Competition Law and Future Dominance Indicators in AI-Driven Economies

1. Introduction

Artificial intelligence is likely to change the traditional concept of market dominance. In conventional markets, dominance is often assessed through market shares, barriers to entry, control over distribution, pricing power, and the ability to act independently of competitors and consumers.

In AI-driven economies, however, competitive power may arise from control over compute, foundation models, proprietary data, AI talent, cloud infrastructure, APIs, distribution channels, ecosystems, feedback loops, and interoperability. A firm may therefore possess substantial competitive power even where its market share in a narrowly defined downstream product market appears modest.

The UK Competition and Markets Authority (CMA), for example, has identified access to compute, data, expertise and funding, together with feedback loops, economies of scale and powerful partnerships, as important competitive issues in foundation-model markets.

The future question will therefore increasingly be:

Which indicators demonstrate that an AI firm possesses durable and difficult-to-replicate competitive power?

2. Meaning of Dominance in AI Markets

Dominance does not simply mean that an undertaking is successful or technologically superior.

Traditionally, dominance means the possession of economic strength enabling an undertaking to behave to an appreciable extent independently of competitors, customers and ultimately consumers.

In AI markets, dominance may exist at several layers:

1. Compute layer – GPUs, accelerators, data centres and cloud infrastructure.
2. Data layer – proprietary datasets and continuous user-generated data.
3. Foundation-model layer – large language, multimodal or reasoning models.
4. Application layer – AI assistants, coding tools, healthcare AI, financial AI, etc.
5. Distribution layer – search engines, operating systems, app stores and browsers.
6. Cloud layer – AI training and inference infrastructure.
7. Agent layer – autonomous AI systems capable of performing transactions.
8. Ecosystem layer – integrated combinations of all of the above.

The CMA's work specifically recognises that AI competition can span compute, data, foundation-model development, partnerships, release, search, social media, mobile ecosystems and productivity software.

3. Why Traditional Market-Share Indicators May Become Insufficient

Market share remains relevant, but it can become misleading in AI markets.

For example, an AI company may have:

• relatively low consumer revenue;
• a rapidly growing user base;
• access to enormous computing resources;
• exclusive or preferential cloud arrangements;
• unique training data;
• control over an important API;
• integration into a dominant operating system; and
• substantial switching costs.

Such a company may possess significant future competitive power despite having a relatively small present-day market share.

Consequently, future dominance analysis should consider both:

Static indicators

• current market share;
• sales;
• customer numbers;
• revenue;
• installed base.

Dynamic indicators

• control over essential inputs;
• compute capacity;
• data accumulation;
• model performance;
• switching costs;
• network effects;
• ecosystem integration;
• access to distribution;
• ability to exclude rivals;
• control over standards;
• access to AI talent;
• strategic partnerships.

4. Major Future Dominance Indicators

A. Control Over Compute

Compute may become one of the most important indicators of AI market power.

Training advanced models can require enormous computational resources. A company controlling substantial quantities of advanced processors, cloud capacity or specialised AI infrastructure may be able to restrict competitors' ability to scale.

Relevant indicators include:

• GPU/accelerator capacity;
• access to advanced processors;
• data-centre capacity;
• cloud reservations;
• inference capacity;
• electricity availability;
• geographic concentration of computing resources;
• long-term exclusive supply agreements.

The CMA has expressly emphasised continued access to inputs such as compute as an important condition for competitive AI markets.

Competition concern

If a vertically integrated firm controls both:

compute → cloud → foundation model → AI application

it may potentially disadvantage independent AI developers.

5. Data Control as a Dominance Indicator

Data may function as a competitive input rather than merely an asset.

Important indicators include:

• quantity of proprietary data;
• uniqueness of the data;
• quality and relevance;
• real-time data access;
• frequency of data updates;
• ability to combine datasets;
• exclusive data contracts;
• user-generated data;
• search-query data;
• behavioural data.

The critical question is not merely:

"How much data does the firm possess?"

but:

"Can competitors realistically reproduce the same data advantage?"

A dataset generated continuously through millions of users can create a self-reinforcing feedback loop:

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

That feedback loop may constitute an important future dominance indicator.

6. Model Performance and AI Quality

Traditional competition law often measures market power through prices and output.

AI markets frequently involve products offered at zero monetary price.

Therefore, competition authorities may increasingly examine:

• benchmark performance;
• reasoning capability;
• accuracy;
• hallucination rates;
• latency;
• multimodal capabilities;
• context windows;
• reliability;
• tool-use capabilities;
• agentic performance;
• inference efficiency.

However, technical superiority by itself should not automatically equal legal dominance.

A firm becomes more competitively significant when technological superiority combines with barriers that prevent rivals from catching up.

7. Network Effects

AI markets can develop strong network effects.

For example:

Users → interactions → data → model improvement → better AI → additional users

This can create a positive feedback loop.

Network effects may become particularly powerful where AI is embedded in:

• social networks;
• search;
• operating systems;
• office software;
• e-commerce;
• financial platforms;
• healthcare systems;
• enterprise software.

The CMA has specifically warned about feedback loops and the possibility that early movers could obtain entrenched advantages.

8. Switching Costs

Switching costs may be one of the strongest indicators of future AI dominance.

They can arise from:

• proprietary APIs;
• model-specific fine-tuning;
• incompatible data formats;
• proprietary agent tools;
• customised enterprise workflows;
• cloud commitments;
• employee training;
• accumulated prompts;
• embedded AI applications;
• dependence on proprietary plugins.

For example:

Company A's AI → customised data → employee training → API integration → automated workflows

may make migration to Company B extremely expensive.

The FTC's study of major cloud/AI partnerships specifically identified potential increases in contractual and technical switching costs as a competition concern.

9. Vertical Integration

Future AI dominance analysis will increasingly examine vertical integration.

A firm may control:

Semiconductors → Cloud → Data → Foundation Model → Operating System → Application → Distribution

This can produce efficiencies, but it can also create opportunities for foreclosure.

Potential indicators include:

• preferential access to internal computing;
• discriminatory cloud pricing;
• tying AI models to cloud services;
• tying AI assistants to operating systems;
• preferential API access;
• self-preferencing;
• refusal to interoperate.

The EU's 2026 DMA measures concerning Google's Android AI interoperability illustrate how access to operating-system functionality can become relevant to AI competition.

10. AI Ecosystem Control

Dominance may increasingly be ecosystem-based rather than product-based.

Consider:

Search + browser + operating system + cloud + AI assistant + advertising + data

or:

Cloud + foundation model + developer platform + enterprise software + AI agents

The relevant question becomes whether the undertaking can leverage power from one market into another.

This represents a movement from:

product dominance

toward:

ecosystem dominance.

11. Strategic Partnerships as a Dominance Indicator

AI development increasingly involves partnerships between:

• cloud providers;
• AI developers;
• semiconductor companies;
• software firms;
• data providers;
• application developers.

Partnerships can be pro-competitive, but their structure matters.

Competition authorities may examine:

• exclusivity;
• revenue-sharing;
• equity rights;
• governance rights;
• access restrictions;
• technical dependencies;
• information exchange;
• preferential computing;
• restrictions on alternative suppliers.

The FTC's investigation of major cloud/AI partnerships examined precisely these issues, including access to compute, engineering talent, switching costs and access to commercially sensitive information.

12. Talent as a Dominance Indicator

AI markets depend heavily upon specialised human capital.

Relevant indicators include:

• concentration of AI researchers;
• acquisition of AI teams;
• employment restrictions;
• exclusive recruitment arrangements;
• compensation advantages;
• control over specialised engineering teams.

A firm acquiring an emerging AI company may obtain not merely technology but an entire talent ecosystem.

Consequently, competition authorities may examine whether acquisitions remove potential competitors by eliminating scarce AI expertise.

13. API and Interoperability Control

APIs can become the equivalent of strategic infrastructure.

A dominant AI provider could potentially:

• limit API access;
• impose discriminatory conditions;
• increase API prices;
• restrict interoperability;
• degrade competitors' access;
• prevent portability;
• impose exclusivity.

The future dominance indicator is therefore:

How much competitive activity depends upon access to the firm's AI interface?

Where a large ecosystem depends on one AI API, control over that interface may become a significant source of market power.

14. Autonomous AI Agents and Dominance

AI agents introduce an additional dimension.

An agent may:

• search;
• negotiate;
• purchase;
• book;
• execute financial transactions;
• manage inventories;
• interact with platforms;
• select suppliers.

The dominant firm could therefore control not merely information but economic decision-making infrastructure.

Potential indicators include:

• number of transactions executed by agents;
• access to transaction interfaces;
• control over agent protocols;
• control over identity and authentication;
• agent interoperability;
• ability to determine which suppliers agents recommend.

This may shift competition analysis from:

"Which website does the consumer visit?"

to:

"Which AI agent determines the consumer's choice?"

15. Algorithmic Self-Preferencing

AI systems may rank or recommend products.

A vertically integrated AI company may have an incentive to favour:

• its own applications;
• its own cloud services;
• affiliated merchants;
• affiliated advertising services;
• its own AI models.

Future dominance analysis should therefore examine:

1. ranking algorithms;
2. recommendation systems;
3. access to data;
4. default settings;
5. model-selection rules;
6. visibility of competing services.

The European Commission's 2026 DMA action concerning Google's preferential treatment of its own services in Search demonstrates the continuing importance of self-preferencing in digital competition.

16. Six Important Case Laws

1. United Brands v Commission

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

Principle

The European Court of Justice explained the concept of a dominant position and emphasised the ability of an undertaking to behave independently of competitors and customers.

Relevance to AI

The principle can be adapted to AI ecosystems.

Instead of asking only whether an AI firm has a particular percentage of sales, authorities can examine whether it possesses sufficient economic strength to operate independently because competitors lack realistic alternatives.

Future indicator

Ability to act independently despite competitive constraints.

2. AKZO Chemie BV v Commission

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

Principle

The case established important principles concerning dominance and predatory pricing, including the relevance of cost and pricing evidence.

AI relevance

AI companies frequently operate with:

• large fixed costs;
• low marginal inference costs;
• subsidised services;
• free AI products.

Therefore, future cases may need to distinguish legitimate investment and scale economies from exclusionary below-cost strategies.

Future indicator

Ability to sustain below-cost or zero-price strategies because of financial and ecosystem strength.

3. Bronner v Mediaprint

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

Principle

The case is central to the European essential-facilities/refusal-to-deal doctrine.

AI relevance

Suppose an AI platform controls an infrastructure that competitors cannot reasonably reproduce.

Possible examples include:

• unique AI infrastructure;
• essential APIs;
• proprietary datasets;
• critical interoperability interfaces.

The difficult legal question would be whether access is genuinely indispensable and whether duplication is realistically possible.

Future indicator

Indispensability of infrastructure or data access.

4. Microsoft Corp. v Commission

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

Principle

The case concerned exclusionary conduct involving interoperability information and tying.

AI relevance

It is highly relevant to future AI ecosystems because interoperability may become a principal competitive bottleneck.

Potentially analogous situations include:

• AI assistants and operating systems;
• foundation models and productivity suites;
• AI agents and enterprise platforms;
• proprietary AI protocols.

Future indicator

Control over interoperability necessary for rivals to compete.

5. Google Shopping

Case: Google Search (Shopping), Case AT.39740, European Commission; General Court judgment in Google and Alphabet v Commission, Case T-612/17 (2021)

Principle

The Google Shopping litigation concerned Google's treatment of its own comparison-shopping service within general search results.

AI relevance

The same analytical issue may arise when AI-generated answers replace traditional search rankings.

Imagine an AI assistant answering:

"Which travel service should I use?"

If the AI provider owns a travel platform, competition authorities could examine whether the model systematically favours its own service.

Future indicator

Control over AI-generated rankings and recommendations combined with vertical integration.

6. Intel v Commission

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

Principle

The case addressed exclusionary rebates and the assessment of whether conduct is capable of foreclosing equally efficient competitors.

AI relevance

Cloud providers and AI infrastructure suppliers may use:

• volume discounts;
• loyalty incentives;
• preferential computing agreements;
• capacity commitments.

The economic effects of such arrangements may therefore become important.

Future indicator

Ability of contractual incentives to foreclose rival AI providers.

17. Additional Important Case Laws

7. Google Android

Case: Google Android, Case AT.40099

The case concerned Google's practices involving Android, including tying and restrictions affecting competing services.

AI relevance

AI assistants increasingly operate through:

• mobile operating systems;
• app stores;
• browsers;
• search;
• default settings.

Control of these distribution channels may therefore become a significant indicator of AI-related market power.

8. Qualcomm

Case: Qualcomm (Exclusivity Payments), Case T-235/18

The case concerned exclusivity arrangements and their potential exclusionary effects.

AI relevance

Comparable questions could arise where an AI infrastructure provider obtains commitments that prevent customers from sourcing competing AI technologies.

Indicator

Exclusive access to strategically important customers or inputs.

9. Aspen Skiing

Case: Aspen Skiing Co. v Aspen Highlands Skiing Corp., 472 U.S. 585 (1985)

Principle

The US Supreme Court considered circumstances in which termination of a prior course of dealing could constitute exclusionary conduct.

AI relevance

A dominant AI ecosystem that suddenly withdraws previously available interoperability, API access or technical integration could potentially raise analogous questions.

Indicator

Strategic withdrawal of previously available access.

10. United States v Microsoft

Case: United States v Microsoft Corp., 253 F.3d 34 (D.C. Cir. 2001)

Principle

The case dealt extensively with network effects, operating-system dominance, exclusionary agreements and leveraging control over a platform.

AI relevance

It provides a particularly important conceptual framework for AI ecosystems.

A future AI platform could resemble an operating system if developers, users and applications depend upon it.

Indicator

Platform control + network effects + exclusionary conduct.

18. A Future AI Dominance Indicator Framework

A competition authority could potentially examine the following matrix:

19. Dominance Is Different From Innovation

An important legal distinction must be maintained.

A company does not become dominant merely because:

• its AI model is technically superior;
• it invests heavily in research;
• it has many users;
• it develops a successful product;
• it has a large valuation.

Competition law protects competition rather than guaranteeing identical market shares.

The central issue is whether the firm's economic power, together with its conduct and structural advantages, enables it to restrict effective competition.

20. The Role of Contestability

Future AI competition policy is likely to focus heavily on contestability.

A market is more contestable where:

• new AI developers can obtain compute;
• data can be accessed lawfully;
• customers can switch providers;
• models can interoperate;
• APIs are accessible;
• AI systems are portable;
• cloud services are not excessively locked in;
• acquisitions do not systematically eliminate emerging rivals.

The CMA's AI work has specifically emphasised the importance of preventing early movers, economies of scale and feedback loops from producing entrenched and disproportionate advantages.

21. AI Dominance and Digital Gatekeepers

The distinction between traditional dominance law and ex ante digital regulation is becoming increasingly important.

The EU Digital Markets Act has already designated major technology companies as gatekeepers based on specified statutory criteria.

More recently, the European Commission stated that Amazon Web Services and Microsoft Azure appeared to have entrenched positions in EU cloud services and identified AI portfolios and partnerships as an important factor in cloud procurement.

This demonstrates a broader regulatory development:

competition analysis → digital gatekeeping → AI ecosystem regulation.

22. Future Competition Concerns

Several future scenarios deserve particular attention.

1. AI-cloud foreclosure

A cloud provider may make its own AI models easier or cheaper to access than rival models.

2. AI-search self-preferencing

An AI search engine may systematically recommend its affiliated services.

3. AI-agent foreclosure

An AI agent could exclude competing merchants or service providers.

4. Data accumulation

A dominant platform may combine search, social, commerce and AI data.

5. AI acquisition strategies

Large firms may acquire promising AI developers before they become significant competitors.

6. Compute bottlenecks

Scarcity of advanced computing resources may create structural entry barriers.

7. Talent concentration

Acquisition or hiring strategies may substantially reduce access to specialised AI expertise.

8. Interoperability restrictions

Dominant ecosystems may make it technically difficult for rival AI systems to interact with their platforms.

23. Proposed Multi-Layer Dominance Test

Future AI competition analysis can be conceptualised through five stages.

Stage 1 — Define the relevant market

Identify whether the market concerns:

• AI models;
• AI inference;
• cloud AI;
• AI applications;
• AI agents;
• data;
• compute;
• AI distribution.

Stage 2 — Measure structural power

Assess:

• market shares;
• compute;
• data;
• infrastructure;
• capital;
• talent;
• ecosystem size.

Stage 3 — Assess dynamic advantages

Examine:

• network effects;
• feedback loops;
• learning effects;
• switching costs;
• economies of scale.

Stage 4 — Examine exclusionary mechanisms

Look for:

• tying;
• bundling;
• self-preferencing;
• exclusivity;
• discriminatory access;
• refusal to deal;
• interoperability restrictions;
• predatory strategies.

Stage 5 — Determine effects

Consider whether the conduct:

• excludes rivals;
• raises entry barriers;
• reduces innovation;
• restricts consumer choice;
• increases dependency;
• entrenches market power.

24. A Useful Formula for Future AI Dominance Analysis

A conceptual framework can be expressed as:

AI Dominance Risk = Structural Power + Input Control + Network Effects + Switching Costs + Ecosystem Leverage + Exclusionary Conduct

The formula is not a legal test. It is an analytical framework for identifying circumstances requiring deeper competition-law investigation.

25. Remedies

Where dominance and anticompetitive conduct are established, possible remedies may include:

Structural remedies

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

Behavioural remedies

• non-discrimination;
• interoperability;
• API access;
• data portability;
• prohibition of tying;
• restrictions on exclusivity.

Data remedies

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

Technical remedies

• open interfaces;
• model portability;
• interoperability protocols.

Merger remedies

• behavioural commitments;
• access commitments;
• divestiture;
• restrictions on exclusive arrangements.

26. Conclusion

The future concept of dominance in AI-driven economies is likely to move beyond a simple assessment of market share.

The most important indicators may increasingly be:

1. control over compute;
2. control over unique data;
3. foundation-model capabilities;
4. network and feedback effects;
5. switching costs;
6. cloud and infrastructure dependence;
7. AI talent concentration;
8. API and interoperability control;
9. distribution and default access;
10. vertical integration;
11. strategic partnerships and exclusivity;
12. ecosystem-wide leverage; and
13. the ability to foreclose emerging competitors.

The existing jurisprudence of United Brands, AKZO, Bronner, Microsoft, Google Shopping, Intel, Qualcomm, Aspen Skiing and US Microsoft provides useful legal foundations, but AI markets require these principles to be applied to new economic realities.

The central future competition-law question will therefore not simply be:

"How large is the AI company?"

It will increasingly be:

"How difficult is it for an equally efficient or innovative competitor to obtain the inputs, data, compute, distribution, interoperability and customers necessary to challenge it?"

That shift—from market share to durable competitive infrastructure and ecosystem power—is likely to be one of the defining issues in competition law for AI-driven economies.