meta representative stated their costs are “a long way north of” third-party estimates of $10-20 million per year.

The allocation of gatekeeper personnel to DMA compliance is also considerable. According to Meta, the company has involved more than 11,000 employees and invested nearly 600,000 engineering hours. In its compliance workshop, Apple claimed its engineers have spent “hundreds of thousands of hours”, with thousands of employees involved in the effort to comply within highly “compressed timelines”. Similarly, Google assigned approximately 3,000 people to work full-time for two years on compliance for a single article of the DMA.

These expenses involve tradeoffs and opportunity costs. Indeed, gatekeepers have argued this massive expenditure is a “hidden tax” that diverts resources from innovation. As Amazon’s representatives explained during its workshop, money spent on compliance “takes away that work from other areas that you could be innovating and providing benefits to customers in Europe”.

Undefined concepts and conflicting interpretations

Second, compliance reports and dedicated websites consistently highlight a frustration with the DMA’s ambiguity. A significant part of gatekeepers’ communication is dedicated to the challenges of complying with a regulation they describe as ambiguous and lacking clear guidance from the Commission. Gatekeepers have consistently argued that DMA is characterized by undefined key concepts and conflicting interpretations. Apple pointedly observed during its workshop that when “no two people agree on what the DMA’s substantive obligations mean”, the resulting ambiguity undermines the rule of law.

In line with this, gatekeepers have repeatedly pleaded for clearer compliance guidance from the Commission. These requests have, however, largely been met with silence under the misguided philosophy that the entire burden of DMA compliance—including interpreting the terse statute—should be on the gatekeepers. This forces companies to navigate a “Kafkaesque regulatory environment”, where they can be fined for noncompliance without clear standards (Barczentewicz (2025a).

These costs are exacerbated by legal fragmentation. The DMA’s promise of a “single European rulebook” does not appear to have materialized. Companies like Amazon and Google reported facing parallel national enforcement actions, such as from Germany’s Federal Cartel Office, on matters squarely covered by the DMA, such as price filters (Colangelo, 2025). This fragmentation multiplies compliance costs and uncertainty.

Harms to EU users and businesses

Third, gatekeepers have argued in their workshops that the Commission’s DMA enforcement is “decontextualized from impact” and is causing tangible harm to European users and businesses. The Commission might view these arguments as self-interested, rather than as genuine concern for the public interest. Yet the two are not necessarily incompatible. Because gatekeepers have a personal stake in keeping their platforms competitive—through low prices, better user experience, and innovative products—their self-interest may, in fact, align with consumer welfare.

Accordingly, gatekeepers claim the DMA’s enforcement leads to “inferior digital services” for European consumers. In its workshop, for instance, Google mentioned “withholding of innovations from Europe” as a direct consequence of the DMA. Apple warned that being forced to rush compliance creates the risk of “premature solutions with bugs” and could force companies to “hit the pause button” on European innovation.

It is possible—even likely—that some of these claims are true. For example, Apple delayed the launch of Apple Intelligence in Europe for six months, and is now reportedly withholding new features, such as live translation, from the AirPods 3.

Significant privacy and security concerns

Fourth, another central theme of the compliance workshops is the Commission’s systematic dismissal of privacy and security concerns.

Apple stated that cybersecurity agencies were “nowhere to be found” in key decisions and that some third parties are exploiting interoperability requirements for data harvesting, requesting the ability to “read the contents from each and every message and email on the user’s device”. Amazon’s vetting process for data access found that more than 75% of applicants were from outside the EU, many appearing to be “data aggregators with opaque privacy policies”. Google presented data showing users are 50 times more likely to encounter malware off the Google Play Store, underscoring the real-world risks of sideloading mandates (Barczentewicz, 2025a; 2025b).

Technical difficulties

Fifth, Gatekeepers have also highlighted the technical difficulties of compliance. For example, Google explained that its use of frequency thresholding as a “state of the art” anonymization method for sharing search-query data is required by law, pushing back against competitors who claimed the method rendered the data useless. The Commission acknowledged the legal requirement for anonymization but stated it should not significantly reduce data quality, which might be a contradictory proposition.

In its workshop, Meta explained the limited data used for its “less personalised ads” (LPA) option, but noted that the Commission’s strict interpretation of “no data combination” risks imposes an “unviable business model” that is also largely useless to SMEs. In other words: complying with the Commission’s rigid reading of an already rigid regulation risks destroying precisely what makes the regulated companies successful—and what draws business users to them in the first place.

Similarly, Microsoft explained that its controversial “Recall” feature processes data entirely locally on a user’s PC for security. Despite this, the Commission is exploring data portability mandates for the feature, which Microsoft suggests raises serious security concerns.

Conclusion

While the specific technical challenges detailed by Google, Apple, Meta, and Microsoft in their respective compliance workshops demonstrate their efforts to balance compliance with core responsibilities like user security and business viability, they also reveal a fundamental impasse in the DMA’s implementation. The core issue is not simply technical difficulty, but a deep-seated risk that the Commission will systematically dismiss these efforts as self-serving attempts to undermine the regulation.

There is an uncomfortable truth at the heart of the DMA: Measures that enhance user privacy and security often align with the gatekeepers’ economic interests in maintaining a controlled “walled garden” ecosystem. An enforcement approach that remains blind to this duality of interest is likely to see only the self-interest and dismiss the genuine user benefits of such actions. The workshops suggest this is already happening, with the Commission engaging in a systematic dismissal of privacy and security concerns (Barczentewicz, 2025a; 2025b).

Do you have any concrete examples on how the DMA has positively and/or negatively affected you/your organisation?

We have observed a range of significant negative impacts across various digital services, affecting companies’ operational efficiency and users’ experiences. While the DMA aimed to foster competition, innovation, and consumer choice, our impression, supported by recent reports (Cennamo et al., 2025; Jebelli & Ledford, 2024), suggests that the implementation of its provisions has, in many instances, led to degraded services, substantial economic losses, increased security risks, and a noticeable slowdown in the rollout of new innovations within the European Union.

Degraded user experiences and service functionality

The DMA’s prohibitions on integrated services have created a more fragmented digital landscape, directly impacting the seamlessness of user experiences. For instance, the changes mandated by Article 6(5) of the DMA, which prohibits “non-discriminatory conditions”, have significantly altered online search services.

Google, as a designated gatekeeper, has removed the smooth integration of products such as Google Maps into search results. Previously, a search for a local business or restaurant would display results plotted on Google Maps, often with a dedicated “Maps” tab, allowing for immediate booking or directions. Users are now often forced to navigate multiple websites and apps for tasks that were once streamlined, increasing friction and requiring additional steps to access basic information. A study found that this specific change led to a 21% increase in searches for mapping services in the EU (Pape & Rossi, 2024). Meanwhile, competing map services have not experienced a notable uptick in traffic (Ibid), suggesting that users continue to prefer Google Maps—even if it takes them longer to get there.

In the accommodation sector, Google’s changes under Article 6(5) have made hotel offers less organized, clear, and intuitive. Beyond the evident qualitative impact on the consumer experience, the change has also reduced hotels’ ability to reach customers directly and diverted traffic toward intermediaries (Delgado, 2024). The DMA has therefore produced clear winners and losers—with potential beneficiaries such as Booking.com, itself a Gatekeeper, among those gaining from the shift.

Finally, the new regulatory framework has also introduced increased consumer confusion and complex choices. For example, Article 5(2) of the DMA requires users to navigate extensive pop-ups to confirm their preferences for service integration versus separate functionality. This change complicates previously simple tasks, such as setting default apps or browsers on devices, and forces users to scroll lengthy lists of options (Jebelli & Ledford, 2024).

Substantial economic losses for businesses

The DMA’s adverse effects could translate into measurable economic losses for businesses across the EU—not just gatekeepers (Cennamo et al., 2025). For example, potential revenue losses up to €114 billion are projected for firms in service sectors across the EU, corresponding to a loss of up to 0.64% of total turnover in the relevant sectors. This also translates to an estimated drop in revenue per worker across these service sectors by up to €1,122 per year.

According to Cennamo and coauthors, a major contributor to these losses stems from the impact on online-advertising services. Article 5(2) prohibits gatekeepers from combining and using personal data across core platform services for advertising without explicit user consent, overriding other legal justifications like legitimate interest. This restriction significantly lowers the effectiveness of personalized and targeted advertising due to the loss of valuable information signals. Studies indicate that personalized marketing can reduce customer acquisition costs by up to 50%, and personalized ads can be three times more valuable than non-personalized ones.

The shift from personalized to generic marketing can also reduce clickthrough rates, with some field experiments showing a drop from 25% to 12%. Following GDPR compliance, which is less stringent than the DMA’s requirements, web publishers experienced a 5.7% decrease in revenue per click and a 2% reduction in online sales revenues, alongside an 8% reduction in profits due to compliance costs. Marketing experts predict that smaller firms, in particular, will face higher customer acquisition costs because they lack the resources to efficiently target audiences without the platforms’ data-aggregation capabilities. This has also led to a potential forced pivot toward more intrusive, less personalized ads or subscription-based models for platforms.

Additionally, new compliance and cybersecurity costs arise from managing explicit user consents and potentially collecting first-party data, with some small businesses reporting spending between €1,000 and €50,000 on GDPR compliance alone, increasing their regulatory costs by 20-30%.

The retail sector also faces substantial impacts, with estimated losses of between €4.4 billion and €59 billion, or up to 1.1% of the sector’s total turnover. These losses stem from reduced ad-targeting efficiency, a decline in organic traffic due to the stronger presence of intermediaries in search results, less-efficient recommender systems on marketplaces, and the loss of useful integrations like maps and review systems.

The accommodation sector has been particularly hard hit due to its heavy reliance on digital-platform services. Changes in Google Search results to comply with the DMA have led to a 36% drop in direct bookings for hotels through Google Hotel Ads for hotels (Delgado, 2024). Reduced visibility of Free Business Listing, which previously offered a costless and efficient discovery tool, has further diminished booking opportunities. These changes have inadvertently increased the prominence of Online Travel Agencies (e.g., Booking.com), leading to higher distribution costs and greater dependency on third-party platforms for hotels.

Stifled innovation and delayed services

Most concerning is the way the DMA has begun to stifle innovation in Europe. Some have referred to this as a “digital curtain” that separates European citizens from innovative digital services available elsewhere (Jebelli & Ledford, 2024). Indeed, regulatory uncertainty caused by the DMA has led companies to delay launching new products and features in the EU, or even bypass the European market altogether (Auer, 2024).

Concrete examples include delays in the rollout of advanced AI products and new social-media platforms. Google Gemini (an advanced AI for enhanced search) and Meta’s Threads (a new social-media platform) faced months-long delays in Europe, with some users waiting half a year for access. Google’s Gemini-powered AI Overviews—which offers multi-step reasoning capabilities in search results and has benefited more than a billion global users—were not immediately available in Europe.

Similarly, Meta announced in July 2024 that its new multimodal AI capabilities, which can interpret combinations of video, audio, images, and text for products like smartphones and smart glasses, would not be launched in the European Union due to regulatory uncertainties, despite being available in the UK and Brazil.

Apple, for reasons tied to Article 6(7)—which mandates interoperability with rival services—has also announced it will not roll out new AI features like Phone Mirroring, SharePlay Screen Sharing enhancements, and Apple Intelligence for its devices in the EU. These features, designed to enhance productivity and personalization, will leave EU consumers behind their global counterparts, affecting daily activities and EU’s users familiarity with cutting-edge technologies.

Conclusion

While the DMA was envisioned to promote a more competitive and fair digital market, its current implementation has imposed significant costs and functional disruptions on gatekeepers’ core platform services and, by extension, on businesses and consumers across the EU. We observe a clear tradeoff where the pursuit of market contestability has come at the expense of platform efficiency, user experience, economic vitality, and technological innovation. The evidence points to a critical need for a more nuanced and adaptable regulatory framework that carefully weighs these costs against the promised benefits, ensuring that European citizens and businesses are not disadvantaged in the global digital economy.

Do you have any comments in relation to the impact and effectiveness of the DMA?

Effectiveness is not the same as “success”. If a regulation is poorly designed, “effective compliance” or “effectiveness” may look more like failure than achievement. An act can be “effective” in a narrow sense, while still causing harm. With the DMA, compliance does not erase the significant costs and unintended consequences that cloud its supposed success. Similarly, “impact” can be positive or negative, intended or unintended, and either aligned with the DMA’s objectives or entirely external to them.

With that said, the Commission’s recently published second annual enforcement report on the DMA focused, as expected, on what has been achieved. The report celebrates concrete and measurable steps the Commission has taken to rein in “gatekeeper” platforms: investigations launched, compliance workshops held, and remedies imposed.

This is a selection of the seen effects: the visible indicators of action and the metrics that suggest movement (although not necessarily progress). A holistic analysis of the DMA’s impact and effectiveness must, however, also consider the “unseen” effects: the costs, tradeoffs, and unintended consequences that do not show up in press releases or enforcement statistics, but which may matter just as much.

The DMA’s achievements, according to the Commission

According to the Commission’s reports, the DMA’s visible achievements fall into three categories: designating new core platform services as “gatekeepers”; monitoring the implementation of substantive obligations; and addressing noncompliance with enforcement actions.

In its first year, six firms’ core platform services were designated as “gatekeepers” under the DMA’s terms—five American (Alphabet, Amazon, Apple, Meta, and Microsoft) and one Chinese (ByteDance). Last year saw two gatekeepers added to the list: the online-reservation platform Booking.com and Apple’s iPad OS.

On the monitoring front, the Commission initiated proceedings against Apple, alleging that it had not enabled “genuine” consumer choice (through choice screens) for selecting web browsers on iOS. It also scrutinized the integration and connectivity of Apple’s ecosystem (e.g., Apple Watch and iPhone) and initiated “specification proceedings” to demonstrate how Apple could remedy alleged deficits in interoperability. Regarding Alphabet and Microsoft, the Commission “corrected” the firms’ compliance with default settings and preinstalled apps on Android and Windows devices.

The Commission also initiated further enforcement action in some cases. For example, it took issue with Meta’s “pay-or-consent” model, whereby Facebook users could continue using the social-media platform for free in exchange for agreeing to be served ads, or they could pay a monthly fee for an ad-free experience. In April 2025, the Commission found Meta in violation of the DMA for not giving users a third choice: one that uses less personal data for targeted advertising.

But these enforcement statistics are relevant largely for regulators’ end-of-year reports and periodic evaluations. The cases brought may or may not be misguided. They may or may not pass a cost-benefit analysis. In other words: the fact that an infringement decision was issued or that a compliance workshop was held says nothing about whether the DMA benefits the public. For this, one must look under the hood.

Unseen effects

Details of the DMA’s implementation paint a rosy picture of accomplishment. But there are at least four kinds of other less obvious effects that deserve closer scrutiny:

• The DMA has thus far had mixed effects for consumers and businesses;
• The DMA competes with competition law for scarce enforcement resources;
• The DMA may divert attention from matters that have a greater impact on the “fairness and contestability” of European digital markets; and
• The DMA may exacerbate problems in an already-inflamed transatlantic trade relationship.

Effects on consumers and businesses

Among the dubious effects that can be attributed to the DMA are that mandatory third-party app stores and sideloading mandates have enabled the proliferation of pornography apps on iOS. Moreover, features like Apple Intelligence and the AI Suite were delayed in the EU due to concerns about DMA compliance. Others like advanced screen mirroring and SharePlay have not been released at all for much the same reason.

To avoid “self-preferencing” accusations, Google has removed or degraded popular features from its search engine, such as quick flight lookups in search results. This came after rivals complained that the one-box Google Flights search widget was unfair. The change caused some to initially witness traffic drops of up to 30% after Google’s first tweaks. Google was not displaying direct links to flights and hotels, as this would presumably be “unfair” to hotel and flight intermediaries (Delgado, 2024). Like Apple, Google also reportedly held back certain AI features or products in Europe pending regulatory clarity. For example, AI overviews were launched in only eight EU member states, and after nine months of delay.

A recent report by the Chamber of Progress (Jebelli & Ledford, 2025) provides a more detailed qualitative overview of the degradation of services due to the DMA, concluding that “[all] consumers have to show for [the DMA] are second-class digital services lagging behind the rest of the world”. The report identified increased user friction, reduced search efficiency, confusing and tedious choices, more irrelevant ads, less private and secure services, and delayed or unavailable innovations.

These consequences, as well as the follow-on effect of higher costs imposed on mostly U.S. firms, will be felt by European companies. According to one study based on industry interviews (Suominen, 2022), increased regulatory costs could lead EU firms to spend an additional 5% on technology services. “The DMA and DSA alone could imply an immediate cost increase of €71 billion ($71 billion) on European companies, equivalent to 0.3 percent of EU GDP”. (Ibid). Of special note, much of this would be incurred by European SMEs—an amount “equivalent to some 40,000 European jobs (measured as revenue over employment)”, according to the report.

Another report (Cennamo et al., 2025) found potential revenue losses of up to €114 billion “for firms in service sectors across the EU from the loss of efficiency of the most widely used digital services platforms. This corresponds to a loss up to 0.64% of the total turnover of the sectors considered”. Especially hard hit would be the accommodation sector (revenue losses of between €1 billion and €14 billion, with annual lost revenue per-worker of up to €3,579) and the retail sector (which could lose between €4.4 billion and €59 billion in revenues, with annual lost revenue per-worker of up to €1,122).

More generally, a recent study found that the market perceives a 10-20% negative impact on businesses from ex-ante regulation (Arai, 2025). Indeed, “over 60 percent of micro and small European firms stated that a 5 percent technology cost increase would be much worse or worse than inflation, slowing demand, or supply chain backlogs”. (Suominen, 2022).

None of this was noted in the Commission’s annual enforcement report.

Competition for enforcement resources

One must also consider the cost—not in Euros, but in institutional bandwidth—of enforcement itself. The DMA is enforced primarily by the European Commission’s Directorate-General for Competition (DG COMP), the same body responsible for applying the EU’s competition-law framework. This matters because the resources poured into DMA monitoring and enforcement—technical teams, legal experts, data scientists, policy analysts—are resources not available to investigate potential abuses of dominance, cartels, or anticompetitive mergers. Enforcement doesn’t scale effortlessly and regulatory capacity is finite, especially for understaffed public authorities.

The irony here is striking: DG COMP has already made digital markets a top enforcement priority for traditional competition law. The 2024 Competition Policy Report stated that the Commission is pursuing “several cases in digital markets against large digital companies”. These include cases that completely overlap with the “achievements” claimed under the DMA.

The glaring example here is Apple’s App Store practices, which faced a €1.8 billion fine for “preventing [developers] from informing iOS users about alternative and cheaper subscription services”. This demonstrates that the Commission is deeply engaged with the digital space through competition enforcement, regardless of the DMA.

The point is not to present competition and DMA enforcement as substitutes; the legislative history and text clearly present the two mechanisms as complementary. It is also true, however, that the DMA was originally pitched by invoking an imaginary consensus (Radic, 2025) that traditional competition tools are too slow or unwieldy for fast-moving digital markets.

That may be true in some cases. But if it is the DMA’s primary justification, we should ask whether the DMA actually fixes a flaw in the system, or simply displaces it with a different, possibly less accountable model. If traditional enforcement is “unworkable” (and the Commission’s own enforcement record raises doubts), then perhaps the answer is to reform how competition law is applied, rather than building a parallel regime that consumes the same limited resources. The risk here is one of duplication with little additional value. Indeed, application of both the DMA and competition law could result in legal fragmentation by treating the same conduct differently.

Diversion from core issues and enforcement opportunity costs

The prevailing sense that the DMA constitutes a significant institutional accomplishment may also distort policy priorities. It is no secret that the Commission sees itself not just as a market regulator, but as a standard setter for the world. The DMA has been presented as a model for proactive tech regulation in action. There is satisfaction, perhaps even pride, in being the global frontrunner in digital rulemaking.

But this attention and energy may come at the expense of less headline-grabbing tasks. One such task is the completion and enforcement of the single common market—a project that remains unfinished decades after its launch. Despite harmonization efforts, many European firms still struggle to scale across borders. Barriers in taxation, licensing, procurement, and logistics persist. Funding ecosystems remain fragmented, particularly for startups and small and medium-sized enterprises (SMEs). When asked by the European Investment Bank what makes scaling so hard for European firms, SMEs answered: availability of skilled staff, fragmented business regulations, and access to funding (EIB, 2023). Perhaps the answers to the fairness, contestability, and competitiveness of European digital markets should be sought here.

Indeed, in the current climate, the DMA risks becoming a source of regulatory displacement. It channels focus onto “gatekeepers” like Google and Apple—powerful and photogenic targets—while deflecting attention from structural problems that make life harder for European entrepreneurs. To put it more bluntly, most European startups are not failing because of platform dominance. They are failing because of fragmented markets and shallow capital pools (Garicano, 2025; see also Draghi, 2024). Yet these root causes are harder to tackle, less media-friendly, and more politically complex.

Harming the transatlantic trade relationship

The DMA’s early enforcement unfolds against an unusually incendiary transatlantic backdrop. In Washington, EU digital rules are increasingly framed as trade-salient measures that disproportionately burden U.S. firms. The Office of the U.S. Trade Representative’s 2025 national trade estimate flags EU digital measures as barriers, and administration statements and aligned advocacy portray the DMA as an “anti-American” regime that could warrant tariff or Section 301 responses. That perception—whether it is right or wrong—matters for stability, and it has plainly entered the U.S. policy conversation.

In parallel, commentary from various observers has underscored that tariff threats are already being brandished as leverage in response to European digital regulation, with the DMA repeatedly cited. These signals elevate the risk that individual DMA actions will be read in Washington through a trade lens, not one of competition policy.

Recent Commission decisions amplify those optics. The Commission has stressed nationality-neutral application and has, in fact, designated non-U.S. firms (e.g., ByteDance and Booking.com) and declined designation where warranted (e.g., iMessage). Yet the most visible enforcement milestones to date—the first noncompliance decisions and fines—have centred on U.S. firms, which inevitably reverberates in the U.S. trade debate. In a charged trade environment, these headlines can be—and routinely are—read to support retaliatory measures.

Against this backdrop, it would be prudent to accompany enforcement with a calibrated de-escalation and transparency package that safeguards the regulation’s objectives while lowering the temperature on the trade front.

First, a time-limited pause on launching new noncompliance proceedings on novel or highly contestable issues—framed as a window for clarifying standards—would create space to publish administrable guidance on the respective roles of Articles 3(5) and 3(8) (including what “manifestly call into question” means in practice and how forward-looking qualitative factors are weighed). This pause could also be used to commit to prompt, detailed, and nonconfidential versions of designation and noncompliance decisions.

Second, sequencing remedies to minimize trade salience—prioritizing proportionate, behaviour-based fixes and supervisory commitments, reserving structural measures only for egregious and repeated noncompliance and, absent urgency, only after judicial review—would reduce the scope for mischaracterization of the act as industrial policy, while remaining squarely within the DMA’s remedial toolbox.

Third, a communications and transparency track—regular plain-language case notes on evidentiary reasoning, and a simple “neutrality dashboard” showing the origin and mix of designated and non-designated services—would make the regime’s even-handedness legible to trade audiences. Taken together, these steps could help convey a sense of nationality-neutral enforcement.

Conclusion

In conclusion, whether right or wrong, the DMA is now the law of the land. As the European Commission celebrates the second year of DMA enforcement, Europe should not lose sight of what is often unseen (especially in official reports): the effects on consumers and businesses, the opportunity costs within DG COMP, the distraction from deeper single-market integration, and the risks to transatlantic alignment.

Additional Comments and Attachments

Bibliography

Arai, K. (2024). Comparative impacts of ex-ante and ex-post regulation in digital markets: An event study analysis of stock-price reactions. Asian Journal of Law & Economics.

Auer, D. (2024, March 12). The broken promises of Europe’s digital regulation. Truth on the Market. https://truthonthemarket.com/2024/03/12/the-broken-promises-of-europes-digital-regulation.

Auer, D. & Zúñiga, M. (2025). AI partnerships and competition: Damned if you buy, damned if you don’t (ICLE white paper 2025-08-19). International Center for Law & Economics. https://laweconcenter.org/wp-content/uploads/2025/08/AI-partnerships-and-competition.pdf.

Bania, K. (2023). Fitting the Digital Markets Act in the existing legal framework: The myth of the “without prejudice” clause. European Competition Journal, 19, 116.

Barczentewicz, M. (2022, July). Minimizing privacy risks in regulating digital platforms: Interoperability in the EU DMA. CPI Antitrust Chronicle. https://ssrn.com/abstract=4190028.  

Barczentewicz, M. (2023). German Big Tech actions undermine the DMA (ICLE issue brief). International Center for Law & Economics. https://laweconcenter.org/wp-content/uploads/2023/06/MB-DMA-v-Germany.pdf.  

Barczentewicz, M. (2025a). EU DMA workshops: Google, Amazon, Apple, Meta, and Microsoft. EUtechReg. https://eutechreg.com/p/eu-dma-workshops-google-amazon-apple.

Barczentewicz, M. (2025b, July 8). The Digital Markets Act as an EU digital tax: When compliance costs dwarf regulatory estimates. Truth on the Market. https://truthonthemarket.com/2025/07/08/the-digital-markets-act-as-an-eu-digital-tax-when-compliance-costs-dwarf-regulatory-estimates.  

Barnett, J. M. (2024). The case against preemptive antitrust in the generative artificial intelligence ecosystem. In A. Abbott & T. Schrepel (Eds.), Artificial intelligence and competition policy. Concurrences.

Boscheck, R. (2024). The EU’s Digital Markets Act: Regulatory reform, relapse or reversal? Intereconomics: Review of European Economic Policy, 59(3), 154.

Bowman, S. & Manne, G. (2021, March 4). Self-Preferencing can be Good for Consumers. Truth on the Market. https://truthonthemarket.com/2021/03/04/platform-self-preferencing-can-be-good-for-consumers-and-even-competitors.

Cappai, M., & Colangelo, G. (2021). Taming digital gatekeepers: The “more regulatory approach” to antitrust law. Computer Law & Security Review, 41, 105559.

Cennamo, C., Kretschmer, T., Constantiou, I., & Garcés, E. (2025, June). Economic impact of the Digital Markets Act on European businesses and the European economy [Working paper].

Chilson, N. (2025, April). The vibrant AI competitive landscape. Abundance Institute. https://cdn.hub-abundance.institute/pdf/AbundanceInstitute_2025_EmergingTech_Publications_APR_TheVibrantAICompetitiveLandscape_Sans%20(1)%20(1).pdf.

Colangelo, G. (2022). The Digital Markets Act and EU antitrust enforcement: Double & triple jeopardy (ICLE white paper 2022-03-23). International Center for Law & Economics. https://laweconcenter.org/wp-content/uploads/2022/03/Giuseppe-Double-triple-jeopardy-final-draft-20220225.pdf.

Colangelo, G. (2025, June 11). Still haven’t found what the Bundeskartellamt is looking for: Thoughts on the German Amazon case. Truth on the Market. https://truthonthemarket.com/2025/06/11/still-havent-found-what-the-bundeskartellamt-is-looking-for-thoughts-on-the-german-amazon-case.

Colangelo, G., & Ribera Martinez, A. (2025). The metrics of the DMA’s success. European Journal of Risk Regulation. Advance online publication.

Comte, J. (2025, April 8). “Grossly understaffed” EU competition enforcers must prioritize, Guersent says. MLex Insight. https://www.mlex.com/mlex/articles/2322385.

Delgado, P. (2024, June 11). Who are the winners and losers of Google’s implementation of the DMA? Mirai. https://www.mirai.com/blog/who-are-the-winners-and-losers-of-googles-implementation-of-the-dma/.

Draghi, M., et al. (2024). A competitiveness strategy for Europe (The Draghi Report). European Commission. https://commission.europa.eu/topics/eu-competitiveness/draghi-report_en.

Epic Games Inc. v. Apple Inc., No. 4:20-cv-05640-YGR, 2025 WL 1508 (N.D. Cal. Apr. 30, 2025).

European Commission. (2024, September 19). Commission starts first proceedings to specify Apple’s interoperability obligations under the Digital Markets Act [Press release]. https://digital-markets-act.ec.europa.eu/commission-starts-first-proceedings-specify-apples-interoperability-obligations-under-digital-2024-09-19_en.  

European Investment Bank. (2023). EIB investment survey 2023. https://www.eib.org/attachments/lucalli/20230285_econ_eibis_2023_eu_en.pdf.

Fletcher, A., Cremer, J., Heidhues, P., Kimmelman, G., Monti, G., Podszun, R., Schnitzer, M., Scott Morton, F., & de Streel, A. (2024). The effective use of economics in the EU Digital Markets Act. Journal of Competition Law & Economics, 20(2), 1.

Franck, J., & Peitz, M. (2024). The Digital Markets Act and the whack-a-mole challenge. Common Market Law Review, 61(2), 299.

Garicano, L. (2025, May 21). How to build a capital markets union. Silicon Continent. https://www.siliconcontinent.com/p/how-to-build-a-capital-markets-union.  

Hovenkamp, H. (2023). Antitrust interoperability remedies. Columbia Law Review Forum, 123, 1.

Jebelli, K., & Ledford, H. (2024). Europe’s digital curtain: How the Digital Markets Act is turning Europeans into second-class digital citizens (Consumer impact report). Chamber of Progress. https://progresschamber.org/wp-content/uploads/2024/12/CHoP_DMA_Consumer-Impact-Report.pdf.  

Kerber, W., & Schweitzer, H. (2017). Interoperability in the digital economy. Journal of Intellectual Property, Information Technology and Electronic Commerce Law, 8, 39.

Kohlenberger, J. (2025, September 5). When interoperability mandates weaken security—JIT happens. Trusted Future. https://trustedfuture.org/when-interoperability-mandates-weaken-security-jit-happens.  

Lamadrid, A., & Fernández, N. B. (2021). Why the proposed DMA might be illegal under Article 114 TFEU, and how to fix it. Journal of European Competition Law & Practice, 12(7), 576.

Manne, G. (2020). Against the Vertical Discrimination Presumption. Concurrences n. 2-2020. https://www.concurrences.com/en/review/issues/no-2-2020/editorial/against-the-vertical-discrimination-presumption-94267-en.

Monti, G. (2021). The Digital Markets Act: Improving its institutional design. European Competition & Regulatory Law Review, 2, 90.

Petit, N. (2021). The proposed Digital Markets Act (DMA): A legal and policy review. Journal of European Competition Law & Practice, 12(7), 529.

Radic, L. (2025). The imaginary antitrust consensus. European Competition and Regulatory Law Review 9(1), 22.

Radic, L., Manne, G. A., & Auer, D. (2025). Regulate for what? A closer look at the rationale and goals of digital competition regulations. Berkeley Business Law Journal, 22(1), 201.

Radic, L. & Stout, K. (2024). What is the relevant product market in AI?. In A. Abbott & T. Schrepel (Eds.), Artificial intelligence and competition policy. Concurrences.

Radic, L. & Manne, G. (2022, January 11). Amazon Italy’s Antitrust Efficiency Offense. Truth on the Market https://truthonthemarket.com/2022/01/11/amazon-italys-efficiency-offense.

Robertson, v. H. S. E. (2024). The complementary nature of the Digital Markets Act and the EU antitrust rules. Journal of Antitrust Enforcement, 12, 325.

Scott Morton, F., Crawford, G. S., Cremer, J., Dinielli, D., Fletcher, A., Heidhues, P., & Schnitzer, M. (2023). Equitable interoperability: The “supertool” of digital-platform governance. Yale Journal on Regulation, 40, 1013.

Suominen, K. (2022). Implications of the European Union’s digital regulations on U.S. and EU economic and strategic interests. Center for Strategic & International Studies. https://csis-website-prod.s3.amazonaws.com/s3fs-public/publication/221122_EU_DigitalRegulations.pdf?VersionId=iuEl9KteAl_SKhjPCEWN8LlvqqORV02X.

Teece, D. (1986). Profiting from technological innovation. Research Policy, 15, 285.

Unekbas, S. (2023). Verticalization of data sharing and the difficult path to “Eunnovation.” Market and Competition Law Review, 7(2), 131.

van den Boom, J. (2023). What does the Digital Markets Act harmonize? Exploring interactions between the DMA and national competition laws. European Competition Journal, 19, 57.

Vezzoso, S. (2024). “Super-apps” and the Digital Markets Act. Journal of Antitrust Enforcement, 12(2), 331.

Witt, A. C. (2023). The Digital Markets Act: Regulating the Wild West. Common Market Law Review, 60(3), 625.

Consultation on AI in the Context of DMA Review

Question 6: What are the main obstacles to developing AI models and commercialising AI-based products and services?

The discussion about competition in the AI realm and the alleged main obstacles to developing AI models and commercialising AI-based products and services is often framed in pessimistic terms, assuming that digital markets are prone to converge on “entrenched monopolies”; and, therefore, often ending with proposals of “pre-emptive enforcement” or to enact specific sectoral regulation (Barnett, 2024).

A more balanced assessment, however, reveals a market that is nonetheless vibrant, competitive, and dynamic at multiple layers of the value chain—even if there are some reasonable concerns about possible input foreclosure at some points. At the level of consumer-facing applications and services, such as large language models (LLMs) and other AI-powered tools, competition is especially intense. New entrants are proliferating, with novel products being launched at a rapid pace and often gaining millions of users within weeks.

The possibility of multi-homing—where consumers use two or more services simultaneously—further amplifies these dynamics. For instance, users can easily switch between or combine services like ChatGPT, Claude, and Gemini. This creates a fluid environment where no single provider can rely on entrenched dominance, since users face low switching costs and often experiment with competing offerings. The result is a high degree of rivalry on both features and performance, which benefits consumers and pushes firms to innovate continuously. As commentators have noted, these competitive pressures are visible in the constant cycle of upgrades and feature releases across providers of LLM-based applications (Auer & Zúñiga, 2025; Chilson, 2025).

At the model-development layer, the picture is more nuanced, but still more competitive than is sometimes assumed. It is true that the costs of developing frontier foundation models are substantial, involving millions of dollars in compute and data expenses. Yet the narrative that only a handful of incumbents can realistically operate in this space is overstated. Several competing players—such as OpenAI, Anthropic, Google DeepMind, Meta, Mistral, Cohere, xAI and Stability AI—are actively building and releasing new models. Moreover, the rise of open-source models has significantly lowered barriers to entry for new firms and research groups. Open-source projects like Meta’s LLaMA, Google’s Gemma, Mistral AI, and Falcon, among others, have made it possible for smaller organisations to adapt and deploy competitive models without incurring prohibitive training costs.

The case of DeepSeek is illustrative: by leveraging open-source foundations and focusing on fine-tuning and optimisation, it was able to produce a model that rivalled much larger efforts, demonstrating that creativity and efficient resource use can substitute, to some extent, for sheer scale. This open-source dynamic injects competitive tension into the model-development layer, ensuring that innovation is not monopolised by the largest players (Auer & Zúñiga, 2025; Chilson, 2025).

At the infrastructural layer, there are more credible risks of bottlenecks, particularly concerning access to specialised chips and to cloud-computing services. Nvidia currently maintains a commanding lead in the market for GPUs used to train and deploy AI models. Its CUDA ecosystem and advanced chips have made it the default choice for most developers. Nevertheless, this position should not be misinterpreted as uncontested dominance.

There is active competition from hyper-scalers developing their own chips: Google’s Tensor Processing Units (TPUs), Amazon’s Trainium and Inferentia processors, and Microsoft’s ongoing investments all represent attempts to reduce dependence on Nvidia and to diversify the supply of high-performance compute (although Microsoft has postponed that investment). In addition, startups such as Cerebras and SambaNova are pursuing alternative architectures that, while still niche, could prove disruptive if technological breakthroughs enable them to scale. Policymakers, therefore, should be cautious in treating the chip market as if it were inherently monopolistic.

A similar dynamic is observable in cloud computing. It is correct that a small number of providers—Amazon Web Services (AWS), Microsoft Azure, and Google Cloud—currently dominate AI workloads. This dominance is, however, checked by both competitive churn and new entry. Cloud customers frequently switch providers in response to pricing, service, and performance differences, and firms are increasingly adopting multi-cloud strategies to avoid lock-in (Chilson, 2025).

Furthermore, Oracle—once considered a marginal player—has invested heavily in expanding its cloud infrastructure and has begun to win significant AI-related contracts. This demonstrates that even markets with a concentrated structure are subject to competitive forces when customers retain credible alternatives. In short, while cloud computing represents a possible chokepoint for AI development, competition in this area is currently active and evolving.

Critically, many of the perceived bottlenecks in AI reflect not a lack of competition but the reality of high fixed costs, technological uncertainty, and economies of scale inherent to cutting-edge innovation. These are structural features of the industry, rather than evidence of anticompetitive conduct. Over time, as technologies mature, costs decrease, and knowledge diffuses, barriers to entry typically diminish, as has been observed in prior waves of general-purpose technologies.

Considering these dynamics, the main obstacles to AI development and commercialisation are best understood as challenges of scale and technological sophistication, rather than structural foreclosure imposed by dominant firms. Access to compute, chips, and data are genuine hurdles, but they are not insurmountable, and they are being actively mitigated by competitive entry, open-source innovation, and user multihoming. Policymakers should be cautious about adopting interventionist measures that assume a static, monopolistic landscape, as such measures could inadvertently stifle the very dynamism that has characterised AI markets thus far.

This review of the DMA provides an opportunity to reflect on how best to balance vigilance against genuine risks of input foreclosure, while also recognizing the robust competitive forces already at play. Overstating the risks of concentration could lead to regulatory overreach, undermining incentives for investment in infrastructure and innovation. A more prudent approach is to monitor potential chokepoints closely while allowing competitive dynamics—including those spurred by open-source models, cloud churn, and custom chip development—to continue shaping the market. This would ensure that regulation supports, rather than impedes, the competitive vibrancy essential for AI innovation.

Do you have any further comments or observations on the DMA’s role to ensure fairness and contestability in the AI sector?

In line with the above, it would be premature and potentially counterproductive to include AI services or products (such as LLMs or generative AI) to the list of core platform services (CPS) under the DMA.

First, it is important to recognize that AI is a general-purpose technology that is already embedded in several of the services expressly listed as CPS under the DMA, such as search engines and social-networking services (Art. 2(2) DMA). In these contexts, AI’s functionality is already covered by existing CPS designations, making the creation of a separate AI-specific category redundant. Moreover, introducing an additional category would risk regulatory duplication and legal uncertainty, since embedded AI services within existing CPS are already subject to gatekeeper obligations, and any standalone treatment would overlap with parallel frameworks such as the AI Act.

If the Commission were nonetheless to pursue inclusion of a specific category of AI service or product (for example, LLMs or foundational models) within the CPS list, this should follow a proper regulatory process. Such a process must begin by establishing a solid evidentiary basis that demonstrates relevant markets have tipped in favour of a few entrenched providers, due to the presence of strong network effects, the services’ multisided nature, a significant degree of dependence on both business and end users, lock-in effects, or vertical integration, as contemplated in Recital 2 of the DMA.

Any such regulatory process should subsequently include a thorough cost–benefit analysis of the proposed inclusion. In particular, it must be acknowledged that the obligations and prohibitions triggered by CPS and gatekeeper designation under the DMA are substantial, both in terms of compliance costs and in their effects on firms’ ability to monetize platforms. Extending these obligations to AI could undermine the very dynamism that currently defines these markets. As noted above, the AI ecosystem is characterised by rapid entry, robust open-source alternatives, and significant multihoming by users. In such a competitive environment, imposing gatekeeper obligations risks freezing a static conception of market power that fails to reflect the fluid and evolving nature of technological progress.

Finally, overlap with other regulatory initiatives—most prominently the EU Artificial Intelligence Act—raises a further concern. The AI Act already establishes horizontal rules governing transparency, safety, and risk management in AI systems. Adding a CPS category under the DMA would create duplicative, and potentially inconsistent, obligations. Rather than expanding the DMA’s scope, a more prudent approach is to rely on the AI Act to address systemic risks associated with AI, and to use the DMA only insofar as AI functionalities are already captured through existing CPS categories. This would preserve regulatory coherence, while avoiding unnecessary burdens on a still-developing and highly competitive sector.

Robust competition in the AI sector is manifest; there is no basis for incorporating AI services or products into the DMA.

In order to assist the Commission in understanding the competitiveness of this sector, we attach to this submission the following writings by the authors of these comments:

1. Stout & Hemphill, A Framework for Understanding and Evaluating AI Commercialization Strategies
2. Auer & Zuniga, AI Partnerships and Competition: Damned if You Buy, Damned if You Don’t
3. Manne, Albrecht, Auer, Radic, & Zúñiga, ICLE Comments on the CMA’s Provisional Findings on the Cloud Services Market
4. Manne, Auer, Stout, Radic, & Zúñiga, ICLE Comments on JFTC Request for Information and Comments Concerning Generative AI and Competition
5. Radic & Stout, What Is the Relevant Product Market in AI?
6. Manne, Auer, Stout, Radic, & Zúñiga, ICLE Comments to DOJ on Promoting Competition in Artificial Intelligence
7. Manne, Auer, Wudrick, & Zúñiga, ICLE and Macdonald-Laurier Institute Comments to Competition Bureau Canada Consultation on AI and Competition
8. Manne, Auer, & Zúñiga, ICLE Comments to UK Competition and Markets Authority on AI Partnerships
9. Manne & Auer, ICLE Comments to European Commission on AI Competition
10. Manne & Auer, From Data Myths to Data Reality: What Generative AI Can Tell Us About Competition Policy (and Vice Versa)

How is the DMA affecting AI rollout in the EU?

Perhaps most concerning is the way the DMA has begun to stifle innovation in Europe. Some have referred to this as a “digital curtain” that separates European citizens from cutting-edge digital services available elsewhere (Jebelli & Ledford, 2024). Indeed, regulatory uncertainty caused by the DMA has led companies to delay launching new products and features in the EU, or even to bypass the European market altogether (Auer, 2024). This is particularly true in the realm of generative AI.

Concrete examples include delays to Apple Intelligence and related capabilities: Apple publicly said in 2024 that it would hold back Apple Intelligence, iPhone Mirroring, and enhanced SharePlay Screen Sharing in the EU due to DMA interoperability requirements and the risk of compromising privacy and security. These features all were rolled out in the United States on schedule, even as they were explicitly withheld in Europe pending clarity.

Google’s rollout of AI services has likewise been staggered. When Google launched the Gemini mobile apps in May 2024 and made them ostensibly available “globally”, they were withheld in the EU and UK, with Google saying it would expand only where consistent with local rules. Moreover, while Google’s AI Overviews debuted in the United States in May 2024 and later expanded to many markets, EU availability remained limited amid heightened regulatory scrutiny. Indeed, several outlets reported EU holds or limited availability well into 2025. The upshot for European users was months of delayed access compared to peers in other regions.

Meta offers a parallel illustration of the broader regulatory chill around launching AI assistants in Europe. In June 2024, it paused the EU launch of Meta AI after Ireland’s Data Protection Commission (DPC) requested a delay over training-data issues. While that pause was seemingly driven more by the GDPR than the DMA, it underscores how overlapping EU regimes have combined to make firms sequence AI launches outside Europe first—precisely the pattern the DMA’s uncertainty exacerbates.

Taken together, these time-lags map onto the “digital curtain” that critics describe. Europeans get powerful AI capabilities later, or not at all, while the same products proceed elsewhere. That is the very dynamic many warned the DMA would catalyse when layered atop existing EU rules.