Competition Law and Evolving Theories of Dominance in Intelligence Ecosystems

1. Introduction

The emergence of intelligence ecosystems—built around artificial intelligence (AI), foundation models, cloud computing, semiconductor accelerators, data, operating systems, application stores, developer tools, and digital distribution—requires competition law to move beyond the traditional idea of dominance in a single, clearly defined product market.

An intelligence ecosystem may involve:

Data → Computing infrastructure → AI chips → Cloud → Foundation model → AI application → Operating system → Distribution platform → Users → Feedback/data → Improved model

This creates a form of multi-layered and interconnected market power. A firm may not dominate every layer, but control over one critical layer can enable leverage into adjacent layers.

The U.S. Federal Trade Commission has specifically examined Microsoft–OpenAI, Amazon–Anthropic and Google–Anthropic arrangements, identifying potential concerns involving access to computing resources, switching costs, engineering talent and access to competitively sensitive information.

The evolving theory therefore asks not merely:

“Does the undertaking have a dominant share in a relevant market?”

but also:

“Does its position across an interconnected intelligence ecosystem allow it to control entry, interoperability, inputs, distribution, innovation or competitive pathways?”

2. Meaning of an Intelligence Ecosystem

An intelligence ecosystem can be understood as a network of complementary technologies and markets through which computational intelligence is developed, trained, deployed and distributed.

Principal layers

The important characteristic is interdependence.

A foundation-model provider may need enormous computing resources; the cloud provider may simultaneously operate a competing AI model; the cloud provider may also control enterprise software through which the AI service reaches customers.

This produces potential vertical, horizontal and conglomerate theories of harm simultaneously.

3. Traditional Dominance Versus Ecosystem Dominance

Traditional competition law generally examines:

1. relevant product market;
2. geographic market;
3. market share;
4. barriers to entry;
5. countervailing buyer power;
6. conduct of the dominant undertaking; and
7. effects on competition.

Ecosystem competition complicates each element.

Traditional model

Market → Dominant undertaking → Exclusionary conduct → Competitive harm

Ecosystem model

Multiple markets → Strategic bottleneck → Cross-market leverage → Foreclosure → Entrenchment

For example, a cloud provider may possess substantial market power in cloud computing and use contractual arrangements with an AI developer to influence competition in foundation models.

The FTC's AI partnerships study specifically identified the possibility that such arrangements could increase switching costs and affect access to important AI inputs.

4. Evolution of Theories of Dominance

A. From Market Share to Strategic Position

Market share remains important, but ecosystem dominance requires examination of strategic position.

A firm with a moderate share may nevertheless possess substantial power if it controls an indispensable:

• API;
• cloud infrastructure;
• operating system;
• AI accelerator;
• dataset;
• distribution channel;
• developer ecosystem; or
• technical standard.

Thus:

Economic importance can sometimes be more informative than numerical market share alone.

5. Bottleneck Dominance

An intelligence ecosystem frequently contains bottleneck resources.

Examples include:

• advanced AI chips;
• hyperscale computing capacity;
• high-quality training data;
• search indexes;
• app-store distribution;
• enterprise software distribution;
• specialized AI infrastructure.

Where competitors cannot reasonably reproduce the bottleneck, the owner may possess significant gatekeeping power.

The European Commission's recent work concerning cloud services illustrates the growing regulatory interest in cloud infrastructure as an important gateway for businesses and digital services, particularly because cloud computing is fundamental to AI development.

6. Ecosystem Leveraging

Ecosystem leveraging occurs where power acquired in one market is used to strengthen a position in another.

For example:

Cloud dominance → AI infrastructure advantage → AI model advantage → Enterprise distribution advantage

or:

Operating-system dominance → Default AI assistant → User data → Model improvement → Reinforced AI position

The legal concern is not simply vertical integration.

Vertical integration can generate substantial efficiencies. The competition question is whether integration is accompanied by conduct that forecloses competitors or restricts competitive access.

7. Defensive Foreclosure

One of the most important emerging theories is defensive foreclosure.

This describes conduct designed not merely to obtain immediate profits but to prevent a future competitor from becoming a competitive threat.

For example:

Dominant platform → identifies emerging AI competitor → restricts access to distribution/data/cloud → competitor cannot scale → future competitive threat disappears.

Academic analysis of ecosystem competition has identified blocking entry paths and defensive foreclosure as important theories for analysing modern ecosystems.

This is particularly significant for AI because today's small foundation-model developer may become tomorrow's major platform competitor.

8. Data-Based Dominance

Data can generate several forms of competitive advantage:

First-mover data advantage

More users → more interaction data → better AI → more users.

Cross-market data advantage

Search data + consumer behaviour + enterprise data + AI interactions may create advantages unavailable to independent competitors.

Feedback-loop dominance

Users → data → model improvement → better product → more users

This creates a self-reinforcing ecosystem.

However, possessing large quantities of data should not automatically constitute dominance. Competition authorities must examine:

• uniqueness;
• substitutability;
• quality;
• timeliness;
• accessibility;
• portability; and
• actual competitive significance.

9. Computing Power as a Competitive Input

AI differs from many earlier digital markets because sophisticated models require enormous computational resources.

Consequently, compute access itself may become a competition parameter.

Potential concerns include:

• exclusive cloud agreements;
• preferential allocation of GPUs;
• discriminatory pricing;
• capacity reservations;
• restrictions on multi-cloud deployment;
• technical incompatibility;
• high switching costs;
• preferential access to new AI hardware.

The FTC has specifically identified computing resources as a potentially important input affected by major AI-cloud partnerships.

10. AI–Cloud Vertical Integration

Consider:

Cloud Provider A

↓ owns infrastructure

↓ finances

AI Developer B

↓ develops model

↓ distributed through

Cloud Provider A's enterprise ecosystem

If A also competes with B's AI products, competition law may examine whether A can use its infrastructure position to disadvantage rival AI developers.

Relevant questions include:

1. Is the cloud provider dominant?
2. Is the AI developer dependent upon it?
3. Are exclusivity provisions present?
4. Are switching costs substantial?
5. Can rival cloud providers realistically supply capacity?
6. Does the arrangement give the cloud provider competitively sensitive information?
7. Does the arrangement restrict model distribution?

The FTC's 2025 report specifically discussed these types of concerns concerning major AI-cloud partnerships.

11. Interoperability as a Competition Issue

AI ecosystems increasingly depend upon interoperability.

Examples:

• model APIs;
• cloud portability;
• operating-system access;
• data portability;
• plugin interoperability;
• identity systems;
• AI-agent interoperability.

A dominant undertaking could theoretically weaken competition by preventing rivals from interoperating with its ecosystem.

Therefore, interoperability can become an important remedy as well as a competition parameter.

The European Commission has identified AI-related interoperability and access to search data among areas receiving regulatory attention under the DMA framework.

12. Tying and Bundling

AI may be bundled with:

• operating systems;
• browsers;
• productivity software;
• cloud services;
• search engines;
• cybersecurity products;
• enterprise software.

The fundamental competition-law issue is:

Does the undertaking use dominance in Product A to restrict competition in Product B?

The issue is particularly relevant where consumers cannot realistically obtain the dominant product without receiving the AI product.

13. Six Important Case Laws

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

Facts

Microsoft possessed substantial power in the market for Intel-compatible PC operating systems and engaged in practices concerning Internet Explorer and competing browsers.

Legal significance

The case is important for intelligence ecosystems because it demonstrates how dominance in one technological layer can be used to influence competition in an adjacent layer.

Relevance to AI

The analogy can arise where:

Operating-system dominance → AI assistant integration → distribution advantage

or:

Enterprise software dominance → AI productivity tool → rival AI foreclosure

The case illustrates the importance of examining ecosystem leverage and distribution advantages, rather than treating each technological product as completely isolated.

14. Google LLC v. United States, 2024 District Court Proceedings

Facts

The U.S. litigation concerning Google's search business examined agreements under which Google obtained important default positions for search distribution.

Legal significance

The proceedings demonstrate the competitive importance of defaults and distribution channels.

Relevance to intelligence ecosystems

AI services may similarly compete for:

• default assistant status;
• browser integration;
• search integration;
• smartphone placement;
• enterprise defaults.

An AI model may be technologically strong but commercially disadvantaged if a rival controls the principal distribution gateway.

Thus, distribution dominance can become AI dominance.

15. Google Android / European Commission, Google Android Case

Case

Google Android, Case AT.40099

Facts

The European Commission examined Google's conduct concerning the Android ecosystem, including tying and distribution arrangements involving Google applications.

Legal significance

The case demonstrated how competition authorities can examine an ecosystem containing:

• operating systems;
• applications;
• app stores;
• search;
• mobile distribution.

Relevance to intelligence ecosystems

AI may become another layer inside an operating-system ecosystem.

For example:

Android/OS → AI assistant → search → applications → data

The case therefore provides an important conceptual framework for analysing multi-layer technological ecosystems.

16. Google Search (Shopping), Case AT.39740

Facts

The European Commission found that Google had systematically favoured its comparison-shopping service in general search results.

Legal significance

The case is important for self-preferencing.

Relevance to AI ecosystems

An AI-platform operator may potentially prefer:

• its own model;
• its own AI applications;
• its own agents;
• its own cloud;
• its own data services;

over competing services.

The key competition-law question becomes whether control over a gateway enables the operator to systematically advantage its own downstream products.

17. Qualcomm, Case AT.39711

Facts

The European Commission's Qualcomm investigation concerned exclusionary payments and the competitive position of Qualcomm in baseband chipsets.

Legal significance

The case illustrates the importance of strategic inputs and conditional commercial arrangements in technology markets.

Relevance to AI

The AI ecosystem similarly depends upon critical upstream inputs such as:

• GPUs;
• accelerators;
• networking;
• memory;
• specialised chips.

Control over an essential or highly significant technological input can therefore affect competition downstream.

The emerging AI competition debate has particularly focused on the role of advanced accelerators and other critical AI inputs.

18. Google AdSense, Case AT.40411

Facts

The European Commission examined Google's contractual restrictions concerning online advertising intermediation.

Legal significance

The case demonstrates how contractual restrictions can become exclusionary where a dominant platform controls an important intermediary layer.

Relevance to AI

AI ecosystems may develop similar intermediary layers:

AI model → API → AI application → enterprise customer

or:

Cloud → model marketplace → AI application

Restrictions imposed at the intermediary level may therefore affect competitors throughout the ecosystem.

19. Amazon Marketplace – European Commission

The European Commission's Amazon investigation concerned the use of non-public marketplace seller data and Amazon's dual role as marketplace operator and competing retailer.

Legal significance

The case illustrates the competition problem created when an ecosystem operator simultaneously:

1. controls infrastructure;
2. receives information from ecosystem participants; and
3. competes against those participants.

Relevance to AI

The same structural concern may arise where:

AI platform → hosts rival developers → receives their usage/business information → competes with them

This makes information asymmetry a significant emerging theory of ecosystem power.

20. Microsoft Teams, European Commission

The European Commission investigated Microsoft's tying of Teams with Microsoft 365 and Office 365.

In September 2025, the Commission adopted a commitment decision addressing concerns under Article 102 TFEU concerning Teams and Microsoft's SaaS productivity applications.

AI significance

This is especially relevant because productivity ecosystems are increasingly becoming AI ecosystems.

A similar structure could arise where:

Office/Productivity dominance → AI assistant → enterprise users → AI distribution

The principal competition question becomes whether customers are effectively induced or required to adopt the dominant firm's AI product because of its position in an adjacent productivity ecosystem.

21. Emerging AI Partnership Cases and Regulatory Proceedings

Although not all current AI competition matters have produced final judicial judgments, they are highly relevant to the evolving doctrine.

The FTC's 2025 study examined:

• Microsoft–OpenAI;
• Amazon–Anthropic; and
• Google–Anthropic.

The FTC identified potential concerns involving:

• computing resources;
• engineering talent;
• switching costs;
• contractual restrictions;
• access to sensitive information; and
• possible future concentration. 

These developments demonstrate that AI partnerships themselves may become competition-law objects, even where they are not conventional mergers.

22. New Theory: Investment-Based Dominance

Traditional dominance analysis usually focuses on ownership and control.

AI ecosystems introduce another possibility:

Strategic minority investment + infrastructure dependence + contractual rights + technical integration

may collectively produce significant competitive influence without formal acquisition.

Consequently, authorities may need to examine:

• voting rights;
• board rights;
• consultation rights;
• exclusivity;
• revenue sharing;
• cloud commitments;
• intellectual-property rights;
• model access;
• information rights.

The FTC specifically examined consultation, control and exclusivity provisions in major AI partnerships.

23. Switching Costs and Lock-In

AI ecosystems can generate switching costs at several levels.

Technical

Moving a model between cloud providers may require substantial engineering work.

Economic

Existing infrastructure investments may become stranded.

Data

Users may lose accumulated prompts, workflows and customisations.

Organisational

Employees may be trained around one AI system.

Contractual

Long-term cloud or licensing commitments may restrict migration.

Therefore:

Low consumer monetary price does not necessarily mean low switching costs.

This is especially important in enterprise AI.

24. Network Effects

Intelligence ecosystems can exhibit several network effects.

Direct network effect

More users → more interactions.

Indirect network effect

More developers → more applications → more users.

Data network effect

More usage → more data → improved system.

Compute network effect

Greater scale → better infrastructure utilisation → lower cost → greater scale.

Ecosystem network effect

More integrations → greater functionality → greater user dependence.

This can produce self-reinforcing dominance.

25. Dynamic Competition and Innovation

AI markets require particular attention to innovation competition.

A firm may not currently possess overwhelming market share but could acquire power by preventing the development of future competing technologies.

This gives rise to:

• nascent competition;
• innovation foreclosure;
• killer-acquisition concerns;
• strategic investments;
• interoperability restrictions;
• access restrictions;
• talent acquisition.

The central question becomes:

Could today's conduct prevent tomorrow's competitive architecture from developing?

26. Talent as a Competitive Input

AI competition is unusually dependent upon:

• machine-learning researchers;
• engineers;
• data scientists;
• chip architects;
• safety researchers;
• specialised technical personnel.

Consequently, concentration in AI labour markets can potentially reinforce product-market power.

The FTC has expressly identified engineering talent among the inputs potentially affected by major AI partnerships.

This expands traditional dominance analysis beyond products and consumers to labour and innovation inputs.

27. AI Ecosystems and Essential-Facility Theory

A highly important future issue concerns whether certain AI infrastructure could constitute an essential facility.

Possible candidates might include:

• specialised computing infrastructure;
• critical interoperability interfaces;
• unique datasets;
• technical standards.

However, not every valuable input is legally an essential facility.

Authorities would need to examine factors such as:

1. indispensability;
2. lack of realistic alternatives;
3. duplication feasibility;
4. competitive necessity;
5. objective justification;
6. proportionality of access obligations.

28. Refusal to Deal and AI Access

Potential disputes could arise if a dominant ecosystem refuses access to:

• APIs;
• computing resources;
• datasets;
• operating-system interfaces;
• app distribution;
• model marketplaces.

The legal assessment would depend upon the relevant jurisdiction and applicable doctrine.

A refusal is not automatically unlawful merely because the supplier is important.

The critical question is whether the refusal has an exclusionary effect without adequate objective justification.

29. Self-Preferencing in Intelligence Ecosystems

Self-preferencing may take several forms:

Search engine → own AI answer

Cloud → own foundation model

App store → own AI application

OS → own assistant

Enterprise suite → own AI copilot

AI marketplace → own agent

The competition concern arises where the platform controls the gateway through which rivals must compete.

This builds upon the logic developed in digital-platform cases involving search, operating systems and marketplaces.

30. Algorithmic and AI-Enabled Collusion

Intelligence ecosystems may also facilitate coordination.

AI systems can:

• monitor competitors;
• change prices rapidly;
• predict competitor behaviour;
• communicate through automated systems;
• optimise pricing algorithms.

The traditional distinction between:

independent parallel conduct

and

concerted practice

may therefore become more difficult to apply.

Competition law must distinguish legitimate algorithmic optimisation from evidence of an agreement or coordinated conduct.

31. Competition Between Ecosystems

Competition may increasingly occur between ecosystems rather than individual products.

For example:

Ecosystem A

Cloud + chips + model + operating system + productivity suite

versus

Ecosystem B

Cloud + model + marketplace + enterprise applications

The relevant competitive unit may therefore be broader than a single product.

Nevertheless, authorities must avoid defining the market so broadly that genuine competitive constraints disappear from analysis.

32. Possible Theories of Harm

The principal theories include:

1. Tying
2. Bundling
3. Exclusive dealing
4. Self-preferencing
5. Refusal to deal
6. Input foreclosure
7. Customer foreclosure
8. Data foreclosure
9. Compute foreclosure
10. Interoperability restrictions
11. Predatory pricing
12. Discriminatory access
13. Leveraging
14. Nascent-competitor foreclosure
15. Strategic investment
16. Information advantage
17. Killer acquisition
18. Algorithmic coordination

33. Competition-Law Test for Intelligence Ecosystems

A useful analytical framework is:

Step 1 — Identify the ecosystem

Map:

Infrastructure → Inputs → Model → Distribution → Application → Users

Step 2 — Identify bottlenecks

Ask:

• What cannot easily be replicated?
• Who controls it?

Step 3 — Define relevant markets

Consider:

• product substitutability;
• technology;
• geographic scope;
• supply-side substitution.

Step 4 — Measure market power

Examine:

• market share;
• entry barriers;
• switching costs;
• network effects;
• data advantages;
• control over infrastructure.

Step 5 — Identify leverage

Determine whether power in Market A affects Market B.

Step 6 — Examine conduct

Analyse:

• tying;
• exclusivity;
• refusal;
• discrimination;
• self-preferencing;
• acquisition;
• investment;
• interoperability restrictions.

Step 7 — Assess effects

Consider:

• foreclosure;
• innovation;
• prices;
• quality;
• consumer choice;
• entry;
• technical development.

Step 8 — Consider efficiencies

Potential efficiencies include:

• improved security;
• lower costs;
• better model performance;
• infrastructure integration;
• interoperability;
• faster innovation.

34. Remedies

Competition authorities may consider several remedies.

Structural remedies

• divestiture;
• separation of business units;
• restrictions on acquisitions.

Behavioural remedies

• non-discrimination;
• interoperability;
• access obligations;
• prohibition of exclusivity;
• data portability;
• transparency.

Ecosystem remedies

• multi-cloud portability;
• API access;
• model interoperability;
• developer neutrality;
• separation of infrastructure and downstream competition.

The appropriate remedy depends on the demonstrated competitive harm; simply requiring access can itself create issues concerning incentives to innovate.

35. India and Intelligence Ecosystems

In India, the principal statutory framework is the Competition Act, 2002, particularly:

• Section 4 — abuse of dominant position;
• Section 5 — combinations;
• Section 6 — regulation of combinations;
• Section 19 — inquiry;
• Section 26 — investigation;
• Sections 27 and 28 — orders and structural remedies.

The Indian framework can accommodate many intelligence-ecosystem concerns through established concepts such as:

• relevant market;
• dominance;
• denial of market access;
• discriminatory conditions;
• leveraging;
• tying;
• unfair conditions;
• combinations.

India's digital competition experience also demonstrates the importance of analysing platform ecosystems rather than simply looking at conventional price effects.

36. European Union Approach

The EU provides two complementary frameworks:

Article 102 TFEU

Addresses abuse of dominance.

Digital Markets Act

Introduces ex ante obligations for designated gatekeepers.

The distinction is significant because ecosystem regulation increasingly moves from:

ex post punishment

toward:

ex ante preservation of contestability.

The European Commission's current regulatory work specifically identifies cloud computing and AI as important areas for maintaining competitive digital markets.

37. United States Approach

The United States primarily relies upon:

• Sherman Act;
• Clayton Act;
• FTC Act;
• DOJ enforcement;
• FTC enforcement.

The U.S. approach remains strongly influenced by:

• monopolization;
• exclusionary conduct;
• mergers;
• tying;
• exclusive dealing;
• foreclosure;
• innovation competition.

The Microsoft litigation remains particularly important because it demonstrates how competition law can address conduct designed to protect an established technological position against emerging technological threats.

38. Evolving Concept of Dominance

The evolution can be summarized as follows:

39. Key Doctrinal Proposition

The most important conceptual development is the movement from:

“Who dominates a market?”

toward:

“Who controls the competitive architecture through which multiple markets operate?”

This does not mean that every large AI ecosystem is legally dominant.

Rather, competition law must distinguish between:

legitimate ecosystem integration

and

strategic ecosystem control that forecloses competitive alternatives.

That distinction is particularly important because AI integration can produce genuine efficiencies while simultaneously creating opportunities for exclusion.

40. Conclusion

The rise of intelligence ecosystems is transforming the theory of dominance from a relatively market-centred concept into a more ecosystem-sensitive concept.

The emerging competition-law analysis increasingly considers:

• computing power;
• AI chips;
• cloud infrastructure;
• data;
• foundation models;
• APIs;
• operating systems;
• application stores;
• enterprise software;
• distribution;
• interoperability;
• strategic investments;
• network effects;
• switching costs; and
• innovation pathways.

The major cases involving Microsoft, Google, Qualcomm, Amazon and other technology ecosystems provide the doctrinal building blocks. Current AI developments involving Microsoft–OpenAI, Amazon–Anthropic and Google–Anthropic demonstrate how those established principles may be applied to the emerging AI architecture. The FTC has expressly warned that such partnerships can affect access to key AI inputs, switching costs and competitively sensitive information.

Accordingly, the modern theory of dominance in intelligence ecosystems can be conceptualised as:

Infrastructure power
↓
Input control
↓
Model advantage
↓
Distribution control
↓
Network/data effects
↓
Ecosystem lock-in
↓
Potential foreclosure of current and future competitors

The central legal challenge is therefore to preserve contestability and innovation without treating integration, scale or technological success as unlawful in themselves.