The DMA Meets the New Intermediaries: AI Agents and the Future of Gatekeeper Regulation
Executive Summary
AI-enabled applications may reshape competitive dynamics and the core forms of digital intermediation on which digital markets depend. That prospect has significant implications for the regulatory regimes recently adopted to govern those markets. Because these instruments were designed without AI specifically in mind, they risk becoming outdated within a short time.
The Digital Markets Act (DMA) illustrates the problem. The regulation reflects a Big Tech-centered understanding of digital markets, focused on entrenched platform ecosystems, vertically integrated gatekeepers, and self-preferencing risks. Those concerns remain important: incumbent firms still control key inputs and access points, including cloud infrastructure, data, distribution channels, and core platform services.
But AI also complicates the DMA’s premises. Downstream AI markets have so far shown substantial dynamism, with new entrants attracting significant investment and developing services that may disrupt established positions. Assistive and agentic AI could also create new gateways through which users access information, goods, services, and content—without necessarily fitting within the DMA’s existing categories of core platform services or replicating the economic logic of traditional digital platforms.
As a result, only a few years after its enactment, the DMA’s role, rationale, and claimed future-proof character are already under scrutiny. The deployment of AI applications raises a basic question for policymakers: whether the DMA can adapt through interpretation and targeted adjustment, or whether its legislative framework must be reopened.
Against this backdrop, and in the context of the DMA’s first review, this paper argues that the rise of AI applications calls for reconsidering the regulation’s overall architecture. The challenge is not merely to fine-tune existing categories, but to develop a competition-policy framework capable of addressing AI-driven markets on their own terms—while avoiding premature ex ante intervention that could chill innovation before market structures have stabilized.
I. Introduction
The rapid diffusion of artificial-intelligence technologies has brought considerable attention to their capacity to transform—and potentially disrupt—existing competitive dynamics. In this context, competition-policy debates increasingly ask whether existing tools can both preserve the conditions for innovation and address anticompetitive concerns arising from AI deployment.[1]
Similar debates have accompanied earlier technological revolutions. But the current debate differs in one important respect: it extends not only to competition law in its conventional form, but also to the newer regulatory regimes adopted to govern digital markets and ecosystems.
Those regimes rest on the premise that traditional antitrust enforcement, because of its ex post character, often moves too slowly for fast-moving and complex digital markets. By the time intervention occurs, market power may already have become entrenched and difficult to dislodge.[2]
The rapid emergence of AI, however, casts doubt on the future-proof character of these ex ante frameworks and, to some extent, on their underlying rationale. Because regulators designed these instruments without AI specifically in mind, they risk becoming outdated within a short time frame.
These concerns are especially acute with respect to assistive and agentic AI, which may reshape core digital-intermediation functions, including web browsing, online search, and e-commerce. Users increasingly rely on applications powered by large foundation models (FMs), which enable more natural interaction with digital environments, facilitate the retrieval and synthesis of information, support task automation and decision-making, and allow users to delegate bounded autonomous actions across interconnected platforms and services.
As a result, AI-enabled assistants and agents are emerging as a potentially significant locus of competitive power. They increasingly operate as full-fledged interfaces through which users access third-party goods, services, and content without leaving the conversational environment.[3] To the extent that these systems mediate user access to downstream markets, they are likely to assume growing importance in the competitive structure of the digital economy.[4]
This backdrop exposes a tension between the adaptive capacity of competition law and the rigidity of sector-specific regulation. Competition law may operate slowly, but it is structurally flexible. Its open-ended standards can accommodate shifting market realities over time. Bespoke regulatory regimes, by contrast, may deliver more immediate and targeted responses, but they are more vulnerable to technological change and regulatory obsolescence.
The European Union’s Digital Markets Act (DMA) illustrates this tension.[5] Despite its recent entry into force and still-nascent implementation, the regime already risks rapid aging because it was designed against the background of a digital environment that AI-driven intermediation may materially alter.
In the DMA’s first review, the European Commission launched a consultation to assess whether the regulation is adequately equipped to address the rollout of AI-powered services and whether the list of core platform services and related obligations may need adjustment.[6]
At the heart of the issue lies the respective market positions of new AI entrants. When existing gatekeepers incorporate AI-enabled functionalities into their ecosystems, those services may fall within the DMA’s scope. But the regulation does not extend to new AI operators, regardless of their market significance or growth potential, if standalone AI applications do not fit within one of the core platform-service categories enumerated in the DMA.
As a result, unless AI services can be subsumed within existing categories of core platform services—such as online search engines, web browsers, or virtual assistants—providers cannot be designated as gatekeepers solely on the basis of their AI offerings. The literature has therefore begun to divide over whether this regulatory gap can be bridged through an expansive interpretation of the existing framework or instead requires reopening the legislative process.[7]
The European Commission itself appears conscious of the DMA’s limits and of the comparative advantages of traditional antitrust enforcement. It recently stated that the “DMA will not be able to tackle every competition issue in the AI value chain” and that, for novel conduct falling outside the DMA’s obligations, competition-law enforcement must serve as a complementary tool.[8]
Against this background, this paper advances a different perspective: the emergence of agentic AI calls for reconsidering the DMA’s overall architecture, not merely fine-tuning its margins. The regulation was designed for a fundamentally different technological and economic environment, in which specific structural features informed the identification of both gatekeepers and core platform services. In particular, the DMA reflects a competition-policy paradigm shaped largely by a Big Tech-centered understanding of digital markets.[9]
But there is no reason to assume AI will replicate the developmental trajectory of earlier digital technologies. To the contrary, important differences separate established digital platforms from FMs.[10]
Traditional competitive concerns remain. They stem especially from constrained access to upstream inputs, including computing power, data, and highly skilled labor, as well as from the strong incentives large incumbent digital conglomerates have to integrate AI functionalities into their existing core products and services.[11]
At the same time, competition in downstream AI markets appears vibrant and intense. New entrants have proliferated, their services have achieved commercial success, and they have attracted unprecedented levels of investment. Many barriers to entry for AI developers therefore appear less significant than initially anticipated, while the diversity of downstream applications suggests that FMs are unlikely to produce uniform winner-takes-all outcomes across sectors.
The rapid expansion of AI services and the multiplicity of market participants instead point to a significant degree of dynamism and contestability. It therefore remains uncertain whether incumbent digital firms will be able to entrench or extend their market power in AI-related markets.
If AI systems do not exhibit the same technological and economic characteristics as established digital platforms—and if they can reshape market dynamics, unsettle established positions, and foster new business models and novel forms of ecosystem competition—they should not be treated merely as an extension of earlier digital technologies. The rise of AI may require not marginal adaptation of existing regulatory categories, but a distinct competition-policy framework capable of addressing the technological and economic logic of AI-driven markets.
The paper proceeds as follows. Section II sets out the economic and legal premises underlying the DMA’s enactment and explains how that background may support applying the regulation, in its current form, to AI markets. Section III examines the economic features of the AI value chain, with particular attention to its similarities to, and differences from, traditional digital markets. It also shows how certain distinctive technological and economic characteristics of AI systems, together with the emergence of assistive and agentic AI, depart from the DMA’s original framework. Section IV addresses the strategic policy choice facing the European Commission as it seeks to keep the DMA fit for purpose in the age of AI. Section v concludes.
II. The DMA’s Regulatory Logic and the AI Challenge
AI has unsurprisingly taken center stage in the European Commission’s consultation on the effectiveness of the Digital Markets Act (DMA).[12] In the context of the regulation’s review, a substantial number of respondents raised concerns about AI and large language models (LLMs), noting their growing significance in digital markets.[13]
Before assessing whether the DMA is suited to address AI, it is useful to revisit the legal and economic context surrounding the regulation’s enactment, as well as the policy objectives that shaped it.
The DMA represents the first—and most significant—response to the broader international debate over whether existing antitrust rules and enforcement tools can address the rise of large technology platforms and effectively scrutinize their practices and business models. The regulation rests expressly on the premise that competition law, standing alone, cannot respond effectively to the challenges and systemic concerns posed by the platform economy.
On this view, antitrust rules have limited reach. They apply only to certain forms of market power, such as dominance in a relevant market, and to specific instances of anticompetitive conduct.[14] Their enforcement is also predominantly ex post, requiring extensive case-by-case investigation of often highly complex facts.[15] Competition law also does not—or at least does not adequately—address risks to the proper functioning of markets that stem from gatekeeper conduct falling short of dominance in the antitrust sense.[16]
For these reasons, lawmakers considered regulatory intervention necessary to complement traditional antitrust enforcement. The DMA therefore introduced a set of ex ante obligations for online platforms designated as gatekeepers, dispensing with the need to define relevant markets, establish dominance, and demonstrate anticompetitive effects.
The DMA identifies several economic characteristics that may allow the provision of a service to confer a gatekeeping position on its provider. These include extreme economies of scale, very strong network effects, multisidedness, lock-in effects, vertical integration, and limited multi-homing.[17] According to the regulation, these features have enabled a small number of large undertakings to emerge as gatekeepers: firms with considerable economic power and a significant impact on the internal market, often operating as vertically integrated actors that provide gateways through which business users reach end users across markets.[18]
This position allows such firms to leverage their advantages—including privileged access to data—from one area of activity into another. Some exercise control over entire platform ecosystems and are structurally difficult to challenge.[19] The combination of these market features may also generate significant bargaining-power imbalances, allowing gatekeepers to impose terms and conditions unilaterally to the detriment of both business users and end users.[20]
Gatekeepers may also occupy a dual role, acting both as intermediaries for third-party undertakings and as direct suppliers of products and services. In those circumstances, they may have both the ability and the incentive to favor their own offerings over those of rivals.[21]
Against this backdrop, the DMA establishes a catalogue of obligations that can be read as a condensed restatement of issues previously addressed in antitrust proceedings brought by European competition authorities.[22] Nearly every obligation can be linked to a specific practice adopted by a particular Big Tech firm and scrutinized by an antitrust authority.
Within this framework, many obligations seek to curb self-preferencing, a practice that has come to symbolize contemporary competition-policy concerns with online markets and is often portrayed as a pervasive feature of digital ecosystems.[23] On this view, self-preferencing enables Big Tech firms to consolidate their positions in core markets and extend them into neighboring markets, further strengthening the boundaries of their ecosystems.
This policy choice—centered on detailed and inherently backward-looking rules—appeared from the outset to sit uneasily with the complexity and rapid evolution of digital markets. To address that concern and preserve some measure of future-proofing, the regulation authorizes the Commission to revise the list of core platform services and relevant practices following a market investigation.[24]
Despite AI’s disruptive potential and the apparent success of new entrants, competition policy has so far remained primarily focused on the competitive risks associated with incumbent Big Tech firms. More specifically, policymakers worry that AI may consolidate the power of existing large digital ecosystems and drive markets toward concentration around a limited number of actors.[25]
That concern stems from the view that incumbent digital firms hold significant market power across several layers of the AI stack. This position may allow them both to insulate themselves from AI-driven disruption and to leverage preexisting positions into adjacent and nascent markets.[26] On this account, AI markets exhibit features comparable to those of traditional digital markets, including economies of scale and scope, network effects, data-feedback loops, and limited multi-homing. These characteristics may generate winner-takes-all or winner-takes-most dynamics, making such markets particularly prone to tipping in favor of a small number of dominant firms.[27]
Competition authorities have therefore directed growing scrutiny toward the strategies adopted by Big Tech firms, reflecting distrust of both partnership arrangements with emerging AI developers and the integration of AI solutions into core products and services.
On the one hand, such partnerships may produce procompetitive effects by giving new developers access to financial resources, distribution channels, and critical inputs—including specialized computing capacity—needed to accelerate innovation and support the development and deployment of AI systems.[28] But they may also neutralize or eliminate nascent competitive constraints.[29]
On the other hand, embedding generative AI (GenAI) solutions within the core products and services of large digital ecosystems raises self-preferencing concerns.[30] These concerns may arise through tying practices, when access to or use of products and services is conditioned on adopting the platform’s own AI solutions. They may also arise through preferential treatment, including pre-installation, exclusive integration, or reduced interoperability.
III. Rethinking the AI Value Chain
This Big Tech-centric approach is consistent with the DMA’s assessment of competitive risks. But it offers only a partial account of the competitive landscape because it assumes that AI markets will replicate the developmental trajectory previously observed in digital markets.
To be clear, concerns about access to critical upstream inputs, as well as the presence of Big Tech firms across the AI value chain—particularly the vertical integration of some firms across the infrastructure, model, and application layers—cannot be dismissed.
The AI stack is not linear. It is a multilayered and interconnected ecosystem in which computational, infrastructural, and application-based components depend on one another.[31] At its foundation lies the hardware layer, consisting of large-scale clusters of graphics processing units (GPUs) and tensor processing units (TPUs), equipped with specialized accelerator chips capable of processing vast quantities of data and billions of parameters in parallel.
Closely linked to this is the layer of cloud and edge infrastructure and services, through which much of the development, training, fine-tuning, and deployment of advanced AI models takes place. These services allow firms to access computational capacity flexibly and scale investment according to operational needs.
The value chain also includes data centers, where AI-system requests are processed and which depend, in turn, on underlying communications networks and related services. At the core of generative AI (GenAI) are foundation models (FMs): large deep-learning models pre-trained to generate specific forms of content and capable of adaptation to a wide range of downstream tasks and uses.
These models are supported by model hubs and machine-learning tools, which perform essential functions such as hosting, curating, fine-tuning, and managing models, thereby facilitating the development of downstream applications. Additional relevant layers include electronic-communications networks and services, which enable data transmission and user interaction with AI systems, and applications that incorporate FMs either as general-purpose tools or in forms tailored to specific use cases.
In this context, the development of AI—and especially FMs and GenAI—depends on four essential building blocks: computing power, data, financial resources, and technical expertise.[32] Antitrust authorities are therefore concerned that control over these key inputs may confer substantial competitive advantages and, where access is exclusive or privileged, create significant barriers to entry.[33] In their view, ecosystem effects and vertical integration amplify these risks because firms active across multiple layers of the AI value chain may leverage market power from adjacent markets into AI-related activities.
Access to cloud infrastructure and distribution channels has emerged as a central competitive concern.[34] Cloud services, which are indispensable for the training, fine-tuning, and deployment of advanced AI models, remain highly concentrated in the hands of a small number of providers.[35] At the same time, large digital platforms increasingly integrate AI functionalities into their ecosystems. This may turn incumbent platforms into key gateways to end users, reducing third parties’ ability to offer competing or customized services, increasing switching costs, and creating opportunities for self-preferencing, foreclosure, and user lock-in.[36]
Access to data raises comparable concerns. Public datasets remain important for model development, but proprietary and first-party data—particularly user-interaction data—may prove decisive for contestability in AI markets.[37] Large technology firms are uniquely positioned to combine exclusive contractual arrangements with privileged access to vast quantities of data generated through their existing services. Network effects, economies of scale and scope, and feedback loops may reinforce these advantages, as AI-system use generates additional data that further improves model performance and entrenches incumbent positions.
Against this background, more alarmist perspectives find “barriers to entry everywhere” and predict that AI’s disruptive force will originate from within “the nest of existing tech ecosystems.”[38]
Yet the literature has pointed out that some technological and economic features of AI systems diverge sharply from those of traditional digital markets.[39] Network effects in FMs, in particular, appear substantially weaker than those typically observed in platform markets because individual users derive little direct benefit from the participation of additional users.
While data-feedback loops matter in both platform and FM markets, data are becoming increasingly abundant, and their relative strategic value appears to be declining. Significantly, antitrust authorities themselves have acknowledged that technological developments—including the growing use of smaller and more efficient datasets, as well as synthetic data—are reducing dependence on large volumes of proprietary data and computing power.[40]
Partnership agreements have also enabled new AI entrants to secure access to critical upstream inputs. In addition, the market valuation of AI services offered by new entrants has risen rapidly and consistently, as shown by their ability to secure substantial investment over successive funding rounds. Those investments reflect strong expectations about the long-term economic impact and commercial viability of these emerging firms.[41]
Moreover, concerns arising from Big Tech firms’ vertical integration are already being addressed through DMA enforcement. The interoperability measures imposed on Google rest on the premise that interoperability with key capabilities of Google Android is essential for third-party AI service providers to compete effectively with Alphabet’s own AI-enabled services, including Gemini.[42]
Finally, contrary to some assumptions, large technology companies have so far appeared largely unable to convert their extensive data advantages into decisive competitive superiority over AI startups. If incumbent gatekeepers had been able seamlessly to transfer their existing dominance into emerging AI markets, the rise of firms such as OpenAI and Anthropic would have been far less likely.
Many of the anticipated barriers to entry for AI developers therefore appear less significant than initially presumed, while the diversity of downstream applications suggests that FMs are unlikely to produce winner-takes-all dynamics across sectors. The rapid expansion of AI markets, together with the continued success of new entrants, casts doubt on the central assumptions informing current competition-policy debates. Taken together, these developments suggest that competitive pressures remain robust and that AI services retain substantial disruptive potential vis-à-vis established market positions.
At the downstream level, AI-enabled applications are transforming competitive dynamics and reconfiguring organizational processes across a wide range of business functions. AI assistants and AI agents are especially likely to alter core forms of digital intermediation, including web browsing, online search, and e-commerce.[43]
AI assistants generally operate as reactive, language-based systems that process user inputs and perform tasks such as content generation or question answering within the limits set by the prompt. Agentic systems, by contrast, exhibit greater autonomy and proactivity. They can independently initiate actions, interact with external software and online environments, and adapt their conduct over time to achieve predefined objectives.[44]
Rather than functioning merely as standalone chatbots, AI assistants and agentic systems are increasingly evolving into fully fledged platforms that allow users to access third-party services directly without leaving the conversational interface.[45] This development appears in the rise of AI-powered web browsers, such as OpenAI’s ChatGPT Atlas and Perplexity’s Comet, as well as in OpenAI’s introduction of “apps in ChatGPT,” which allow users to browse for and purchase products directly through ChatGPT, repositioning it as a digital shopping hub.[46]
These economic features of AI markets and the emergence of AI-enabled applications—developed primarily by new entrants—raise significant questions about whether the DMA’s current framework can address potential competition issues in AI.
If certain technological and economic characteristics of AI systems differ from those of traditional digital markets, competitive dynamics in AI may not replicate the developmental trajectory of earlier digital technologies. Those differences have direct implications for the identification of core platform services under the DMA, which in turn determines gatekeeper designation.
In other words, if AI services do not share the economic features of the core platform services currently listed in the DMA, revising or adapting the regulation cannot be accomplished simply by adding new services to the existing list. That conclusion inevitably affects gatekeeper designation, which remains closely tied to the characteristics of the relevant services.
The DMA is fundamentally animated by the goal of constraining vertically integrated firms that occupy a gatekeeping position and operate in a dual role. For that reason, a substantial share of its obligations targets the leveraging risks associated with self-preferencing. But that framework does not fit easily with the profile of new AI entrants.
Expectations regarding AI assistants and agents indicate that these applications may emerge as new gateways, distinct from those identified in the DMA. They may, indeed, place AI providers in a potential gatekeeping position without requiring vertical integration or preferential treatment.
This point is further borne out by the fact that, despite some vertical integration along the AI stack, most Big Tech firms do not appear to be serious competitors in the critical FM layer. Following the restructuring of its longstanding partnership with OpenAI, Microsoft has only recently begun developing proprietary in-house models.[47] Apple and Amazon, through their respective partnerships with Google and OpenAI, appear largely to have chosen to remain outside the FM market.[48] Meta, meanwhile, does not appear to have invested on a scale comparable to the most successful new AI entrants.
IV. The Policy Choice: Adaptation, Extension, or Restraint
Given AI’s strategic importance in digital markets, the DMA’s first review has inevitably brought to the forefront whether the regulation is sufficiently future-proof and remains fit for purpose in the age of AI.[49] As part of that process, the European Commission has consulted stakeholders on whether the DMA is adequately equipped to address the deployment of FMs and AI-powered services, and whether adjustments to the regulatory framework may be necessary.
At its core, the debate over DMA reform turns on the role to be assigned to new AI entrants.
AI services may fall within the DMA’s scope through two main avenues. First, an AI provider may itself be subject to the regime if it offers a core platform service and satisfies the criteria for designation as a gatekeeper. Second, AI functionalities may be embedded within core platform services that have already been designated, in which case those functionalities are indirectly governed by the obligations imposed on the relevant service.
The latter scenario does not appear to raise novel difficulties for applying the DMA, in its current form, to AI markets. Competition concerns arising in this context—namely, gatekeepers’ structural advantages stemming from privileged access to cloud infrastructure and data, as well as the integration of AI assistants and agents into their core platform services—can be addressed through the regulation’s existing provisions.
A recent specification proceeding initiated by the European Commission against Google confirms this point.[50] In that proceeding, the Commission seeks to ensure, first, that Google affords third-party AI-service providers access to the same Android operating-system features and functionalities available to its own services; and second, that it grants third-party providers of online search engines, including AI-chatbot providers offering search functionalities, access on fair, reasonable, and nondiscriminatory (FRAND) terms to anonymized ranking, query, click, and view data. The Commission is also monitoring whether the integration of AI Overviews within Google Search complies with the DMA, particularly Article 6(5)’s prohibition on self-preferencing.[51]
By contrast, the DMA does not, as such, apply to new AI operators, regardless of their market position, if their standalone AI applications do not fall within any of the core platform-service categories enumerated in the regulation. The DMA allows the Commission to designate as gatekeepers undertakings that, although not yet enjoying an entrenched and durable position, are expected to attain such a position in the near future.[52] But even when foreseeable developments are considered, that assessment may be undertaken only with respect to services falling within the regulation’s category of core platform services.[53] Accordingly, unless AI services can be brought within one of the existing core platform-service categories, providers cannot be designated as gatekeepers solely by virtue of their AI offerings.
Questions about the proper legal status of AI services arise not only under the DMA, but also under other significant components of the European digital regulatory framework, such as the Digital Services Act (DSA).[54] GenAI applications may display platform-like characteristics, operate as intermediaries, and offer functionalities comparable to search engines, while still resisting classification within preexisting legal categories.[55]
In the DMA context, the High-Level Group has expressly acknowledged—when considering AI’s impact on the contestability and fairness of digital markets—that these technologies may both entrench existing gatekeepers and give rise to new ones.[56] Particular attention has focused on AI agents because they may emerge as new gateways, challenging the gatekeeping position of players that currently control the main access points, such as app stores, operating systems, search engines, and online marketplaces.
The literature has advanced two opposing approaches to preserving the DMA’s adaptability. Some scholars argue that policymakers should reopen the legislative framework by swiftly initiating a market investigation under Article 19 to determine whether new categories of AI-related core platform services should be introduced.[57] Others contend that the regulation, in its current form, can already address agentic AI because its wording permits AI applications to be brought within existing listed-service categories, most notably virtual assistants.[58]
As shown above, however, the rise of AI-enabled applications—and the potential transformation they may bring to competitive dynamics and the core organizational forms of digital intermediation—calls into question the DMA’s foundational premises. The regulation reflects a competition-policy paradigm shaped by a Big Tech-centered conception of digital markets. If AI systems do not exhibit the same technological and economic characteristics as established digital platforms, and if AI markets are unlikely to follow the developmental trajectory previously observed in digital markets, then DMA review calls not for mere fine-tuning, but for a distinct competition-policy framework.
At the same time, the DMA’s own logic supports its extension to new AI players. The regulation rests on the premise that conventional antitrust enforcement in digital markets may come too late—after market positions have become entrenched and difficult to challenge. That premise may also apply to AI technologies, given their potential to generate new gatekeepers.[59] Unless European policymakers are willing to repudiate the strategy that has characterized the Brussels approach to digital markets, they must consider activating the DMA as the principal body of law governing competitive dynamics in digital markets in the age of AI.
Still, the risks of premature regulatory intervention deserve careful attention. The AI sector remains at a comparatively early stage of development and has not yet reached a stable phase with respect to technological trajectories, business models, or market structures. In such circumstances, extending ex ante regulation to FMs and AI services by classifying them as core platform services may unduly inhibit innovation, restricting developers’ ability to experiment with alternative functionalities, design configurations, and platform architectures.
Such a strategy would also require policymakers to anticipate both the pace and direction of technological change in markets marked by rapid evolution and substantial uncertainty. Those forward-looking judgments are inherently difficult and heighten the risk of mischaracterizing market power—whether by underestimating emerging competitive constraints or overestimating the extent and persistence of market concentration. These errors may carry significant consequences for innovation, competition, and consumer welfare.
Policymakers therefore must confront not only whether to intervene under the DMA, but when. The challenge is to act early, but not too early.
The European Commission does not appear to share these concerns, at least not fully. In the DMA’s first review, the Commission has thus far opted for a cautious approach.[60] At this stage, it considers the DMA fit for purpose and not in need of amendment, taking the view that the regulation, in its current form, has proved well-suited to a fast-changing environment and able to keep pace with technological developments such as AI.
Yet the Commission’s focus remains directed primarily at the strategies of existing gatekeepers and the deployment of AI tools within designated core platform services. From the Commission’s perspective, then, if the future described in the previous pages is taking shape, it is doing so slowly and is not expected to materialize in the near term. Time will tell which prediction proves correct.
V. Conclusion
Only a few years after its enactment, the Digital Markets Act’s role, rationale, and claimed future-proof character are already under scrutiny. By challenging existing competitive dynamics, AI may rapidly render obsolete the regulatory initiatives recently adopted to govern digital markets.
The DMA’s first review therefore offers more than an occasion to evaluate enforcement results or assess the regulation’s early impact. It requires policymakers to confront a deeper question about the regime’s future. The issue is not simply what has worked, what has failed, or which obligations require adjustment. The rise of AI applications—and the role they may play in reshaping digital competition—raises the broader question whether, and to what extent, the DMA’s legislative framework should be reopened.
The stakes are significant. The recent wave of digital regulation has been shaped by a Big Tech-centered conception of digital markets, focused on entrenched platform ecosystems, vertically integrated gatekeepers, and self-preferencing risks. AI does not eliminate those concerns. Incumbent firms still control critical inputs, including cloud infrastructure, data, distribution channels, and key ecosystem access points. Existing gatekeepers may use those advantages to incorporate AI into core platform services and reinforce their positions.
But AI also complicates the DMA’s premises. New AI entrants have already shown that downstream AI markets can be dynamic, well-funded, and contestable. Assistive and agentic AI may create new gateways to users, content, goods, and services that do not fit neatly within the DMA’s existing categories of core platform services. These systems may exercise intermediation power without replicating the same economic logic as traditional digital platforms—or the same vertically integrated, dual-role structure that the DMA was built to constrain.
That possibility exposes the central tension running through the DMA. Competition law may act slowly, but its open-ended standards allow it to adapt to technological change. Sector-specific regulation can move faster and impose more targeted obligations, but it risks becoming obsolete when markets evolve in unexpected ways. The DMA’s reliance on detailed and inherently backward-looking rules raised this concern from the outset. AI makes it harder to ignore.
For that reason, the limits of the current regulatory framework should be understood not only as a challenge, but also as an opportunity to rethink the DMA’s architecture. The choice is not between complacency and reflexive expansion. Premature ex ante regulation could chill experimentation in a market whose technological trajectories, business models, and competitive structure remain unsettled. But waiting too long may allow new gatekeeping positions to harden before policymakers respond.
The European Commission has, for now, chosen caution. It acknowledges that AI warrants special vigilance, but remains largely committed to the DMA’s existing Big Tech-centered approach and to monitoring the deployment of AI tools within designated core platform services. That choice may buy time. It does not resolve the underlying dilemma.
If AI markets develop along familiar lines, the DMA may prove sufficiently adaptable. If they instead produce new forms of intermediation, new sources of competitive power, and new gatekeepers outside the regulation’s existing categories, the DMA will require more than fine-tuning. It will require a competition-policy framework capable of addressing AI-driven markets on their own terms.
[1] See, e.g., Bertin Martens, Why Artificial Intelligence Is Creating Fundamental Challenges for Competition Policy, Bruegel Pol’y Brief (July 18, 2024), https://www.bruegel.org/policy-brief/why-artificial-intelligence-creating-fundamental-challenges-competition-policy.
[2] See, e.g., Regulation (EU) 2022/1925 of the European Parliament and of the Council of 14 September 2022 on Contestable and Fair Markets in the Digital Sector and Amending Directives (EU) 2019/1937 and (EU) 2020/1828 (Digital Markets Act), recital 5, 2022 O.J. (L 265) 1.
[3] See, e.g., Autorité de la Concurrence, Conversational Agents: the Autorité Starts Inquiries Ex Officio with a View to Issuing an Opinion (2026), https://www.autoritedelaconcurrence.fr/en/press-release/conversational-agents-autorite-starts-inquiries-ex-officio-view-issuing-opinion.
[4] S See, e.g., Org. for Econ. Co-operation & Dev. (OECD), Artificial Intelligence and Competitive Dynamics in Downstream Markets (2025), https://www.oecd.org/en/publications/artificial-intelligence-and-competitive-dynamics-in-downstream-markets_ccf0624a-en.html.
[5] Digital Markets Act, supra note 2.
[6] Pursuant to Article 53 of the Digital Markets Act, the European Commission must evaluate the regulation by May 3, 2026, and every three years thereafter, and report its findings to the European Parliament, the Council, and the European Economic and Social Committee. See Eur. Comm’n, Consultation on the First Review of the Digital Markets Act (2025), https://digital-markets-act.ec.europa.eu/consultation-first-review-digital-markets-act_en; Eur. Comm’n, Review of the Digital Markets Act, Call for Evidence, Ares(2025)6881572 (2025), https://eur-lex.europa.eu/legal-content/EN/TXT/?uri=intcom:Ares(2025)6881572.
[7] See, e.g., Friso Bostoen & Jan Kramer, How Future-Proof Is the DMA? A Case Study of AI Agents, J. Competition L. & Econ. (forthcoming); Peter Georg Picht, DMA–Innovation (2026), https://papers.ssrn.com/sol3/papers.cfm?abstract_id=6409478; Jan-Frederick Gohsl, Future Proofing the DMA for Agentic AI: Lessons from the AI Act, 48 World Competition 315 (2025); Ayse Gizem Yasar, Andrew Chong, Evan Dong, Thomas Krendl Gilbert, Sarah Hladikova, Carlos Mougan, Xudong Shen, Shubham Singh, Ana-Andreea Stoica & Savannah Thais, Integration of Generative AI in the Digital Markets Act: Contestability and Fairness from a Cross-Disciplinary Perspective, LSE L., Soc’y & Econ. Working Papers No. 4 (2024), https://papers.ssrn.com/sol3/papers.cfm?abstract_id=4769439; Philipp Hacker, Johann Cordes & Janina Rochon, Regulating Gatekeeper Artificial Intelligence and Data: Transparency, Access and Fairness under the Digital Markets Act, the General Data Protection Regulation and Beyond, 15 Eur. J. Risk Reg. 49 (2024); Andreas Schwab, Digital Markets Act and Artificial Intelligence Services, 3 Concurrences 1 (2024).
[8] Eur. Comm’n, Comm’n Staff Working Document Accompanying the Report on the Review of Regulation (EU) 2022/1925 of the European Parliament and of the Council on Contestable and Fair Markets in the Digital Sector and Amending Directives (EU) 2019/1937 and (EU) 2020/1828 (Digital Markets Act), in Accordance with Article 53 Thereof, SWD(2026) 123 final, at 48.
[9] See, e.g., Marco Cappai & Giuseppe Colangelo, Taming Digital Gatekeepers: the More Regulatory Approach to Antitrust Law, 41 Computer L. & Sec. Rev. 105559 (2021).
[10] See, e.g., Andrei Hagiu & Julian Wright, Artificial Intelligence and Competition Policy, 103 Int’l J. Indus. Org. 103134 (2025); Anton Korinek & Jai Vipra, Concentrating Intelligence: Scaling and Market Structure in Artificial Intelligence, 40 Econ. Pol’y 227 (2025); Zach Meyers & Marc Bourreau, A Competition Policy for Cloud and AI, Ctr. on Regul. in Eur. (2025), https://cerre.eu/publications/a-competition-policy-for-cloud-and-ai; Thibault Schrepel & Alex “Sandy” Pentland, Competition Between AI Foundation Models: Dynamics and Policy Recommendations, 34 Indus. & Corp. Change 1085 (2025); Catherine Tucker, How Does Competition Policy Need to Change in a World of Artificial Intelligence?, 40 Oxford Rev. Econ. Pol’y 834 (2024).
[11] See, e.g., Austl. Competition & Consumer Comm’n, Digital Platform Services Inquiry—Final Report (2025), https://www.accc.gov.au/inquiries-and-consultations/digital-platform-services-inquiry-2020-25/march-2025-final-report; Org. for Econ. Co-operation & Dev. (OECD), Competition in Artificial Intelligence Infrastructure (2025), https://www.oecd.org/en/publications/competition-in-artificial-intelligence-infrastructure_623d1874-en.html; Autorité de la Concurrence, Opinion on the Competitive Functioning of the Generative Artificial Intelligence Sector (2024), https://www.autoritedelaconcurrence.fr/en/press-release/generative-artificial-intelligence-autorite-issues-its-opinion-competitive; Korea Fair Trade Comm’n, Generative AI and Competition (2024), https://www.ftc.go.kr/viewer/synap/skin/doc.html?fn=BBS_202502130426599090&rs=/viewer/synap/preview/; Klaus Kowalski, Cristina Volpin & Zsolt Zombori, Competition in Generative AI and Virtual Worlds, Eur. Comm’n Competition Pol’y Brief No. 3 (2024), https://op.europa.eu/en/publication-detail/-/publication/5530c8ca-7a1f-11ef-bbbe-01aa75ed71a1/language-en; UK Competition & Mkts. Auth., AI Foundation Models—Updated Paper (2024), https://www.gov.uk/government/publications/ai-foundation-models-update-paper; Autoridade da Concorrência, Competition and Generative Artificial Intelligence (2023), https://www.concorrencia.pt/en/articles/adc-warns-competition-risks-generative-artificial-intelligence-sector.
[12] Eur. Comm’n, Consultation on the First Review of the Digital Markets Act, supra note 6.
[13] Eur. Comm’n, DMA Review—Summary of the Contributions to the Targeted Consultation (2026), https://digital-markets-act.ec.europa.eu/commission-publishes-summary-and-responses-consultation-ongoing-review-digital-markets-act-2026-01-08_en.
[14] Digital Markets Act, supra note 2, rec. 5.
[15] Id.
[16] Id.
[17] Id., rec. 13.
[18] Id., rec. 3, 6, 51.
[19] Id., rec. 3.
[20] Id., rec. 4.
[21] Id., recs. 46, 57.
[22] See, e.g., Cristina Caffarra & Fiona Scott Morton, The European Commission Digital Markets Act: A Translation, VoxEU (Jan. 5, 2021), https://voxeu.org/article/european-commission-digital-markets-act-translation.
[23] See, e.g., Giuseppe Colangelo, Antitrust Unchained: The EU’s Case Against Self-Preferencing, 72 GRUR Int’l 538 (2023).
[24] Digital Markets Act, supra note 2, art. 19.
[25] See, e.g., Margrethe Vestager, Speech, Making Artificial Intelligence Available to All—How to Avoid Big Tech’s Monopoly on AI? (Feb. 18, 2024), https://ec.europa.eu/commission/presscorner/detail/en/speech_24_931.
[26] See, e.g., Autorité de la Concurrence, supra note 11; Can. Competition Bureau, Artificial Intelligence and Competition: Discussion Paper (2024), https://publications.gc.ca/site/eng/9.935280/publication.html; Press Release, Eur. Comm’n, UK Competition & Mkts. Auth., U.S. Dep’t of Just. & U.S. Fed. Trade Comm’n, Joint Statement on Competition in Generative AI Foundation Models and AI Products (2024), https://competition-policy.ec.europa.eu/about/news/joint-statement-competition-generative-ai-foundation-models-and-ai-products-2024-07-23_en; Korea Fair Trade Comm’n, supra note 11; Kowalski et al., supra note 11; UK Competition & Mkts. Auth., supra note 11; Autoridade da Concorrência, supra note 11.
[27] See, e.g., Press Release, Taiwan Fair Trade Comm’n, Taiwan Fair Trade Commission Publishes Public Consultation Report and Competition Policy Statement on Generative AI (2026), https://www.ftc.gov.tw/internet/english/doc/docDetail.aspx?uid=179&docid=18369; Press Release, Eur. Comm’n, Commission Launches Calls for Contributions on Competition in Virtual Worlds and Generative AI (2024), https://ec.europa.eu/commission/presscorner/detail/en/IP_24_85.
[28] See, e.g., Press Release, Anthropic, Anthropic Expands Partnership with Google and Broadcom for Multiple Gigawatts of Next-Generation Compute (2026), https://www.anthropic.com/news/google-broadcom-partnership-compute; Ashley Capoot & Kate Rooney, Google to Invest up to $40 Billion in Anthropic as Search Giant Spreads Its AI Bets, CNBC (Apr. 24, 2026), https://www.cnbc.com/2026/04/24/google-to-invest-up-to-40-billion-in-anthropic-as-search-giant-spreads-its-ai-bets.html; Press Release, OpenAI, OpenAI and Amazon Announce a Strategic Partnership (Feb. 27, 2026), https://openai.com/index/amazon-partnership?utm_source=chatgpt.com.
[29] See, e.g., Press Release, Conselho Administrativo de Defesa Econômica, CADE to Investigate Big Techs’ Acquisitions of AI Startups (Sept. 4, 2024), https://www.gov.br/cade/en/matters/news/cade-to-investigate-big-techs2019-acquisitions-of-ai-startups; Eur. Comm’n, supra note 27; Press Release, UK Competition & Mkts. Auth., CMA Seeks Views on AI Partnerships and Other Arrangements (Apr. 24, 2024), https://www.gov.uk/government/news/cma-seeks-views-on-ai-partnerships-and-other-arrangements; Press Release, U.S. Fed. Trade Comm’n, FTC Launches Inquiry into Generative AI Investments and Partnerships (Jan. 25, 2024), https://www.ftc.gov/news-events/news/press-releases/2024/01/ftc-launches-inquiry-generative-ai-investments-partnerships.
[30] See, e.g., Conselho Administrativo de Defesa Econômica, Cade Abre Inquérito contra Meta e Aplica Medida Preventiva Suspendendo Novos Termos do WhatsApp sobre IA (Jan. 1, 2026), https://www.gov.br/cade/pt-br/assuntos/noticias/cade-abre-inquerito-contra-meta-e-aplica-medida-preventiva-suspendendo-novos-termos-do-whatsapp-sobre-ia; Press Release, Eur. Comm’n, Commission Notifies Meta of Possible Interim Measures to Reverse Exclusion of Third-Party AI Assistants from WhatsApp (Feb. 8, 2026), https://ec.europa.eu/commission/presscorner/detail/en/ip_26_310; Press Release, Autorità Garante della Concorrenza e del Mercato, The Italian Competition Authority Launches Investigation into Meta over Abuse of Dominant Position (July 30, 2025), https://en.agcm.it/en/media/press-releases/2025/7/A576.
[31] See, e.g., Hagiu & Wright, supra note 10; OECD, supra note 11.
[32] See, e.g., U.S. Fed. Trade Comm’n, Generative AI Raises Competition Concerns (June 29, 2023), https://www.ftc.gov/policy/advocacy-research/tech-at-ftc/2023/06/generative-ai-raises-competition-concerns.
[33] See, e.g., Press Release, High-Level Group for the Digital Markets Act, Joint Paper on Artificial Intelligence—Mapping out Regulatory Interplay Related to AI Issues (Dec. 12, 2025), https://digital-markets-act.ec.europa.eu/fifth-meeting-digital-markets-act-high-level-group-2025-12-12_en; Autorité de la Concurrence, supra note 11; UK Competition & Mkts. Auth., supra note 11.
[34] See, e.g., Press Release, Autorité de la Concurrence, Cloud Computing: the Autorité de la Concurrence Issues Its Market Study on Competition in the Cloud Sector (June 29, 2023), https://www.autoritedelaconcurrence.fr/en/press-release/cloud-computing-autorite-de-la-concurrence-issues-its-market-study-competition-cloud.
[35] Press Release, Eur. Comm’n, Commission Launches Market Investigations on Cloud Computing Services under the Digital Markets Act (Nov. 18, 2025), https://digital-markets-act.ec.europa.eu/commission-launches-market-investigations-cloud-computing-services-under-digital-markets-act-2025-11-18_en; see also Judith Arnal, Cloud Competition as a Prerequisite for Competitiveness: Rethinking EU Digital Regulation, J. Eur. Competition L. & Prac. (forthcoming); Antonio Manganelli, Foundation Models and Generative AI Applications: What Competitive Concerns?, 22 Eur. Competition J. 264 (2026); Hagiu & Wright, supra note 10.
[36] See, e.g., Kowalski et al., supra note 11.
[37] See, e.g., High-Level Group for the Digital Markets Act, supra note 33.
[38] Vestager, supra note 25; see also Pierre Azoulay, Joshua Krieger & Abhishek Nagaraj, Old Moats for New Models: Openness, Control, and Competition in Generative Artificial Intelligence, in Entrepreneurship and Innovation Policy and the Economy 7 (Benjamin Jones & Josh Lerner eds., 2025) (arguing that incumbent firms’ control over key complementary assets is likely to produce highly concentrated markets and relegate entrants to the application layer, much as occurred in the smartphone sector.)
[39] See, e.g., Hagiu & Wright, supra note 10; Korinek & Vipra, supra note 10; Meyers & Bourreau, supra note 10; Tucker, supra note 10.
[40] See, e.g., Austl. Competition & Consumer Comm’n, supra note 11; Autoridade da Concorrência, supra note 11; Autorité de la Concurrence, supra note 11.
[41] See, e.g., OpenAI, OpenAI Raises $122 Billion to Accelerate the Next Phase of AI (Mar. 31, 2026), https://openai.com/index/accelerating-the-next-phase-ai; Kate Clark, Anthropic Raising $10 Billion at $350 Billion Value, Wall St. J. (2026), https://www.wsj.com/tech/ai/anthropic-raising-10-billion-at-350-billion-value-62af49f4.
[42] Eur. Comm’n, Case Summary—Case DMA.100220—Alphabet—OS—Google Android—Art. 6(7) Features for AI or AI-Related Services (Apr. 27, 2026), https://digital-markets-act.ec.europa.eu/dma100220-consultation-proposed-measures-interoperability-google-android-article-67-dma_en.
[43] OECD, supra note 4, at 25. See also Nicolas Padilla, H. Tai Lam, Anja Lambrecht & Brett Hollenbeck, The Impact of LLM Adoption on Online User Behavior (2026), https://papers.ssrn.com/sol3/papers.cfm?abstract_id=5393256 (finding that AI tools may significantly alter how users engage with online information and substantially reduce online search activity); Hagiu & Wright, supra note 10 (arguing that AI-based search engines pose the most significant disruptive threat to Google Search in years and that AI agents could disrupt online marketplaces by offering alternative supplier-access channels).
[44] See, e.g., UK Gov’t, Guidance AI Insights: Agentic AI (HTML) (Mar. 13, 2025), https://www.gov.uk/government/publications/ai-insights/ai-insights-agentic-ai-html (describing agentic AI as systems composed of autonomous agents—specialized software programs capable of making decisions and acting independently or collaboratively to achieve defined objectives).
[45] See, e.g., Autorité de la Concurrence, supra note 3.
[46] OpenAI, Introducing Apps in ChatGPT and the New Apps SDK (Oct. 6, 2025), https://openai.com/index/introducing-apps-in-chatgpt?utm_source=chatgpt.com.
[47] See Janakiram MSV, Microsoft Builds Its Own AI Model Stack To Reduce OpenAI Dependence, Forbes (Apr. 2, 2026), https://www.forbes.com/sites/janakirammsv/2026/04/02/microsoft-builds-its-own-ai-model-stack-to-reduce-openai-dependence?utm_source=chatgpt.com; Microsoft, The Next Chapter of the Microsoft–OpenAI Partnership (Oct. 28, 2025), https://blogs.microsoft.com/blog/2025/10/28/the-next-chapter-of-the-microsoft-openai-partnership.
[48] See Press Release, Amazon, Amazon and Anthropic Expand Strategic Collaboration (Apr. 20, 2026), https://www.aboutamazon.com/news/company-news/amazon-invests-additional-5-billion-anthropic-ai; Press Release, Google, Joint Statement from Google and Apple (Jan. 12, 2026), https://blog.google/company-news/inside-google/company-announcements/joint-statement-google-apple; Press Release, OpenAI, OpenAI and Amazon Announce Strategic Partnership (Feb. 27, 2026), https://openai.com/index/amazon-partnership.
[49] Eur. Comm’n, Review of the Digital Markets Act, supra note 6.
[50] Press Release, Eur. Comm’n, Commission Opens Proceedings to Assist Google in Complying with Interoperability and Online Search Data Sharing Obligations under the Digital Markets Act (Jan. 26, 2026), https://ec.europa.eu/commission/presscorner/detail/en/ip_26_202. In late April 2026, the Commission adopted preliminary findings setting out proposed measures that remain subject to public consultation. See Press Release, Eur. Comm’n, Commission Seeks Feedback on Measures to Ensure Interoperability with Google’s Android under the Digital Markets Act (Apr. 26, 2026), https://ec.europa.eu/commission/presscorner/detail/en/ip_26_887.
[51] Press Release, Eur. Comm’n, Commission Sends Preliminary Findings to Alphabet under the Digital Markets Act (Mar. 18, 2025), https://ec.europa.eu/commission/presscorner/detail/en/ip_25_811.
[52] Digital Markets Act, supra note 2, art. 3(1)(c).
[53] Id. art. 3(8).
[54] Regulation (EU) 2022/2065 of the European Parliament and of the Council of 19 October 2022 on a Single Market for Digital Services and Amending Directive 2000/31/EC (Digital Services Act), 2022 O.J. (L 277) 1.
[55] See, e.g., Marco Bassini & Andrea Palumbo, Beyond Intermediaries—Generative AI in Search of Their Legal Status, Verfassungsblog (Apr. 12, 2026), https://verfassungsblog.de/genai-dsa; Lilian Edwards, Igor Szpotakowski, Gabriele Cifrodelli, Joséphine Sangaré & James Stewart, Private Ordering, Generative AI and the “Platformisation Paradigm”: What Can We Learn from Comparative Analysis of Models Terms and Conditions?, 1 Cambridge Forum on AI: L. & Governance 1 (2025). See also Christoph Busch, Enabling Innovation and Protecting Consumers in the Agentic Economy: Why the Digital Fairness Act Should Regulate Agentic AI (2025), https://papers.ssrn.com/sol3/papers.cfm?abstract_id=5369055 (arguing that agentic AI remains a blind spot in debates surrounding the proposed Digital Fairness Act); Eur. Comm’n, Call for Evidence for an Impact Assessment, Ares(2025)5829481, https://ec.europa.eu/info/law/better-regulation/have-your-say/initiatives/14622-Digital-Fairness-Act_en.
[56] Press Release, High-Level Group for the Digital Markets Act, Public Statement on Artificial Intelligence (May 22, 2024), https://digital-markets-act.ec.europa.eu/high-level-group-digital-markets-act-public-statement-artificial-intelligence-2024-05-22_en.
[57] See, e.g., Gohsl, supra note 7; Yasar et al., supra note 7.
[58] See, e.g., Bostoen & Kramer, supra note 7; Picht, supra note 7.
[59] See, e.g., Hagiu & Wright, supra note 10 (noting that AI agents may simply replace one type of gatekeeper with another, as providers with larger consumer bases could develop superior recommendation capabilities through data-driven feedback loops and then charge suppliers substantial access fees).
[60] Eur. Comm’n, Report on the Review of Regulation (EU) 2022/1925 of the European Parliament and of the Council on Contestable and Fair Markets in the Digital Sector and Amending Directives (EU) 2019/1937 and (EU) 2020/1828 (Digital Markets Act), in Accordance with Article 53 Thereof, COM(2026) 178 final, at 9–11. See also Eur. Comm’n, supra note 8, § 4.2.3.