ICLE Comments to the Taiwan Fair Trade Commission on Generative AI
Introduction and Summary
The International Center for Law & Economics (ICLE) respectfully submits these comments on the Taiwan Fair Trade Commission’s (TFTC) Public Consultation Report and Policy Statement on Generative Artificial Intelligence and Competition.[1] We commend the TFTC for a thoughtful, economically grounded document whose method and conclusions reflect the best of the modern law & economics tradition.[2]
The Policy Statement adopts a sound analytical approach. It opens with the maxim ex facto jus oritur—law arises from the facts—and honors that commitment throughout.[3] It adopts four guiding principles: issue-driven and evidence-based analysis, attention to a nexus with Taiwan’s markets, preservation of market contestability (the ability to enter markets and challenge incumbents), and case-by-case assessment under the rule of reason.[4]
The Policy Statement cautions against equating scale, data, or computing power with market dominance. It rejects reliance on market structure alone, recognizes the potential benefits of partnerships, and distinguishes domestic firms’ competitive disadvantages from competition-law violations. The TFTC’s commitment to evidence-based, case-specific assessment offers a valuable alternative to the European Union’s prescriptive approach to digital competition under the Digital Markets Act.[5]
Part I explains why these principles deserve support. Part II suggests clarifications to help firms and enforcers apply them consistently, particularly to ecosystems, entry, talent, and integration. Our recommendations seek to preserve the flexibility and attention to evidence that make the Policy Statement valuable. Part III concludes.
I. Support for the TFTC’s Evidence-Based Approach to AI Competition
A. Ground Enforcement in Evidence of Competitive Harm
The Policy Statement’s insistence on a concrete theory of competitive harm to guide enforcement priorities is among its most important features.[6] As the document recognizes, a theory of harm links specific conduct, a mechanism, and a competitive effect. The theory alone does not establish that the mechanism operates in the market concerned.
In rapidly evolving markets, a plausible account of how an incumbent might exclude rivals cannot substitute for evidence that the conduct is likely to harm competition. Prospective assessment nonetheless remains appropriate. Enforcers need not wait for harm to occur, provided evidence supports the predicted harm.
Enforcement in artificial intelligence (AI) markets should replace hype and preconceived views with careful, evidence-based analysis.[7] Uncertainty in nascent markets increases the risk that premature intervention will suppress beneficial conduct.[8] The TFTC’s approach appropriately requires sufficient evidence to justify intervention while leaving room for innovation and experimentation.
B. Assess Market Power Across Distinct AI Activities
The Policy Statement rightly cautions against equating the scale and volume of data and computing power with market dominance or a lack of contestability.[9] The OpenAI–Cerebras computing agreement, reported discussions involving OpenAI and Anthropic over biotech data, the Adobe–Runway partnership, and the BMW–Innoviz collaboration illustrate the importance of specialized suppliers and complementary capabilities. Scale in one activity does not automatically establish market power in another.[10]
The economic reasons for this caution vary across AI activities. Inference—running a trained model to generate outputs—entails variable costs. Those costs do not, however, negate substantial fixed costs, sunk investments, or economies of scale in model development and cloud infrastructure. An entrant also need not match a frontier model’s capabilities across every task to compete effectively for a particular use. Smaller, specialized models may offer an attractive combination of cost and performance.[11] Treating AI as a single market or monolithic input obscures differences in demand, production costs, and opportunities for entry across the “AI stack”—the layers of infrastructure, models, and applications that support AI services.[12]
Oversimplification is particularly problematic when assessing the competitive significance of data access. That significance depends, among other things, on which data a task requires, whether public, licensed, or synthetic alternatives are adequate, and whether claimed advantages persist as methods improve. Data volume alone does not establish a bottleneck that excludes rivals.
Open-weight models, which make their trained parameters available for others to use or adapt, can also reduce development costs for entrants. The TFTC should assess their availability alongside proprietary alternatives without prescribing a preferred business model. Any intervention should identify the particular constraint that prevents effective competition.[13]
C. Assess Consumer Welfare Beyond Market Concentration
The Policy Statement rightly reflects the mainstream economic rejection of the structure-conduct-performance (SCP) paradigm, which links market structure to firms’ conduct and market outcomes.[14] It declines to presume illegality from market structure unless the evidence shows that structure determines the overall competitive effect. It also expressly directs attention to demand and user choice.[15] Concentration alone cannot establish durable market power or consumer harm, as it may also reflect economies of scale, superior performance, or successful innovation.[16]
A consumer-welfare approach should assess price, output, quality, and innovation, including privacy and security where users value them. The number of suppliers or standalone products is not itself a measure of consumer welfare. Moreover, integrated offerings can improve welfare even when they reduce some forms of choice.
D. Assess Acquisitions and Partnerships by Their Competitive Effects
The report records stakeholders’ concern that unnecessary burdens on mergers and acquisitions may weaken innovation, while the Policy Statement recognizes that strategic alliances can promote innovation and competition.[17] These observations support a receptive, effects-based approach to AI partnerships. Acquisition can provide an important exit strategy for startups. Partnerships can supply the funding, computing resources, and complementary capabilities needed to commercialize their work.
The TFTC should be especially wary of treating an incumbent’s acquisition as suspect merely because the incumbent might hypothetically have developed the target’s business itself. That “growth through acquisition” theory rests on a fringe reading of antitrust law that, if generalized, could diminish acquisition-driven growth across the economy.[18] A credible independent-entry counterfactual—a prediction of how the incumbent would have entered without the acquisition—requires evidence of likely entry, its timing, and its competitive significance. Technical or financial capacity alone is insufficient.[19]
Recent empirical work also counsels against categorical assumptions. Florian Ederer, Regina Seibel, and Timothy Simcoe find little support for the often-claimed position that digital incumbents’ acquisitions systematically “kill” innovation.[20] They conclude instead that such transactions often reflect complementary capabilities and wider technology diffusion.[21]
The same error is particularly salient in AI markets. Firms may face criticism for acquiring or partnering to obtain complementary capabilities, and also for building those capabilities internally. This regulatory bind risks chilling the investments and collaborations needed for innovation.[22] The broader literature finds that voluntary vertical integration—combining activities at different stages of production or distribution—frequently benefits consumers. That evidence supports requiring a demonstration of harm before treating integration as anticompetitive.[23]
The appropriate comparison is with the opportunities realistically available without the transaction. Alternative financing may arrive later, cost more, or lack the computing and commercial support a partner supplies. Contractual protections may enable investments tailored to a particular commercial relationship rather than exclude rivals. These considerations require evidence about the agreement’s terms and alternatives. They do not establish that every restriction is efficient. Preventing established technology firms from entering through partnerships can also protect today’s leading AI providers from an important source of rivalry.[24]
E. Distinguish Competition Enforcement from Industrial Policy
Among the Policy Statement’s clearest analytical strengths is its distinction between policies addressing domestic firms’ limited capital or lower turnover and conduct that harms competition.[25] Competition law protects the competitive process and users’ welfare. It should not engineer a particular industrial structure or favor firms because of their nationality.
The report’s discussion of incentives and subsidies need not imply endorsement. Any such measures require their own justification, accounting for their fiscal costs, effects on entry, and risk of protecting less efficient firms.
We likewise commend the TFTC’s competition advocacy under Article 46 of the Fair Trade Act—in particular its warning that proposed data-sharing requirements may weaken firms’ incentives to innovate and compete.[26] Scrutiny of regulatory barriers to entry complements enforcement. The same attention to incentives should inform any proposed access or interoperability remedy—a requirement to provide access to resources or enable systems to work together. Its expected benefits should justify the costs to investment, privacy, security, and implementation. Its scope and duration should also be no broader than necessary to address the demonstrated problem.
Competition advocacy could also examine whether data-localization requirements—rules governing where firms store or process data—and other compliance obligations unnecessarily restrict access to computing resources or raise adoption costs for smaller firms. That inquiry should distinguish customers’ voluntary preferences from regulatory constraints, assess the stated privacy and security objectives, and consider whether less restrictive measures could achieve them.[27]
F. Adapt Comparative Experience to Taiwan’s Markets
The TFTC rightly recognizes that Taiwan’s competitive context differs from that of other jurisdictions and that comparative experience requires adaptation to local conditions. The European model, which some jurisdictions have rushed to adopt, has not demonstrated superior outcomes. Before adopting it, authorities should assess its economic rationale and suitability to local conditions.[28]
The Policy Statement also appropriately notes that earlier Microsoft tying cases—concerning the sale of products together—did not involve generative AI.[29] Those cases may inform analysis, but they cannot establish the effects of a different arrangement under different competitive conditions. Authorities should assess the benefits of integration alongside any restriction of choice. The availability of more standalone products does not by itself demonstrate greater consumer welfare.[30]
II. Clarifications for Consistent, Evidence-Based Enforcement
We offer the following observations in the same constructive spirit as the above. Clarifying how the TFTC will apply its principles would improve predictability for businesses and help preserve the Policy Statement’s evidence-based, case-specific approach.
A. Scrutinize Emerging Ecosystems Case by Case
The Policy Statement places particular importance on promoting competition as AI ecosystems form.[31] Early attention can help the TFTC identify conduct likely to exclude rivals before harm becomes difficult to remedy. The accompanying discussion of the Apple–Google arrangement and Google’s integration across AI activities identifies useful subjects for inquiry. The existence of a partnership or an integrated provider, however, should not itself create a presumption of harm.[32]
We respectfully encourage the TFTC to clarify that this principle calls for heightened vigilance—a readiness to investigate conduct in emerging ecosystems under the rule of reason. It should not authorize preventive rules that prescribe market structure.
The document’s own ecosystem analysis supports this distinction. Where firms compete “for the market”—to establish a leading ecosystem—the relevant focus is competition between ecosystems or brands. Preserving rivalry within an ecosystem is not an end in itself. Competition between ecosystems can discipline conduct within them, while restrictions on rivalry within an ecosystem may support investment and competition between integrated offerings.[33]
The TFTC should test the strength of this competitive constraint in each case, rather than assume it is either absent or sufficient. Rules that dictate how an ecosystem must be organized risk suppressing the integration and experimentation through which providers compete.
B. Define Markets Around How Users Choose Among Offerings
The Policy Statement’s attention to competition within and between ecosystems is welcome.[34] An ecosystem can itself constitute the relevant market where users choose among competing systems and the integration of complementary services determines the offering’s performance. Defining markets solely around individual components can exclude the very competition that constrains a provider.
In AI, customers may choose among combinations of models, computing resources, development tools, and applications whose usefulness depends on how they work together. Common governance, compatibility rules, and coordinated investment help produce that value. They form part of the competitive offering and should not carry a presumption that they restrict competition and require removal.[35]
Market definition should therefore examine users’ willingness to switch among offerings based on the dimensions on which they compete, including quality, functionality, and innovation. The question is whether an increase in price or a decline in performance would cause enough users to switch to make that change unprofitable. The analysis should account for changes in demand and participation across components that materially depend on one another.
Depending on the evidence, the relevant alternatives may include other ecosystems, less integrated combinations, or non-AI products and services. Providers may compete through different combinations of internal development, partnerships, and purchased inputs. Antitrust should not privilege one organizational form over another.[36]
This approach gives practical effect to the TFTC’s attention to demand. Neither a narrow product label nor the description of a collection of products as an ecosystem should predetermine the market.
C. Assess Contestability Through Evidence of Competitive Constraints
The Policy Statement recognizes that current indicators—competitor counts, prices, sales, and market-share changes—may not adequately capture competition in a rapidly evolving sector. It therefore proposes examining the potential and probability of entry.[37] We agree with this forward-looking conception of contestability, associated with William J. Baumol. The threat of entry can discipline incumbents even where few firms operate. The relevant question is whether that threat constrains their conduct, rather than whether the market contains a preferred number of competitors.[38]
Left unspecified, “potentiality and probability of entry” can cut both ways. Without a concrete evidentiary test, it can become an open-ended license to intervene whenever an incumbent appears large. The TFTC should instead specify whether entry or expansion is likely, timely, and sufficient to constrain the conduct at issue. Neither current concentration nor the mere possibility of entry should be considered conclusive.
This economic conception of contestability differs from the European regulatory approach. Reducing an incumbent’s advantages or increasing the number of rivals does not establish that consumers are better off. Network effects (the value a service gains as more people use it), economies of scale, and accumulated capabilities may reflect successful competition. Compelling firms to share or surrender those advantages can weaken investment without strengthening competitive constraints. Contestability should discipline the exercise of market power, rather than protect competitors from the consequences of competition.[39]
Error-cost analysis—weighing the expected costs of mistaken intervention and nonintervention—should inform the enforcement decision. The TFTC should weigh the expected costs of mistaken intervention and nonintervention, including lost innovation and the ability of entry to correct market power.[40] Evidence of multi-homing—using multiple providers—and competition on data localization and security, which the TFTC identifies in Taiwan, is relevant to that assessment.[41] Its significance depends on whether users can shift meaningful workloads at reasonable cost, rather than simply maintain accounts with multiple providers. This distinction would make contestability a practical inquiry into competitive constraints while retaining appropriate caution about intervention.
The TFTC should likewise assess switching costs over the customer relationship and against realistic alternatives. Some costs reflect data transfer, retraining, or the distinctive functionality customers chose. Competition to attract customers may also produce discounts and other benefits that compensate for anticipated migration costs.
The relevant inquiry is whether a particular restriction impairs competition, accounting for total costs, customers’ ability to anticipate contractual terms, and available alternatives. Neither competition at the initial contracting stage nor the existence of switching costs is conclusive. Requiring uniform products or costless switching could sacrifice differentiation without improving consumer welfare.[42]
D. Assess Labor Restraints on Competitive Grounds
The separation of industrial policy and competition law should also guide the treatment of noncompete clauses. The Policy Statement refers to Taiwan’s interest in stemming the outflow of talent and key technologies.[43] We encourage the TFTC to distinguish that industrial-policy objective from competition-relevant justifications, such as protecting trade secrets or enabling investment in training. The TFTC should assess those justifications on their merits, irrespective of the nationality of the firms or workers concerned.
The competitive assessment should also distinguish among types of restraint. A restriction reasonably necessary to a productive collaboration raises different questions from an agreement among competing employers simply to refrain from hiring workers or competing on compensation. The economics support effects-based review of ancillary restraints—restrictions reasonably necessary to a broader productive arrangement. They do not support a blanket rule-of-reason safe harbor for labor-market cartels.[44]
Subject to Taiwan’s applicable legal rules, the analysis should consider the restraint’s scope, duration, effect on workers’ opportunities, and substantiated efficiencies. National talent retention should not substitute for competitive analysis.
E. Assess Acquihires by Their Effects and Alternatives
The Policy Statement lists acquihires—arrangements to recruit a firm’s team—alongside no-poach and noncompete clauses among talent-related practices warranting attention.[45] We encourage the TFTC to resist the view that acquihires are inherently suspect and assess their likely effects under the applicable legal standards. Hiring a team does not itself establish that a competitor has disappeared. These arrangements can serve different economic functions, and their competitive effects depend on their terms and the realistic alternatives available to the firms involved.[46]
Three distinctions would assist that inquiry. First, the labor-market conditions identified in the TFTC’s report may cut either way. They do not necessarily establish a scarcity of AI workers. Rising wages may attract additional workers, while retraining may alleviate shortages. Rising wages may also reflect scarcity, and the TFTC’s record identifies different training requirements for different AI roles that may limit retraining’s effectiveness. The inquiry should examine the specific skills concerned, sources of additional labor (e.g., retraining, education, and immigration), available substitutes, and the time needed to recruit or train alternatives.[47]
Second, the structural-merger framework fits acquihires poorly. Labor is not a commodity that firms can “acquire,” and employment relationships cannot transfer without workers’ consent. A team’s movement also has different competitive effects from the permanent transfer of a productive asset. Retention, employee choice, contractual restrictions, and the capabilities remaining at the target all matter. The typical case nevertheless has few of the hallmarks of a traditional merger.[48]
Third, the transactions most often invoked as cautionary tales are poor examples. In AI inference, the cost of querying a model at GPT-3.5-level performance fell more than 280-fold between November 2022 and October 2024.[49] In such a market, portraying a talent-and-licensing deal as eliminating a uniquely formidable rival assumes an unlikely course of events without the deal. A credible case requires evidence that the target would otherwise have remained a meaningful competitive constraint and that the transaction is likely to remove that constraint.[50]
To be clear, an acquihire can still be anticompetitive. Where a transaction removes a meaningful competitive constraint, the TFTC should examine whether it falls within the applicable legal provisions and whether the evidence establishes harm.
But acquihires can also enable entrepreneurial exit, transfer practical knowledge embedded in cohesive teams, and sustain the labor mobility and reputational incentives on which startup ecosystems depend.[51] These benefits belong in an effects-based assessment and support caution about treating the transaction’s form as evidence of competitive injury.
F. Distinguish Exclusionary Conduct from Growth on the Merits
As noted, the Policy Statement rightly cautions against equating scale with dominance.[52] It also expresses concerns that larger firms may continue to grow and that markets may tend toward winner-takes-all outcomes.[53]
The TFTC should clarify the Policy Statement’s exception for circumstances in which market structure could determine competitive effects. Any such inference should explain why the observed structure reflects or enables a restriction of competition rather than superior performance or efficient scale. An undefined exception risks restoring the very presumption the Policy Statement otherwise rejects. Specifying the necessary evidence and distinguishing merger review from conduct cases would avoid that result.
A large firm’s growth is not inherently a competitive problem, particularly when the sector also grows through more output, lower prices, better integration, and higher quality. The relevant distinction is between exclusionary conduct that harms competition and growth through competition on the merits. Enforcement should target the former. The latter remains lawful even when it enlarges a firm’s market power.
Access to capital, vertical integration, and scale can support lower costs, better products, and innovation. Market power acquired through superior performance is not itself an antitrust violation. Enforcement should protect competition rather than particular competitors.[54]
At the same time, sector-wide growth does not establish that every practice is benign. The relevant inquiry is how the conduct changes competition and consumer welfare compared with realistic conditions without it, accounting for likely efficiencies.[55]
Tying and bundling likewise require attention to likely exclusion and consumer benefits under the applicable standards. Integration alone cannot establish harm. Any proposed remedy should account for the agency’s limited information about product design and the risk of impairing beneficial integration.[56]
G. Distinguish Independent Pricing from Unlawful Coordination
The consultation identifies uncertainty about AI-assisted pricing and recognizes these tools’ potential benefits. The TFTC’s evidence-based approach should distinguish independent use of pricing software from agreements or information exchanges that facilitate coordination. Common software, similar recommendations, or parallel prices should not alone establish concerted action.
Relevant questions include what information firms share, whether it is public or competitively sensitive, whether firms retain independent pricing discretion, and what evidence establishes coordination under the Fair Trade Act. This approach would preserve effective enforcement against unlawful coordination while allowing firms and consumers to benefit from lawful uses of pricing technology.[57]
III. Conclusion
The TFTC’s Policy Statement offers a valuable example of economically grounded, evidence-based competition policy. We commend its insistence on concrete theories of harm, its economically informed treatment of partnerships and acquisitions, its separation of competition law from industrial policy, and its commitment to local evidence and case-by-case assessment.
Our suggested clarifications would help preserve these commitments against exceptions that reinstate presumptions against scale, integration, or entrepreneurial exit. Recognizing competition between ecosystems, assessing entry and switching in their economic context, and requiring a credible assessment of what would happen without an acquisition would make the Policy Statement’s principles more effective in practice.
We also support addressing the identified competition concerns through the existing Fair Trade Act.[58] Before adopting AI-specific prohibitions, the TFTC should identify a concrete deficiency in existing law and assess whether the proposed rule’s benefits justify its compliance and innovation costs. Taiwan can protect consumers and competition while preserving the investment, integration, and experimentation through which AI services improve.
[1] Authored by Geoffrey A. Manne, president and founder of the International Center for Law & Economics (ICLE), and Lazar Radic, senior scholar at ICLE. ICLE is a nonprofit, nonpartisan research center based in Portland, Oregon, dedicated to building the intellectual foundations for sensible, economically grounded policy. ICLE promotes law & economics methods to inform public policy and has filed comments and analyses with competition and regulatory authorities across numerous jurisdictions.
[2] Taiwan Fair Trade Comm’n, Public Consultation Report and Policy Statement on Generative Artificial Intelligence and Competition (Mar. 2026), https://www.ftc.gov.tw/upload/833774aa-edbd-42a5-a41e-5d71045a3e6c.pdf [hereinafter Policy Statement]. Page references follow the document’s printed page numbers.
[3] Policy Statement, supra note 2, at 47 (quoting Louis D. Brandeis, The Living Law, 10 Ill. L. Rev. 461, 467 (1916)).
[4] Id. at 61.
[5] Regulation (EU) 2022/1925, 2022 O.J. (L 265) 1 (Digital Markets Act). See generally Dirk Auer, Geoffrey A. Manne & Sam Bowman, Should ASEAN Antitrust Laws Emulate European Competition Policy?, 67 Sing. Econ. Rev. 1637 (2022), https://doi.org/10.1142/S0217590821430025.
[6] Policy Statement, supra note 2, at 50–51 & n.27 (describing theories of competitive harm as narratives linking conduct, mechanism, and effect).
[7] Lazar Radic & Kristian Stout, What Is the Relevant Product Market in AI?, in Artificial Intelligence and Competition Policy (Alden Abbott & Thibault Schrepel eds., 2024) (author manuscript at 5, 20), https://laweconcenter.org/wp-content/uploads/2024/09/ssrn-4927505.pdf [hereinafter Relevant Product Market]. Manuscript pinpoints count PDF pages beginning with the first page.
[8] Relevant Product Market, supra note 7, manuscript at 14 (discussing Jonathan M. Barnett, The Case Against Preemptive Antitrust in the Generative Artificial Intelligence Ecosystem, in Artificial Intelligence and Competition Policy (Alden Abbott & Thibault Schrepel eds., 2024) (author manuscript at 8)).
[9] Policy Statement, supra note 2, at 58 (cautioning against equating the scale and volume of data and computing power with market dominance or a lack of contestability).
[10] Id. at 57–58 & nn.34–38.
[11] For example, Google’s January 2026 TranslateGemma release includes a translation-specific 12-billion-parameter model that outperformed the general Gemma 3 27-billion-parameter baseline on Google’s reported translation benchmark, using fewer than half as many parameters. See David Vilar & Kat Black, TranslateGemma: A New Suite of Open Translation Models, Google (Jan. 15, 2026), https://blog.google/innovation-and-ai/technology/developers-tools/translategemma.
[12] See Andrei Hagiu & Julian Wright, Artificial Intelligence and Competition Policy, 103 Int’l J. Indus. Org. 103134, at 9–10 (2025), https://doi.org/10.1016/j.ijindorg.2025.103134; Policy Statement, supra note 2, at 54; Relevant Product Market, supra note 7, manuscript at 1–2, 6–8.
[13] See Geoffrey A. Manne et al., ICLE Comments to DOJ on Promoting Competition in Artificial Intelligence §§ I.C–D (Int’l Ctr. for L. & Econ., July 15, 2024), https://laweconcenter.org/resources/icle-comments-to-doj-on-promoting-competition-in-artificial-intelligence [hereinafter DOJ AI Comments]; Hagiu & Wright, supra note 12, at 9–10; Kristian Stout & Thomas A. Hemphill, A Framework for Understanding and Evaluating AI Commercialization Strategies 5–8 (Int’l Ctr. for L. & Econ., Issue Brief, Aug. 27, 2025), https://laweconcenter.org/resources/a-framework-for-understanding-and-evaluating-ai-commercialization-strategies.
[14] See Steven Berry, Martin Gaynor & Fiona Scott Morton, Do Increasing Markups Matter? Lessons from Empirical Industrial Organization, 33 J. Econ. Persp. 44, 45–48 (2019), https://doi.org/10.1257/jep.33.3.44 (explaining why concentration–performance correlations cannot identify the competitive mechanism or its welfare consequences).
[15] Policy Statement, supra note 2, at 58–59.
[16] See Auer, Manne & Bowman, supra note 5, at 1679–80 (examining the limitations of concentration-based inferences); Berry, Gaynor & Scott Morton, supra note 14, at 45–48.
[17] Policy Statement, at 45–46 & 48.
[18] See Geoffrey A. Manne & Justin (Gus) Hurwitz, Build, Buy, or Both?: On the Antitrust Laws’ Supposed Preference for “Internal Growth” over Acquisitions, 26 Nev. L.J. 53, 80–87 (2025), https://scholars.law.unlv.edu/nlj/vol26/iss1/3.
[19] See, e.g., FTC v. Meta Platforms, Inc., 654 F. Supp. 3d 892, 934–35 (N.D. Cal. 2023) (requiring a feasible and reasonably probable path to independent entry rather than an inference from resources alone). See also Lazar Radic, Imaginary Consensus as a Legitimizing Philosophy of the New Antitrust Meta-Narrative 5 (Int’l Ctr. for L. & Econ., Issue Brief, Jan. 31, 2025) [hereinafter Imaginary Consensus].
[20] See, e.g., Cristina Caffarra, Gregory S. Crawford & Tommaso Valletti, “How Tech Rolls”: Potential Competition and “Reverse” Killer Acquisitions, CPI Antitrust Chron., May 2020, at 13, 14–16, https://www.competitionpolicyinternational.com/wp-content/uploads/2020/05/AC-May-II.pdf (arguing that acquisitions can eliminate both the target’s potential challenge and the acquirer’s own innovation effort, and favoring presumptions against such transactions).
[21] Florian Ederer, Regina Seibel & Timothy Simcoe, Digital (Killer?) Acquisitions 51 (Nat’l Bureau of Econ. Rsch., Working Paper No. 35762, 2026) (“[T]he evidence is more consistent with complementarity than with suppression: innovation does not decline after these deals, and the acquired technologies diffuse more widely.”).
[22] See Dirk Auer & Mario A. Zúñiga, AI Partnerships and Competition: Damned if You Buy, Damned if You Don’t 17–21 (Int’l Ctr. for L. & Econ., White Paper, Aug. 19, 2025) (forthcoming Eur. Competition L. Rev.), https://laweconcenter.org/wp-content/uploads/2025/08/AI-partnerships-and-competition.pdf.
[23] See Geoffrey A. Manne, Kristian Stout & Eric Fruits, The Fatal Economic Flaws of the Contemporary Campaign Against Vertical Integration, 68 U. Kan. L. Rev. 923, 957 (2020); Francine Lafontaine & Margaret Slade, Vertical Integration and Firm Boundaries: The Evidence, 45 J. Econ. Lit. 629, 680 (2007), https://doi.org/10.1257/jel.45.3.629. See also Imaginary Consensus, supra note 19, at 5.
[24] See Manne & Hurwitz, supra note 18, pt. III; DOJ AI Comments, supra note 13, at 22–24; Geoffrey A. Manne, Dirk Auer & Mario A. Zúñiga, ICLE Comments to UK Competition and Markets Authority on AI Partnerships (Int’l Ctr. for L. & Econ., May 9, 2024), https://laweconcenter.org/resources/icle-comments-to-uk-competition-and-markets-authority-on-ai-partnerships.
[25] Policy Statement, supra note 2, at 49–50.
[26] Id. at 50 & n.26. See also DOJ AI Comments, supra note 13, § I.D (discussing how mandatory data sharing can weaken investment incentives).
[27] See Policy Statement, supra note 2, at 8–10, 39, 50, 54 (discussing localization, computing access, and competition advocacy).
[28] See generally Auer, Manne & Bowman, supra note 5, at 1640 (arguing that the EU model’s appeal partly reflects its accommodation of politically useful populist narratives about bigness and robber barons that more evidence-based regimes resist).
[29] Policy Statement, supra note 2, at 49.
[30] See Auer, Manne & Bowman, supra note 5, at 1657–58 (criticizing the failure to weigh asserted consumer-choice benefits against integration efficiencies in the European Microsoft litigation).
[31] Policy Statement, supra note 2, at 56.
[32] Id. at 56 & nn.32–33 (noting the Apple–Google arrangement to adopt Gemini for Siri and Apple Intelligence, and concerns about Google’s full-stack vertical integration).
[33] See Auer, Manne & Bowman, supra note 5, at 1664–65; Policy Statement, supra note 2, at 59–60.
[34] Policy Statement, supra note 2, at 59–60.
[35] See Geoffrey A. Manne, Ecosystem Market Definition: Systems Competition, Dynamic Analysis, and the Limits of Ecosystem Theories of Harm, in Conglomerates and Ecosystems in the Digital Era ch. 12, §§ 12.3–12.4 (Justin (Gus) Hurwitz, Geoffrey A. Manne & Jeffrey Prince eds., Cambridge Univ. Press forthcoming) (Sept. 10, 2026 draft).
[36] See Manne, supra note 35, §§ 12.3–12.4; Manne & Hurwitz, supra note 18, pt. III; Relevant Product Market, supra note 7, manuscript at 14–20.
[37] Policy Statement, supra note 2, at 55–56.
[38] See William J. Baumol, Contestable Markets: An Uprising in the Theory of Industry Structure, 72 Am. Econ. Rev. 1, 3–4 (1982); Geoffrey A. Manne, Lazar Radic & Dirk Auer, Regulate for What? A Closer Look at the Rationale and Goals of Digital Competition Regulations, 22 Berkeley Bus. L.J. 201, 235–41 (2025), https://doi.org/10.15779/Z38M03Z022.
[39] See Manne, Radic & Auer, supra note 38, at 235–41.
[40] See Baumol, supra note 38, at 3–4; Frank H. Easterbrook, The Limits of Antitrust, 63 Tex. L. Rev. 1, 2–3 (1984).
[41] Policy Statement, supra note 2, at 53–54, 59–60.
[42] See Brian Albrecht & Geoffrey A. Manne, ICLE Response to the FTC’s Cloud Computing RFI § II.B (Int’l Ctr. for L. & Econ., June 21, 2023), https://laweconcenter.org/resources/icle-response-to-the-ftcs-cloud-computing-rfi; Geoffrey A. Manne et al., ICLE Comments on the CMA’s Provisional Findings on the Cloud Services Market §§ IV–V (Int’l Ctr. for L. & Econ., Feb. 18, 2025), https://laweconcenter.org/resources/icle-comments-on-the-cmas-provisional-findings-on-the-cloud-services-market; Policy Statement, supra note 2, at 12–15, 51–52, 59–60.
[43] Policy Statement, supra note 2, at 54–55 & n.29.
[44] See Brief for International Center for Law & Economics and Scholars of Law and Economics as Amici Curiae Supporting Appellees at 17–19, Giordano v. Saks Inc., No. 23-600 (2d Cir. Nov. 3, 2023), https://laweconcenter.org/wp-content/uploads/2023/11/2023.11.03-Saks-ICLE-Amicus-Motion-and-Brief-As-Filed.pdf. This example applies U.S. law. Although the economics are the same, Taiwan’s law may differ.
[45] Policy Statement, supra note 2, at 52.
[46] See, e.g., Selcukhan Ünekbas, Acquihires and Other Antitrust Ghost Stories, Truth on the Mkt. (Mar. 25, 2026), https://truthonthemarket.com/2026/03/25/acquihires-and-other-antitrust-ghost-stories; Onyeka Aralu & Dirk Auer, Acquihires and Antitrust: When Buying the Team Isn’t Buying the Company, Truth on the Mkt. (Apr. 9, 2026), https://truthonthemarket.com/2026/04/09/acquihires-and-antitrust-when-buying-the-team-isnt-buying-the-company.
[47] Policy Statement, supra note 2, at 52; cf. Ünekbas, supra note 46.
[48] See John F. Coyle & Gregg D. Polsky, Acqui-Hiring, 63 Duke L.J. 281, 293–99, 303–05 (2013), https://scholarship.law.duke.edu/dlj/vol63/iss2/1; Beril Boyacioglu, Mahmut N. Özdemir & Samina Karim, Acqui-hires: Redeployment and Retention of Human Capital Post-Acquisition, 45 Strategic Mgmt. J. 205, 205–07 (2024), https://doi.org/10.1002/smj.3556. See also Aralu & Auer, supra note 46.
[49] See Stanford Inst. for Human-Centered Artificial Intelligence, Artificial Intelligence Index Report 2025 41 (2025), https://hai.stanford.edu/assets/files/hai_ai_index_report_2025.pdf (reporting a decline from $20 to $0.07 per million tokens for GPT-3.5-level MMLU performance, using a 3:1 weighted average of input and output token prices).
[50] See Eric Fruits et al., ICLE Comments to FTC and DOJ on Premerger Notifications § IV.B (Int’l Ctr. for L. & Econ., May 22, 2026), https://laweconcenter.org/resources/icle-comments-to-ftc-and-doj-on-premerger-notifications.
[51] See Coyle & Polsky, supra note 48, at 283–85, 294–95, 309–16; Boyacioglu, Özdemir & Karim, supra note 48, at 205–07; Aralu & Auer, supra note 46; Ünekbas, supra note 46.
[52] Policy Statement, supra note 2, at 58.
[53] Id. at 55, 59.
[54] See Imaginary Consensus, supra note 19, at 5; Lafontaine & Slade, supra note 23, at 680.
[55] Policy Statement, supra note 2, at 59.
[56] Id. at 32, 58–59.
[57] See Policy Statement, supra note 2, at 37, 46, 50 n.26; Eric Fruits et al., ICLE Comments on DOJ/FTC Guidance on Business Collaborations §§ IV.A–B (Int’l Ctr. for L. & Econ., 2026), https://laweconcenter.org/resources/icle-comments-on-doj-ftc-guidance-on-business-collaborations; Ian Adams et al., ICLE Comments to Competition Bureau Canada on Anti-Competitive Conduct and Agreements (Int’l Ctr. for L. & Econ., 2026), https://laweconcenter.org/resources/icle-comments-to-competition-bureau-canada-on-anti-competitive-conduct-and-agreements.
[58] See Policy Statement, supra note 2, at 39–41, 61.