AI Agent Marketplace Ecosystem Dominance Concerns in Europe

1. Introduction

An AI agent marketplace is a digital ecosystem in which users can discover, select, install, purchase, subscribe to, or deploy AI agents for tasks such as:

research;

shopping;

travel booking;

coding;

financial analysis;

customer service;

procurement;

logistics;

scheduling;

autonomous transactions.

An AI agent marketplace ecosystem may include:

Foundation model → cloud infrastructure → operating system → agent store → identity/payment system → data → APIs → end users → merchants/service providers

The competition-law concern arises when one undertaking controls several layers of this ecosystem and uses that position to advantage its own agents or disadvantage rival agents.

There is no mature EU case specifically deciding “AI agent marketplace ecosystem dominance.” The legal analysis therefore relies heavily on digital-platform cases such as Google Shopping, Android, Slovak Telekom, Servizio Elettrico Nazionale and Meta, together with the DMA and emerging AI regulation.

This is particularly current because the European Commission's 2026 DMA work has identified AI, cloud services, interoperability and access to search data as important digital-market issues. In July 2026, the Commission issued binding DMA specification measures concerning interoperability for competing AI services on Android and access by third-party search engines to Google Search data. (Digital Markets Act (DMA))

2. Meaning of AI Agent Marketplace Ecosystem

Consider an ecosystem operated by Platform X.

It provides:

a foundation AI model;

cloud computing;

an operating system;

an AI assistant;

an agent marketplace;

identity services;

payments;

application programming interfaces;

consumer data;

ranking and recommendation systems.

Independent developers can upload their own agents.

The platform then controls the gateway through which users discover and employ those agents.

This creates a potential strategic problem:

The marketplace owner may simultaneously be the marketplace operator and a competitor of the agents sold through that marketplace.

3. Central competition concern

The basic concern can be expressed as:

Platform control

↓

Gatekeeper position

↓

Access to users + data + APIs + computing + payment + operating system

↓

Preferential treatment of own AI agents

↓

Reduced visibility/access for rival agents

↓

Higher switching costs

↓

Ecosystem dependence

↓

Potential foreclosure

This can potentially engage Article 102 TFEU, and for designated gatekeepers, the Digital Markets Act (DMA).

The Commission's current Article 102 framework treats exclusionary conduct by dominant undertakings as the central concern, while emphasising that dominance itself is not unlawful. (Competition Policy)

4. Main forms of ecosystem dominance

A. Self-preferencing

The platform ranks its own AI agent above rival agents.

Example:

Platform's shopping agent appears first even when a rival agent performs the task equally well.

This is closely analogous to the Google Shopping litigation.

B. Preferential API access

The platform gives its own AI agent:

faster APIs;

more tokens;

greater context windows;

privileged system functions;

deeper operating-system integration.

Competitors receive restricted access.

C. Data advantage

The platform's own agent receives access to:

search data;

transaction data;

user histories;

behavioural information;

device data.

Third-party agents cannot access equivalent information.

This may produce a data feedback loop:

More users → more data → better agent → more users.

5. Case Law 1 — Google Shopping

Google and Alphabet v Commission

C-48/22 P

This is probably the strongest existing analogy.

The Court of Justice upheld the Commission's finding that Google had abused its dominant position by favouring its own comparison-shopping service in general search results. The judgment concerned leveraging, potential foreclosure, causal effects and the use of Google's general-search position to favour its specialised service. (curia)

Application to AI marketplaces

Imagine:

Platform X owns the dominant general AI assistant.

It launches its own:

“Travel Agent X.”

Third-party travel agents are also available in the marketplace.

The platform's general assistant preferentially recommends its own travel agent.

This resembles the structural logic of Google Shopping:

dominant gateway → own downstream service → preferential treatment → potential foreclosure.

Principle

A dominant platform's control of a gateway can become problematic when that control is used to systematically favour its own downstream service.

6. Case Law 2 — Google Android

Google and Alphabet v Commission

C-738/22 P

The Court of Justice delivered judgment on 2 July 2026 in the Android case.

The Court upheld Google's liability concerning contractual restrictions involving Android, including pre-installation arrangements, tying and exclusionary effects, and confirmed the fine of approximately €4.1 billion after judicial review. (curia)

Relevance to AI-agent ecosystems

An AI-agent marketplace could similarly combine:

operating-system control;

default settings;

pre-installation;

contractual restrictions;

exclusive placement;

technical integration.

For example:

Android device → default AI assistant → default agent marketplace → preferred agents.

If competing AI agents cannot obtain equivalent access to the operating system, an Article 102 issue may arise.

Principle

Control of an important platform layer can be leveraged into adjacent markets through contractual or technical restrictions.

This is especially relevant to AI agents because agents increasingly depend upon access to operating-system functions.

7. Case Law 3 — Slovak Telekom

Slovak Telekom v Commission

C-165/19 P

The Court considered exclusionary conduct involving access to telecommunications infrastructure and margin squeeze.

The judgment clarified that, for practices other than a pure refusal of access, the absence of strict indispensability is not necessarily decisive when assessing potentially abusive conduct. (Infocuria)

Application to AI-agent marketplaces

Suppose an AI platform controls an important agent interface.

It permits rival agents to use the interface but imposes:

slower access;

higher fees;

technical limitations;

inferior functionality.

This is not necessarily a complete refusal of access.

It may instead be discriminatory or disadvantageous access.

Principle

Partial or inferior access can itself be competitively significant.

8. Case Law 4 — Servizio Elettrico Nazionale

Servizio Elettrico Nazionale and Others

C-377/20

The Court examined exclusionary conduct by an incumbent undertaking and the use of commercially sensitive information within a corporate group to preserve a dominant position.

The Court focused on whether the conduct was capable of producing exclusionary effects and whether the undertaking used means other than those associated with competition on the merits. (Infocuria)

AI ecosystem application

Imagine a platform owns:

the agent marketplace;

a dominant search engine;

a consumer identity system.

It obtains information from rival agents and uses that information to improve its own competing agent.

For example:

Rival shopping agent → transaction data → platform → platform's own shopping agent.

That can create a potentially important information advantage.

Principle

A dominant undertaking's exploitation of information obtained through its existing position may become relevant where it helps preserve or extend dominance through exclusionary means.

9. Case Law 5 — Meta Platforms

Meta Platforms and Others

C-252/21

The Court examined the relationship between competition law and GDPR in the context of Meta's processing of personal data.

The Court accepted that a competition authority may need to consider whether data processing complies with GDPR when examining the conduct of a dominant undertaking, while requiring appropriate cooperation with data-protection authorities. (Infocuria)

AI marketplace relevance

AI agents depend heavily on data.

An agent marketplace may know:

what consumers ask;

which agents they select;

which products they purchase;

which agent responses they reject;

how long users interact with agents;

what tasks users repeatedly perform.

The marketplace operator may then use those insights to improve its own agent.

This can create:

data advantage → better agent → more users → more data.

Principle

Data practices can become relevant to the assessment of competitive power and abuse where data is an important competitive input.

10. Case Law 6 — Bronner

Oscar Bronner GmbH v Mediaprint

C-7/97

Bronner is important for refusal-of-access questions.

The Court established a strict framework for when a dominant undertaking may be required to provide access to infrastructure.

The traditional conditions include considerations such as:

indispensability;

elimination of effective competition;

absence of objective justification.

AI marketplace application

Suppose Platform X owns the only technically viable agent-discovery infrastructure.

A rival asks:

“Give our agent access to your marketplace.”

The answer is not automatically:

“Dominant platform = mandatory access.”

Bronner cautions against turning Article 102 into a general obligation to deal.

However, where the conduct is not a pure refusal of access but discriminatory treatment of already admitted competitors, other Article 102 principles can become more relevant.

This distinction is crucial for AI-agent marketplaces.

11. Case Law 7 — Deutsche Telekom

Deutsche Telekom v Commission

C-280/08 P

The case concerned margin squeeze and the relationship between upstream and downstream markets.

The Court recognised that a dominant vertically integrated undertaking can abuse its position when the pricing structure makes effective downstream competition difficult.

AI marketplace example

Suppose Platform X provides:

upstream AI infrastructure

and

downstream agent marketplace services.

It charges independent agent developers:

€10 for infrastructure access;

€9 marketplace commission;

while providing its own agent with infrastructure at an internal cost unavailable to rivals.

If competitors cannot profitably compete because of the platform's pricing structure, a margin-squeeze-type theory may become relevant.

The exact legal analysis would depend on the market definition, costs, pricing structure and applicable Article 102 principles.

12. Case Law 8 — Intel

Intel v Commission

C-413/14 P

Intel concerns exclusionary rebates.

The Court clarified the importance of examining the actual or potential exclusionary effects of conditional rebates where the undertaking puts forward evidence that its conduct is not capable of restricting competition.

AI marketplace application

Suppose an AI platform tells developers:

“You receive better marketplace placement if you use our cloud infrastructure exclusively.”

Or:

“Agents receive reduced marketplace commissions only if they use our AI model.”

This can create ecosystem lock-in.

A developer may remain on the platform not because it is technically superior but because leaving would cause:

higher costs;

loss of ranking;

loss of customers;

loss of data;

loss of interoperability.

Principle

Conditional commercial incentives can become relevant to Article 102 where they have exclusionary potential.

13. Case-law table

These are analogical authorities. None of these judgments has yet established a specific doctrine governing an AI-agent marketplace as such.

14. The ecosystem problem

AI-agent markets differ from traditional software markets because the ecosystem may contain several layers.

Layer 1 — Compute

Cloud infrastructure.

Layer 2 — Foundation model

Large language or multimodal model.

Layer 3 — Agent framework

Tools allowing agents to execute actions.

Layer 4 — Operating system

Device-level integration.

Layer 5 — Marketplace

Agent discovery and distribution.

Layer 6 — Identity

Authentication and user profiles.

Layer 7 — Payments

Transaction infrastructure.

Layer 8 — Data

Search, behavioural and transactional information.

Layer 9 — Users

The final demand side.

Control over several layers can create ecosystem leverage.

15. Network effects

AI-agent marketplaces can have strong network effects.

More users:

↓

more developers

↓

more agents

↓

more tasks covered

↓

greater user value

↓

more users.

At the same time:

More users

↓

more behavioural data

↓

better recommendation

↓

better agent performance

↓

more users.

This creates a two-sided or multi-sided feedback loop.

A dominant platform may therefore become difficult to challenge even if a rival develops technically strong AI.

16. Switching costs

Users may accumulate:

personal preferences;

memories;

agent configurations;

API connections;

payment credentials;

workflow integrations;

business data;

custom instructions.

Developers may accumulate:

marketplace ratings;

customers;

API integrations;

proprietary tools;

platform-specific code.

Therefore switching from Platform X to Platform Y can be expensive.

This creates ecosystem lock-in.

17. Interoperability

Interoperability may become one of the most important issues.

Imagine Platform X's agent can access:

email;

calendar;

contacts;

payments;

maps;

operating-system functions.

A rival agent can access only:

basic text generation.

Even if the rival agent is technically superior, users may choose Platform X because it has deeper system integration.

This is why the Commission's July 2026 DMA specification measures requiring equal access for competing AI assistants on Android are significant for the emerging AI-agent ecosystem. (Digital Markets Act (DMA))

18. DMA and AI-agent ecosystems

The DMA is especially important for designated gatekeepers.

AI-agent marketplaces can interact with:

operating systems;

online search;

app stores;

cloud computing;

advertising services;

online intermediation services.

The DMA may impose obligations concerning:

self-preferencing;

interoperability;

access;

data use;

switching;

business-user access;

combination of personal data;

anti-steering;

technical restrictions.

The precise DMA obligation depends on the designated core platform service and the conduct involved.

It is therefore incorrect to say:

“Every AI-agent marketplace is regulated by the DMA.”

The legal question is whether the undertaking and service fall within the DMA's relevant designation and obligations.

19. AI Act and AI-agent marketplaces

The AI Act is relevant but performs a different function.

The EU AI Act's official AI Act Service Desk states that AI agents are not a separate category of AI under the Act, but can fall within the general definition of an AI system or, where applicable, a GPAI model. (AI Act Service Desk)

From 2 August 2026, certain transparency obligations apply where AI agents interact with natural persons or generate content, subject to the applicable conditions. Certain high-risk obligations apply on later dates where an agent qualifies as a high-risk AI system. (AI Act Service Desk)

Competition significance

The AI Act is not itself a substitute for Article 102.

Instead:

AI Act → safety/transparency/fundamental rights

DMA → contestability and fairness of designated digital platforms

Article 102 → abuse of dominance

GDPR → personal-data processing

These regimes can operate simultaneously.

20. Self-preferencing in AI-agent stores

Suppose Platform X operates an agent store.

There are:

50,000 third-party agents;

5 Platform X agents.

Platform X's algorithm nevertheless gives its own agents:

first-page placement;

default installation;

higher recommendation frequency;

preferential API access.

The competition question is:

Is the ranking based on objective quality, or is the platform using its gatekeeper position to disadvantage rivals?

Google Shopping demonstrates why this distinction matters. (curia)

21. Tying and bundling

A platform could require:

“To access our agent marketplace, you must use our AI model.”

Or:

“To obtain premium marketplace placement, you must use our cloud.”

Or:

“To access operating-system functions, your agent must use our assistant.”

This may raise tying or bundling concerns.

The Google Android judgment is particularly relevant because the Court's 2026 decision addressed contractual restrictions and tying in a multi-layer digital ecosystem. (curia)

22. Data foreclosure

Imagine the platform gives its own agent access to:

100% of search queries.

Third-party agents receive:

only public search results.

The platform can then train and optimise its own agent using richer data.

This produces:

data foreclosure.

The concern becomes particularly serious when the data is:

difficult to replicate;

generated by the platform's gateway position;

essential to effective competition;

used to improve a downstream competing service.

The Commission's July 2026 DMA action concerning access to Google Search data illustrates the regulatory significance of this issue. (Digital Markets Act (DMA))

23. Agent ranking manipulation

Ranking is particularly powerful because consumers rarely inspect every available agent.

Suppose 100,000 agents are available.

If Platform X determines which 10 agents consumers see first, it effectively controls consumer attention.

Possible manipulation includes:

suppressing rival agents;

artificial popularity;

preferential badges;

default selection;

personalised rankings favouring the platform's own services.

The competitive effect can arise even without formally excluding a competitor from the marketplace.

24. Commission discrimination

Suppose:

Platform-owned agent

Commission = 2%

Independent agent

Commission = 30%

This could create a major competitive disadvantage.

However, different prices are not automatically unlawful.

The analysis would need to consider:

dominance;

objective justification;

cost differences;

discriminatory effects;

exclusionary capability;

relevant market;

actual competitive effects.

25. Agent marketplace and refusal to deal

A platform could simply refuse to list a rival.

But Bronner means that mandatory access is not automatically required.

The legal analysis becomes more complex if:

the platform has already admitted rival agents;

it selectively removes competitors;

it provides materially inferior access;

it controls a critical gateway;

the restriction is designed to favour its own agent.

This is why refusal to deal and discriminatory access should not be treated as identical legal theories.

26. Exclusive dealing

Platform X could offer:

“Agents using our model exclusively receive 90% lower marketplace fees.”

A competing agent developer may technically be free to use another model.

But the commercial incentive may make switching uneconomic.

Potential consequences:

foreclosure of rival models;

reduced multi-homing;

increased entry barriers;

greater ecosystem dependency.

The Intel jurisprudence is useful when analysing conditional commercial incentives and potential foreclosure.

27. Consumer lock-in

Consumers can become locked into an AI ecosystem through:

stored memories;

personalised profiles;

payment systems;

proprietary plugins;

agent workflows;

API permissions;

device integration.

Switching may require rebuilding the entire digital environment.

This can create high switching costs, a factor the Commission considers when assessing dominance. (Competition Policy)

28. Developer lock-in

Developers can similarly become dependent.

For example:

Agent developer → Platform API → Platform marketplace → Platform users.

If the developer leaves:

API compatibility disappears;

marketplace ranking disappears;

customer reviews may disappear;

user relationships may become inaccessible.

The platform may therefore control both sides:

consumer lock-in + developer lock-in.

29. Interoperability as a competition remedy

Possible remedies could include:

API access

Competitors receive equivalent technical access.

Data portability

Users can transfer relevant information.

Agent portability

Users can migrate agent configurations.

Ranking transparency

Marketplace ranking criteria become more transparent.

Choice screens

Users can choose competing agents.

Default neutrality

The platform cannot automatically privilege its own agent.

Multi-homing

Developers can operate across several marketplaces.

30. Competition and privacy intersection

An AI-agent marketplace may know:

what a user asks an agent;

what products the user is considering;

what services the user buys;

what financial decisions the user is contemplating;

what competitors the user is considering.

This creates a powerful behavioural dataset.

Under Meta, C-252/21, competition authorities may need to consider the relationship between dominance and personal-data processing. (Infocuria)

Therefore:

data protection can become part of the competitive structure of an AI ecosystem.

31. Agent-to-agent competition

A new issue arises where AI agents themselves negotiate with other agents.

Example:

Shopping Agent A

negotiates with

Retail Agent B

which negotiates with

Logistics Agent C.

If all are controlled by the same ecosystem, the platform may effectively control several stages of the transaction.

Potential concerns include:

vertical foreclosure;

discriminatory routing;

exclusive dealing;

tying;

preferential transaction fees;

information asymmetry.

32. Autonomous algorithmic coordination

Suppose independent marketplaces use AI agents that continuously monitor:

prices;

commissions;

demand;

inventory.

Their agents independently adjust conduct.

This creates a difficult Article 101 question:

When does autonomous algorithmic adaptation become legally relevant coordination?

The existence of similar algorithms alone does not establish an infringement.

The legal analysis would need to examine:

communication;

information exchange;

concerted practice;

algorithmic instructions;

transparency;

human involvement;

predictability of coordination.

33. Dominance test

The first Article 102 question remains:

Is the undertaking dominant?

The Commission assesses factors including:

market shares;

barriers to entry;

countervailing buyer power;

resources;

vertical integration.

The Commission's current Article 102 materials expressly identify vertical integration and barriers to entry as relevant factors. (Competition Policy)

For AI ecosystems, additional practical indicators may include:

installed user base;

developer base;

data advantages;

switching costs;

cloud capacity;

model access;

operating-system integration;

marketplace network effects.

34. Relevant-market questions

An AI-agent ecosystem may contain several possible markets.

Market 1

Foundation AI models.

Market 2

AI-agent platforms.

Market 3

Agent marketplaces.

Market 4

AI assistant services.

Market 5

Cloud AI infrastructure.

Market 6

Specialised agent services.

Market 7

AI-enabled transactions.

The relevant market cannot simply be assumed.

Different products may be:

substitutes;

complements;

vertically related;

part of the same ecosystem but separate markets.

35. Essential facility argument

A competitor may argue:

“The platform's agent marketplace is essential to compete.”

But Bronner establishes a demanding framework for compulsory access.

Therefore, an “essential facility” argument requires careful proof concerning:

indispensability;

elimination of effective competition;

inability to replicate;

objective justification.

A platform being very important is not automatically the same as being legally indispensable.

36. Economic effects

Potential effects of ecosystem dominance include:

Higher entry barriers

New agent developers cannot obtain sufficient users.

Reduced innovation

Rivals cannot scale.

Higher commissions

Developers become dependent.

Lower consumer choice

Only platform-preferred agents receive visibility.

Data concentration

One firm accumulates disproportionate behavioural information.

Reduced interoperability

Users cannot easily switch.

Higher prices

Developers pass platform fees to consumers.

Lower quality

Competitive pressure decreases.

37. Possible objective justifications

A platform may argue that preferential treatment is based upon:

security;

privacy;

reliability;

latency;

safety;

fraud prevention;

technical compatibility;

quality assurance.

These can be legitimate concerns.

The question is whether the measure is:

genuinely necessary;

objectively justified;

proportionate;

consistently applied.

A platform cannot simply label a discriminatory practice “security” without supporting evidence.

38. Hypothetical

Assume AI Hub Europe operates:

the largest AI operating system;

an AI-agent marketplace;

a cloud platform;

a foundation model.

It launches AI Hub Shopping Agent.

Third-party agents previously controlled 70% of marketplace transactions.

AI Hub changes the ranking algorithm.

Its own agent now appears first for most users.

It also:

gives its own agent privileged access to search data;

gives its own agent unlimited API calls;

charges third-party agents higher commissions;

requires competing agents to use its cloud;

prevents agents from redirecting users to competing marketplaces.

Potential legal issues

Article 102

Possible leveraging/self-preferencing/exclusionary conduct.

DMA

Potential obligations depending on gatekeeper designation and relevant core platform service.

GDPR

Data-combination and profiling issues.

Consumer law

Transparency and choice.

AI Act

Applicable AI-system obligations depending upon classification and functionality.

39. Legal test

A practical European legal test is:

Step 1 — Identify the ecosystem

Map:

model → cloud → OS → marketplace → data → users.

Step 2 — Define the relevant markets

Do not assume the entire ecosystem is one market.

Step 3 — Establish dominance

Assess market power and ecosystem barriers.

Step 4 — Identify the conduct

Is there:

self-preferencing?

tying?

bundling?

discriminatory access?

refusal to deal?

exclusive dealing?

data exploitation?

discriminatory commissions?

Step 5 — Identify competitive effects

Does the conduct have the capability to:

foreclose rivals;

raise entry barriers;

reduce multi-homing;

reduce innovation;

restrict consumer choice?

Step 6 — Examine counterfactual

What would competition look like without the challenged conduct?

Google Shopping and the 2026 Android judgment illustrate the importance of contextual and counterfactual analysis in Article 102 cases. (Infocuria)

Step 7 — Examine objective justification

Is there a genuine technical, security or efficiency reason?

Step 8 — Consider DMA

Is the undertaking a gatekeeper and is the relevant service covered?

Step 9 — Consider GDPR/AI Act

Does the conduct involve personal data or regulated AI functionality?

Step 10 — Consider remedies

Potential remedies include:

interoperability;

non-discriminatory access;

data access;

choice mechanisms;

prohibition of self-preferencing;

contractual changes;

behavioural remedies.

40. Direct and analogical authority

These cases provide legal analogies, not a completed AI-agent-marketplace doctrine.

41. Important 2026 development

The AI-agent ecosystem is moving rapidly toward an infrastructure model in which AI assistants need access to operating-system functions and large-scale data.

The Commission's 16 July 2026 DMA measures concerning Google specifically addressed interoperability for competing AI assistants on Android and access by third-party search engines to Google Search data. (Digital Markets Act (DMA))

Separately, on 25 June 2026, the Commission announced preliminary views that Amazon Web Services and Microsoft Azure should be designated as DMA gatekeepers for cloud services, citing entrenched user bases, lock-in, switching costs, ecosystems and the increasing importance of AI tools in cloud procurement. (Digital Markets Act (DMA))

These developments show why AI-agent competition cannot be examined solely at the agent-store level. Control of cloud, operating systems, search, data and distribution may determine competitive conditions in downstream agent markets.

42. Conclusion

AI Agent Marketplace Ecosystem Dominance is fundamentally an issue of control over gateways.

The most important potential structure is:

AI model + cloud + operating system + marketplace + data + payments + users

When one undertaking controls several of these layers, it may possess significant ecosystem advantages.

The principal competition-law risks are:

self-preferencing of the platform's own agents;

discriminatory API access;

preferential operating-system integration;

exclusive dealing;

tying and bundling;

data foreclosure;

discriminatory commissions;

ranking manipulation;

consumer and developer lock-in;

restriction of interoperability.

The leading authorities—particularly Google Shopping, Google Android, Slovak Telekom, Servizio Elettrico Nazionale, Meta, Bronner, Deutsche Telekom and Intel—provide the existing doctrinal building blocks.

The central principle can therefore be stated simply:

An AI-agent marketplace may compete on the merits, but a dominant ecosystem owner cannot necessarily use control over one layer of the ecosystem to unfairly foreclose competition at another layer.

At the same time, mere ecosystem size, vertical integration, superior AI quality or successful innovation is not itself an Article 102 infringement. The decisive legal analysis remains focused on dominance, the specific conduct, its capability/effects, causation where required, and any objective justification. The Commission's 2026 Article 102 Guidelines reinforce this structured approach. (Competition Policy)

Exam Keywords

AI agent marketplace — ecosystem dominance — Article 102 TFEU — DMA — self-preferencing — leveraging — tying — bundling — interoperability — API access — data foreclosure — algorithmic ranking — agent store — cloud dependency — operating-system control — switching costs — multi-homing — network effects — ecosystem lock-in — essential facilities — Bronner — Google Shopping — Google Android — Slovak Telekom — Meta — Servizio Elettrico Nazionale — Deutsche Telekom — Intel.