A Critical Analysis of Exclusionary Abuses in Multisided Markets and the Role of the Digital Markets Act

Abstract

Article 102 TFEU, the European Union’s prohibition on abuse of dominance, remains doctrinally capable of addressing exclusionary practices in digital platform markets; yet its institutional limitations, slow proceedings, demanding evidentiary standards, and reactive timing have exposed significant enforcement gaps. This article critically evaluates whether Article 102 remains effective in an era of multisided platforms, generative AI, and the new Digital Markets Act (DMA). The analysis reveals a mixed reality: Article 102 is substantively adequate but practically inadequate. Self-preferencing, tying, anti-steering, and interoperability refusals can be addressed within existing law, as demonstrated by recent cases involving Google, Apple, Meta, and Amazon. However, remedies often arrive after competitive structures have already become established or entrenched. The rise of the DMA complements, rather than replaces, Article 102 by enabling faster ex ante intervention against recurring gatekeeper practices. Generative AI further demonstrates why Article 102’s flexibility remains essential: novel AI-driven foreclosure strategies cannot be anticipated by fixed ex ante obligations. The article concludes that Article 102’s future effectiveness depends on four developments: faster interim intervention, technically informed remedies, close coordination with the DMA, and sustained scrutiny of interoperability.

1. INTRODUCTION

Online platforms have fundamentally reshaped competition. Today, a handful of gatekeepers, Google, Apple, Meta, Amazon, and increasingly AI-driven ecosystems, organise the conditions under which millions of businesses and consumers interact. These platforms are not mere market participants. They are markets themselves, controlling visibility, ranking, pricing terms, and user access.

This transformation has created a paradox in EU competition law. Article 102 TFEU, unchanged in substance since the Treaty of Rome of 1957, must now regulate algorithmic ranking, app store gatekeeping, AI assistant pre-installation, and ecosystem control. Can a legal standard of this age speak meaningfully to digital power? This article argues that it can, but not alone.

2. UNDERSTANDING DOMINANCE IN DIGITAL AND MULTISIDED MARKETS

2.1 The Challenge of Free-to-User Dominance

The fundamental challenge in applying Article 102 to digital platforms is that dominance cannot be assessed through conventional industrial metrics. Google Search, Facebook, and Apple’s App Store offer free services to end users, yet exercise enormous market power. Dominance must instead be understood through platform structure.

A multisided platform acts as an intermediary, enabling interaction between interdependent groups. A search engine connects users and advertisers. An app store connects users with developers. A social-network marketplace connects consumers, traders, and service providers. The platform’s power lies in controlling how user groups interact, through ranking, defaults, visibility, interoperability, and access routes.

2.2 The Feedback Loop: Network Effects and Data Accumulation

Market power compounds through three mechanisms. First, indirect network effects create feedback loops: more users attract more advertisers or developers, which in turn attract more users. Second, economies of scale are profound: once scale is achieved, serving additional users costs little while quality improves. Third, data accumulation creates learning advantages: the dominant platform’s superior dataset enables optimisation that disadvantages rivals.

When users single-home on one platform (as many do on their primary device), business users must be present on that platform regardless of whether they multi-home elsewhere. This competitive bottleneck is structural. The platform becomes not just a market participant but an organiser of the conditions under which others compete.

3. THE FOUR PRINCIPAL EXCLUSIONARY ABUSES

3.1 Self-Preferencing and Leveraging

Self-preferencing occurs when a vertically integrated platform uses control over a key service to favour its own adjacent services while disadvantaging competitors. Google’s comparison shopping service, for example, received more prominent placement in general search results than competing comparison shopping services, whose results were simultaneously demoted by adjustment algorithms.

The significance is that self-preferencing can replace competition on the merits with algorithm-controlled visibility. In Case C-48/22 P, Google and Alphabet v Commission (Google Shopping), the Court of Justice on 10 September 2024 dismissed Google’s appeal and upheld the €2.42 billion fine. The Court characterised Google’s favouring of its own comparison shopping service, coupled with the demotion of rivals, as a form of abusive leveraging rather than a refusal to supply. Crucially, it held that the strict Bronner ‘essential facilities’ conditions did not apply in that context, because the case concerned the conditions of access to Google’s general results page access which competitors already enjoyed, rather than the refusal of an indispensable facility.

It is important not to overstate the ruling. The Court did not create a free-standing, self-contained abuse labelled ‘discriminatory leveraging’, and it did not treat self-preferencing as abusive per se. Such conduct is abusive only where, assessed against all the relevant circumstances, it is at least capable of producing anticompetitive foreclosure effects. What the judgment does confirm is that Article 102 does not mandate the as-efficient-competitor test in this context, allowing the Commission to rely on whether the conduct was capable of excluding competitors. That was itself a meaningful clarification of enforcement standards.

3.2 Tying and Bundling

Tying occurs when a dominant firm uses power in one market to push users toward a related service with no meaningful opt-in choice. Google required manufacturers to pre-install its apps in order to access the Google Play Store. Apple’s in-app billing system is unavoidable for developers. Meta tied its Marketplace directly to Facebook’s social network.

The Google Android case illustrates how tying is treated at the highest level, and its procedural history repays careful statement. By decision of 18 July 2018, the Commission fined Google approximately €4.343 billion for, among other conduct, tying the Google Search and Chrome apps to the Play Store and imposing anti-fragmentation restrictions on device manufacturers. In Case T-604/18, Google and Alphabet v Commission, the General Court on 14 September 2022 upheld the substance of the infringement but partially annulled the decision and reduced the fine to approximately €4.125 billion. Google appealed further, and on 2 July 2026, in Case C-738/22 P, the Court of Justice dismissed the appeal in its entirety and confirmed the €4.125 billion penalty, concluding the ordinary EU judicial appeal process.gal options. The judgments confirm that tying of this kind can be condemned without any need to show the tied product is inherently inferior: it succeeds through ecosystem gravity rather than genuine consumer benefit.

Tying has also re-emerged in social-media markets. By decision of 14 November 2024 (Case AT.40684), the Commission fined Meta €797.72 million for tying its Facebook Marketplace classified-ads service to the Facebook personal social network and for imposing unfair trading conditions on rival online classified-ads providers, an official finding of tying and unfair trading conditions, not merely of self-preferencing. The tie was found to give Marketplace a distribution advantage its competitors could not match, while Meta’s terms allowed it to exploit rivals’ advertising data for the benefit of its own service.

3.3 Refusal of Interoperability

Digital platforms derive value from interconnectivity. Restricting access to technical interfaces (APIs, data, software systems) can become a structural way to block downstream competition. Traditionally, access disputes were governed by the Bronner test: access is required only where it is indispensable and denial would eliminate downstream competition. That threshold is notoriously difficult to meet.

In Case C-233/23, Alphabet and Others (Android Auto), decided on 25 February 2025, the Grand Chamber adapted this framework. The Court held that where a platform has been designed to be ‘open’, that is, to work with third-party applications, a refusal to ensure interoperability, or the grant of only degraded access, can be abusive even where the platform is not strictly indispensable. Google had designed Android Auto to accommodate third-party apps, yet declined to make Enel X’s JuicePass app interoperable.

The ruling should not, however, be read as making every refusal on an ‘open’ platform automatically abusive. The Court expressly preserved objective justifications: a refusal may be lawful where interoperability would compromise the integrity or security of the platform, or where it is technically impossible for other reasons. What it would not accept was justification by mere inconvenience; the absence of a software template, or the effort required to build one, does not on its own excuse a refusal. Where no objective justification exists, the dominant undertaking must develop the necessary template within a reasonable period, though it may charge a fair and proportionate fee. The burden of establishing justification rests on the dominant firm.

3.4 Anti-Steering and Discriminatory Conditions

Anti-steering rules prevent app developers from informing customers of cheaper purchasing options outside an app store. Apple’s App Store forbade developers from including links directing users to external payment methods, despite Apple’s commission being substantially higher than many alternatives. Developers had to accept these rules to reach users.

The Commission fined Apple approximately €1.8 billion on 4 March 2024 (Case AT.40437) for imposing anti-steering restrictions on music-streaming app developers. This marked a shift of emphasis from market exclusion toward direct consumer harm. The case treated anti-steering as an unfair trading condition, using quality-based rather than price-based tests appropriate to free services.

4. THE INSTITUTIONAL PROBLEM: SPEED, PROOF, AND REMEDY

Recent case law demonstrates that Article 102 can reach modern exclusionary conduct. Yet enforcement reveals a critical institutional weakness: proceedings are slow, proof is demanding, and remedies arrive after competitive structures have already tipped.

Google Shopping is instructive. The Commission opened its investigation in 2010, issued its decision in 2017, the General Court upheld it in 2021, and only in September 2024 did the Court of Justice deliver final judgment. Fourteen years elapsed between opening and final judgment. Throughout those years, Google’s comparison shopping service consolidated its position.

Google Android followed a similar trajectory. The Commission investigated from 2015 and issued its decision in July 2018, imposing a fine of approximately €4.343 billion. The General Court reduced that fine to approximately €4.125 billion in September 2022 while upholding the core findings, and the Court of Justice dismissed Google’s final appeal only on 2 July 2026, confirming the reduced figure. Roughly eleven years separated the opening of the investigation from the final judgment, and eight years separated the Commission’s decision from the last word in Luxembourg, while the Android ecosystem continued to consolidate.

This timing problem is structural. Article 102 is fundamentally ex post enforcement, requiring proof of dominance, of abuse, and of foreclosure effects, each contestable through successive layers of appeal. In fast-moving digital markets, where competitive outcomes can shift within months, an enforcement timeline measured in the better part of a decade, eight to fourteen years in the Google cases, represents a structural failure of timing. Behavioural remedies also struggle in these markets: once users have habituated to one ecosystem and network effects have locked in complementary services, a late remedy may identify the abuse correctly while failing to restore genuine competition.

5. THE DIGITAL MARKETS ACT AS COMPLEMENT, NOT REPLACEMENT

The Digital Markets Act addresses this institutional gap through a fundamentally different design. The Regulation entered into force in 2022 and became applicable from 2 May 2023; its principal compliance obligations became applicable to the first designated gatekeepers on 7 March 2024. Rather than investigating abuse case by case, the DMA designates firms as ‘gatekeepers’ and imposes ex ante obligations.

A firm is designated as a gatekeeper if it satisfies the size thresholds (an annual EEA turnover of at least €7.5 billion in each of the last three financial years, or an average market capitalisation of at least €75 billion in the last financial year), provides a core platform service in at least three Member States, and constitutes an important gateway between business users and end users. The six original gatekeepers, Alphabet, Amazon, Apple, ByteDance, Meta, and Microsoft, were designated on 6 September 2023. Booking was subsequently designated on 13 May 2024 for its online intermediation service Booking.com, bringing the current total to seven.

Once designated, gatekeepers must comply with a closed list of obligations that ensure interoperability, limit self-preferencing and tying, permit data portability, prohibit anti-steering, and require fair access terms. The DMA’s decisive institutional advantage lies in speed, though that advantage should be stated accurately rather than exaggerated. The Commission opened its first non-compliance investigations into Apple and Meta on 25 March 2024 and adopted its first non-compliance decisions on 23 April 2025, approximately thirteen months later. Apple was fined €500 million for breaching the anti-steering obligation under Article 5(4), and Meta was fined €200 million for its ‘pay or consent’ model under Article 5(2). Thirteen months is not instantaneous, but it is a small fraction of the eight-to-fourteen-year timelines seen in the Article 102 cases above.

Yet the Commission has been explicit that the DMA ‘complements, but does not change’ Article 102 TFEU. Article 1(6) of Regulation (EU) 2022/1925 provides that it applies ‘without prejudice to Articles 101 and 102 TFEU’. This reflects a deliberate legislative choice to maintain dual systems. The DMA is pointed: it applies only to designated gatekeepers and closed-list core platform services. Article 102 remains the broader prohibition, capable of reaching dominant firms outside the DMA architecture, conduct outside the closed list, and novel abuses emerging before regulatory anticipation.

6. GENERATIVE AI AND THE FUTURE FRONTIER

Generative AI’s significance lies not in rendering Article 102 obsolete, but in revealing why its flexibility remains essential. GenAI integration does not create wholly new abuse categories; it re-situates familiar ones, tying, self-preferencing, access refusals, discriminatory conditions, in new technical settings.

Consider AI assistant integration. Google embeds Gemini across Search, YouTube, and Gmail. Apple has integrated Apple Intelligence into compatible models of its iPhone, Mac and iPad product lines. Meta has moved to embed Meta AI within WhatsApp. These are not marginal features. AI assistants may become the primary interface through which users search, ask, compare, and transact. If dominance over the gateway converts into dominance over the answer layer, leveraging concerns become more acute.

The treatment of Meta AI in WhatsApp is a case in point. National competition authorities have begun to scrutinise whether the pre-installation and default prominence of an AI assistant within a dominant messaging service constitutes abuse, applying self-preferencing and tying logic to a setting in which the intensity of foreclosure is new. Whether such conduct is ultimately condemned will turn on the usual analysis of capability to foreclose and objective justification.

Upstream, competitive risks are equally acute. GenAI markets are defined by control of training data, model access, inference capacity, and compute infrastructure. A dominant cloud provider can degrade model access, reserve premium functionality for its own downstream products, restrict interoperability, or bundle model access with cloud commitments that raise switching costs. These risks map directly onto Article 102’s language of unfair trading conditions, tying, and access refusals.

GenAI thus shows why Article 102’s flexibility is more necessary than ever. In AI markets, competitive conflict involves combinations of conduct that no fixed list can anticipate. The same firm may combine pre-installation, interface prominence, proprietary data advantages, model dependency, cloud bundling, and selective interoperability in a single integrated strategy. A rigid taxonomy would miss the interaction between these layers; Article 102 remains valuable precisely because it can connect them.

7. CONCLUSION: A QUALIFIED EFFECTIVENESS

Article 102 TFEU is effective to a meaningful but incomplete extent. It is effective in principle because its open-textured prohibitions remain capable of reaching new exclusionary conduct. It is less effective in practice because enforcement is slow, technically demanding, and often retrospective.

Recent case law is encouraging. Google Shopping confirmed that Article 102 can condemn self-favouring without pretending that digital foreclosure resembles classical price predation. Android Auto showed that Article 102 can extend access obligations to open platforms without mechanically reproducing the Bronner standard, while still preserving objective justifications for refusal. The Apple music-streaming decision and the Meta Marketplace fine show that the framework remains capable of addressing novel harms, tying, unfair trading conditions, and direct consumer impact.

Institutionally, however, the picture is weaker. The principal difficulty lies not in doctrinal incapacity but in timing, proof, and remedy. Proceedings spanning eight to fourteen years are inconsistent with digital-market dynamics, and behavioural remedies struggle once dominant positions have hardened.

Going forward, Article 102’s effectiveness depends on four developments. First, faster interim intervention is essential in fast-tipping digital and AI markets. Second, remedies must become technically informed. Third, DMA coordination must function as a genuinely complementary tool. Fourth, sustained scrutiny of interoperability, AI infrastructure, and data access must continue.

The real question is no longer whether Article 102 can survive the digital economy. It can. The real question is what instrument it now is within a dual system of ex ante and ex post control. The answer is that Article 102 remains the general abuse-of-dominance standard: capable of reaching conduct beyond the DMA’s closed list, capable of addressing novel conduct combinations in AI markets, and capable of protecting competition in settings regulators cannot fully anticipate. It is slower and more demanding than ex ante regulation, but also more flexible. In a world of rapid technological change, that flexibility is essential.

REFERENCES

Primary Sources

Legislation

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 (Digital Markets Act) [2022] OJ L265/1

Case Law

Case C-48/22 P, Google and Alphabet v Commission (Google Shopping) [2024] EU:C:2024:726 (self-preferencing; leveraging; inapplicability of the Bronner conditions)

Case T-604/18, Google and Alphabet v Commission (Google Android) [2022] EU:T:2022:541 (tying; app-store pre-installation; fine reduced to c. €4.125 billion)

Case C-738/22 P, Google and Alphabet v Commission (Google Android), judgment of the Court of Justice of 2 July 2026 (appeal dismissed; €4.125 billion fine confirmed)

Case C-233/23, Alphabet and Others (Android Auto) [2025] EU:C:2025:110 (refusal of interoperability; open platforms; objective justification)

Commission Decision AT.40684 of 14 November 2024, Meta — Facebook Marketplace (tying; unfair trading conditions; fine of €797.72 million)

Commission Decision AT.40437 of 4 March 2024, Apple — App Store Practices (Music Streaming) (anti-steering; fine of c. €1.8 billion)

Commission Decision of 18 July 2018 (C(2018) 4761 final), Google Android (original fine of c. €4.343 billion)

Case 85/76, Hoffmann-La Roche v Commission [1979] EU:C:1979:36 (special responsibility of dominant firms)

Case C-7/97, Bronner [1998] EU:C:1998:569 (essential-facilities test)

Secondary Sources

Akman, P, ‘The Reform of Article 102 TFEU: Mission Accomplished?’ (2014) 81 Antitrust Law Journal 145

Crémer, J, de Montjoye, Y-A and Schweitzer, H, Competition Policy for the Digital Era (Publications Office of the European Union 2019)

Davies, T, ‘Innovation or Infringement? Generative AI and the Potential for Exclusionary Abuse under Article 102 TFEU’ (Centre for Law, Economics and Society Policy Paper Series 1/2025)

Knapstad, T, ‘Digital Dominance: Assessing Market Definition and Market Power for Online Platforms under Article 102 TFEU’ (2024) 20(2) European Competition Journal 412

Mândrescu, D, ‘Designing (Restorative) Remedies for Abuses of Dominance by Online Platforms’ (2025) 13 Journal of Antitrust Enforcement 353

Monti, G, ‘Article 102 TFEU: The Case for a Re-Evaluation’ (2017) 54 Common Market Law Review 145

Peters, V, ‘Double Jeopardy of Article 102 TFEU and the DMA — the Challenges of a Multi-Level Enforcement System’ (2025) 21 Utrecht Law Review 95

Whish, R and Bailey, D, Competition Law (10th edn, Oxford University Press 2021)

Zheng, K, ‘Antitrust in Artificial Intelligence Infrastructure — Between Regulation and Innovation in the EU, the US, and China’ (2025) 59 Computer Law & Security Review 106211