Distributed AI Market Actors and Decentralized Dominance Structures

Introduction

Distributed AI market actors are firms, developers, infrastructure providers, data suppliers, model creators, application developers, cloud providers, platform operators, and autonomous AI agents whose economic activities are spread across multiple entities rather than controlled by one visibly dominant undertaking.

A decentralized dominance structure arises where market power is not concentrated in a single corporation but is nevertheless exercised through a network of interdependent actors, technical standards, APIs, data pools, cloud infrastructure, model ecosystems, developer communities, or algorithmic coordination mechanisms.

This creates a major competition-law challenge: formal decentralization does not necessarily mean effective competition. A market may contain hundreds of nominally independent AI actors while a small number of firms control the critical inputs, interfaces, infrastructure, data, standards, distribution channels, or computational resources on which those actors depend.

The central competition-law question is therefore:

Who possesses economically decisive control, even where legal ownership and decision-making appear distributed?

1. Meaning of Distributed AI Market Actors

Distributed AI markets can contain several layers of actors.

A. Foundation-model developers

These actors develop large AI models and may control:

• model weights;
• training methodologies;
• proprietary datasets;
• safety systems;
• model APIs;
• licensing terms;
• fine-tuning access.

Their power may extend downstream even where applications are developed by independent firms.

B. Cloud and compute providers

AI development frequently depends upon:

• GPUs;
• TPUs;
• cloud computing;
• specialized AI accelerators;
• data-centre capacity;
• networking infrastructure;
• inference infrastructure.

Control over these resources can create upstream bottlenecks.

C. Data suppliers

Data may be supplied by:

• publishers;
• social-media platforms;
• mapping companies;
• financial institutions;
• governments;
• scientific repositories;
• consumer platforms.

Exclusive or preferential access to valuable datasets can produce competitive advantages.

D. AI application developers

These firms build:

• legal AI;
• medical AI;
• financial AI;
• autonomous-driving systems;
• recommendation engines;
• enterprise copilots;
• procurement agents.

They may technically be independent while economically dependent on upstream models or cloud infrastructure.

E. AI agents

Increasingly, AI systems themselves may perform economically significant functions such as:

• price setting;
• purchasing;
• bidding;
• inventory management;
• advertising;
• investment decisions;
• supplier selection.

This creates a distinction between human market actors and machine-mediated market actors.

2. What Is a Decentralized Dominance Structure?

Traditional dominance analysis often looks for a single undertaking with substantial market power.

Distributed AI markets complicate this model.

A dominance structure may instead look like:

Cloud provider → compute access → foundation model → API → application ecosystem → distribution platform → consumers

No single layer necessarily appears to control the entire market.

Yet each layer may become an essential dependency for the next.

For example:

100 AI application companies + 10 model developers + 5 cloud providers

may look competitive.

But if:

• two firms control most advanced compute;
• three firms control leading foundation models;
• one platform controls enterprise distribution;
• one ecosystem controls critical data;

then competitive independence may be substantially weaker than the number of firms suggests.

3. Decentralization Does Not Eliminate Market Power

Competition law traditionally distinguishes between:

structural decentralization and economic decentralization.

A market is structurally decentralized where many legal entities exist.

It is economically decentralized only where those entities possess meaningful independent competitive capacity.

This distinction is particularly important in AI.

A thousand developers may depend on:

• one cloud provider;
• one model API;
• one app store;
• one data source;
• one identity system;
• one enterprise distribution channel.

Consequently:

Many firms do not necessarily equal many competitive centres.

4. Sources of Decentralized AI Dominance

4.1 Compute concentration

Advanced AI requires substantial computing resources.

If access to high-performance computing becomes concentrated, dominant firms may influence:

• model development;
• entry;
• innovation;
• pricing;
• model quality;
• research capacity.

The competition concern is not merely the price of computing.

It is whether compute scarcity creates strategic dependence.

4.2 Data concentration

AI markets may exhibit strong data advantages.

A dominant platform can potentially combine:

• transaction data;
• behavioural data;
• search data;
• location data;
• purchasing information;
• interaction data.

The resulting advantage can create barriers to entry even where data is technically available elsewhere.

4.3 Model/API dependence

Downstream firms may build applications around a particular AI model.

Switching may require:

• rewriting applications;
• retraining;
• changing prompts;
• modifying safety systems;
• validating outputs;
• migrating data;
• recalibrating performance.

This can generate AI-specific switching costs.

4.4 Ecosystem control

A powerful firm may not dominate one narrowly defined market but may control an ecosystem consisting of:

• operating systems;
• cloud services;
• search;
• browsers;
• app stores;
• AI models;
• advertising;
• enterprise software.

Competition problems may therefore arise through ecosystem leverage rather than conventional single-market dominance.

5. Distributed Control Through Standards

Technical standards can become competition-law instruments.

An AI ecosystem may develop standards concerning:

• model interfaces;
• agent protocols;
• data formats;
• identity;
• safety certification;
• model evaluation;
• interoperability.

A technically open standard can nevertheless produce centralized power if one undertaking controls:

• certification;
• implementation;
• updates;
• compatibility;
• access to critical infrastructure.

The relevant question becomes:

Who controls the evolution of the standard?

6. Decentralized AI and Collective Dominance

One of the most difficult issues is whether competition law can address dominance exercised by multiple economically connected actors.

Traditional collective dominance doctrine generally requires more than parallel behaviour.

Relevant considerations may include:

1. market transparency;
2. economic links;
3. structural interdependence;
4. ability to monitor rivals;
5. incentives to maintain coordinated conduct;
6. absence of effective competitive pressure.

AI systems can intensify these characteristics because algorithms can monitor markets continuously.

7. Algorithmic Coordination Without Explicit Agreement

Distributed AI actors might use similar:

• pricing models;
• optimization systems;
• forecasting tools;
• third-party AI providers.

The resulting prices may converge without direct communication between competitors.

This creates the difficult distinction between:

Legitimate algorithmic adaptation

Each firm independently responds to market conditions.

Coordinated algorithmic conduct

AI systems facilitate or sustain behaviour that reduces competitive uncertainty.

Tacit algorithmic coordination

Algorithms independently learn that aggressive competition is disadvantageous and converge on stable outcomes.

The legal challenge is determining when correlation becomes legally relevant coordination.

8. AI Agents as Distributed Market Actors

Autonomous agents create an additional problem.

Suppose thousands of businesses use AI agents that:

• monitor competitors;
• negotiate contracts;
• purchase inventory;
• set prices;
• allocate advertising;
• respond to demand.

The agents may independently optimize for their owners.

But their interaction can generate a market outcome resembling coordination.

Competition authorities therefore increasingly need to examine:

human instructions + algorithmic architecture + data + incentives + resulting market behaviour.

9. Six Important Case Laws

9.1 United Brands v Commission

United Brands Company v Commission, Case 27/76 (1978)

The Court of Justice developed important principles concerning dominance and the ability of an undertaking to behave to an appreciable extent independently of competitors, customers, and consumers.

Relevance to distributed AI

The case provides a useful conceptual test.

An AI undertaking need not have absolute control over the market.

The question is whether it possesses sufficient economic strength to behave independently.

In an AI ecosystem, that power could arise from:

• proprietary models;
• exclusive data;
• compute access;
• distribution;
• ecosystem integration.

Thus, formal dependence on other actors does not necessarily prevent a finding of dominance.

9.2 Hoffmann-La Roche v Commission

Hoffmann-La Roche & Co AG v Commission, Case 85/76 (1979)

The Court emphasized that dominance concerns a position of economic strength enabling an undertaking to hinder effective competition and behave independently to a significant degree.

Relevance to AI

This is particularly important where AI market power is based on network effects and strategic dependencies.

A foundation-model provider might not monopolize every downstream AI application.

Nevertheless, if downstream firms cannot realistically discipline its behaviour, dominance may exist at the relevant upstream level.

9.3 Microsoft v Commission

Microsoft Corp. v Commission, Case T-201/04 (2007)

The General Court upheld major findings concerning Microsoft's conduct involving interoperability information and tying.

Relevance to distributed AI

The case is highly relevant to AI ecosystems because interoperability can determine whether downstream firms can compete effectively.

Potential modern analogues include:

• restricting interoperability between AI models and applications;
• withholding technical interfaces;
• tying AI functionality to operating systems;
• limiting compatibility with competing AI assistants.

The broader principle is that control over an important technological interface can become a source of exclusionary power.

9.4 Google Android

Google LLC and Alphabet Inc. v Commission, Case T-604/18 (2022)

The General Court examined Google's practices concerning Android, including tying and contractual restrictions involving search and app distribution.

Relevance to AI

The case demonstrates how dominance can be exercised through an ecosystem rather than a single product.

An AI ecosystem may similarly involve:

operating system + browser + search + assistant + app store + cloud + AI model.

Competition concerns may arise where control over one layer is used to advantage another.

This makes ecosystem leverage a central concept for AI competition law.

9.5 Google Shopping

Google Search (Shopping), Case T-612/17 (2021)

The General Court upheld the Commission's finding concerning Google's preferential treatment of its own comparison-shopping service in general search results.

Relevance to AI

The case is significant for self-preferencing.

An AI platform may control:

• model access;
• search results;
• recommendation systems;
• AI-generated answers;
• marketplace rankings.

If the platform gives its own AI services preferential visibility over competing services, competition authorities may examine whether control over the distribution layer is being used to disadvantage rivals.

9.6 Intel v Commission

Intel Corp. v Commission, Case C-413/14 P (2017)

The Court of Justice clarified the approach to exclusivity rebates and the importance of assessing whether conduct is capable of producing anticompetitive foreclosure.

Relevance to AI

AI firms could potentially use contractual arrangements involving:

• exclusive cloud commitments;
• preferential model access;
• rebates;
• capacity reservations;
• exclusive distribution;
• minimum-purchase obligations.

The lesson is that competition analysis should examine the actual foreclosure effects rather than simply the existence of contractual exclusivity.

10. Additional Relevant Case Laws

10.1 AKZO v Commission

AKZO Chemie BV v Commission, Case C-62/86 (1991)

AKZO established important principles concerning predatory pricing and abuse of dominance.

AI relevance

AI providers may operate temporarily at low prices because they possess:

• venture financing;
• cross-subsidies;
• cloud revenues;
• advertising revenues;
• ecosystem benefits.

Low or even zero prices therefore require careful analysis rather than automatically being treated as beneficial competition.

10.2 Bronner

Oscar Bronner GmbH & Co KG v Mediaprint, Case C-7/97 (1998)

The Court established a demanding framework for refusal-to-deal claims involving essential facilities.

AI relevance

The doctrine may become relevant where a firm controls:

• indispensable compute;
• unique datasets;
• critical model infrastructure;
• indispensable interfaces.

However, merely being technologically important does not automatically make an AI resource an essential facility. The stringent legal requirements remain significant.

10.3 Servizio Elettrico Nazionale

Servizio Elettrico Nazionale SpA v Autorità Garante della Concorrenza e del Mercato, Case C-377/20 (2022)

The Court addressed the assessment of exclusionary conduct by a dominant undertaking and emphasized the need to distinguish competition on the merits from conduct capable of excluding rivals.

AI relevance

This provides a useful framework for evaluating whether an AI firm's use of proprietary advantages represents:

competition on the merits

or

leveraging of entrenched market power.

11. Case-Law Synthesis

12. Decentralized Dominance Through Vertical Integration

An AI company may operate simultaneously as:

• cloud provider;
• chip purchaser;
• model developer;
• API provider;
• application developer;
• enterprise software supplier;
• marketplace operator.

This creates vertical control across the AI stack.

The competitive concern is not necessarily that vertical integration is unlawful.

Rather, the question is whether upstream control is used to disadvantage downstream competitors.

Examples include:

• preferential compute allocation;
• discriminatory API access;
• higher prices for rival applications;
• technical interoperability restrictions;
• self-preferencing;
• tying;
• exclusionary contracts.

13. Decentralized Dominance Through Network Effects

AI markets may exhibit several network effects simultaneously.

Direct network effects

More users may increase the value of an AI platform.

Data network effects

More users → more data → better model → more users.

Developer network effects

More developers → more applications → more users → more developers.

Compute network effects

More demand → larger infrastructure → greater efficiency → stronger competitive position.

Distribution network effects

More enterprise customers → stronger integration → higher switching costs → greater customer retention.

These effects can create self-reinforcing dominance.

14. The "Hidden Centre" Problem

The most important conceptual issue is the existence of a hidden centre of control.

An AI market can appear decentralized because:

• models are open source;
• developers are independent;
• data is distributed;
• agents are autonomous;
• protocols are interoperable.

But the real competitive bottleneck may be:

compute + distribution + data + standards + capital.

Consequently, competition authorities should not assess decentralization solely by counting market participants.

15. Competition-Law Tests for Distributed AI Markets

A useful analytical framework is:

Step 1 — Identify the AI stack

Map:

chips → compute → data → models → APIs → applications → distribution → users

Step 2 — Identify dependencies

Ask:

• Who depends upon whom?
• What cannot easily be substituted?
• What creates switching costs?

Step 3 — Identify control points

Determine who controls:

• access;
• interoperability;
• standards;
• data;
• compute;
• distribution.

Step 4 — Assess market power

Consider:

• market shares;
• barriers to entry;
• network effects;
• economies of scale;
• data advantages;
• switching costs;
• multi-homing.

Step 5 — Examine conduct

Look for:

• tying;
• bundling;
• self-preferencing;
• exclusionary agreements;
• discriminatory access;
• predatory pricing;
• refusal to deal;
• interoperability restrictions.

Step 6 — Examine actual effects

Determine whether conduct:

• forecloses rivals;
• increases entry barriers;
• reduces innovation;
• increases dependence;
• suppresses interoperability;
• consolidates ecosystem power.

16. Decentralized Dominance vs Collective Dominance

These concepts should not be treated as identical.

Decentralized dominance

Power is distributed across several layers or actors, but one or more firms may possess decisive control over critical dependencies.

Collective dominance

Two or more economically independent undertakings jointly occupy a dominant position because their market structure enables them to behave collectively in a manner resembling a single dominant entity.

Algorithmic coordination

Independent algorithms may converge on similar conduct without a conventional agreement.

The legal tests for these situations remain distinct.

17. Regulatory Difficulties

A. Attribution

If an autonomous AI agent makes an exclusionary decision, who is responsible?

• developer?
• deployer?
• platform?
• model provider?
• user?

B. Transparency

Complex AI systems may make it difficult to determine why a competitor was:

• ranked lower;
• denied access;
• charged more;
• excluded from an ecosystem.

C. Market definition

Traditional product markets may become unstable where one AI system performs multiple functions.

D. Dynamic competition

Today's small AI firm may become tomorrow's major competitor.

E. Multi-layer dependence

A conduct decision made at one level can affect several downstream markets.

18. Remedies

Competition authorities could potentially consider:

Structural remedies

• divestiture;
• separation of business units;
• limits on vertical integration.

Behavioural remedies

• non-discrimination;
• interoperability;
• access obligations;
• transparency;
• prohibition of self-preferencing.

Data remedies

• data portability;
• controlled data access;
• interoperability requirements.

Compute remedies

Where legally justified:

• access obligations;
• non-discriminatory allocation;
• transparency concerning capacity.

Ecosystem remedies

Authorities may also consider whether a dominant AI ecosystem should be prevented from using control over one market to foreclose competition in another.

19. Indian Competition-Law Perspective

Under the Competition Act, 2002, the issues can potentially engage:

• Section 4 — abuse of dominant position;
• Section 3 — anti-competitive agreements;
• Section 5 — combinations;
• Section 6 — regulation of combinations.

For AI, Section 4 concerns could arise around:

• denial of market access;
• discriminatory conditions;
• tying and bundling;
• leveraging;
• unfair conditions;
• exclusionary practices.

Section 3 may become relevant where apparently independent AI firms coordinate through:

• common algorithms;
• information-sharing systems;
• common intermediaries;
• restrictive technical arrangements.

The Competition Commission of India would therefore need to distinguish genuine technological interdependence from arrangements that materially restrict competition.

20. Overall Legal Principle

The central lesson from competition jurisprudence is that legal decentralization and economic decentralization are not synonymous.

An AI market can contain thousands of independent developers while remaining highly concentrated because a few undertakings control the resources that determine whether those developers can effectively compete.

The most important analytical shift is therefore:

From counting market actors to identifying control points.

In distributed AI markets, competition authorities should examine who controls compute, data, models, interfaces, standards, distribution and switching costs, and how those control points interact.

Conclusion

Distributed AI markets challenge the traditional assumption that dominance must be visibly concentrated in a single firm.

Future AI competition problems may instead arise from networked dominance, where market power emerges from relationships among cloud infrastructure, data, foundation models, APIs, applications, platforms and autonomous agents.

The relevant question is not simply:

“Is there one dominant AI company?”

It is:

“Where is effective competitive control located within the AI ecosystem, and can independent market actors realistically escape that control?”

The cases involving United Brands, Hoffmann-La Roche, Microsoft, Google Android, Google Shopping, Intel, AKZO, Bronner and Servizio Elettrico Nazionale provide the principal doctrinal tools for answering that question. Their combined significance is that competition law can examine economic power, foreclosure, interoperability, leveraging, contractual exclusion, ecosystem effects and competitive dependence, even where technological structures are complex and apparently decentralized.