Competition Law and Synthetic Intelligence Ecosystems and Antitrust

1. Introduction

Synthetic Intelligence Ecosystems may be understood as emerging technology ecosystems in which artificial or machine-generated intelligence is combined with synthetic data, generative AI, autonomous agents, algorithmic decision-making, foundation models, cloud infrastructure, AI chips, application programming interfaces (APIs), digital platforms, robotics, and automated commercial transactions.

From a competition-law perspective, the principal concern is not merely whether an AI system is powerful. The concern is whether control over one layer of the ecosystem enables an undertaking to foreclose rivals, raise entry barriers, restrict access to essential inputs, discriminate against competing applications, facilitate coordination, or extend market power into adjacent markets.

Competition authorities therefore increasingly need to examine AI ecosystems vertically and horizontally rather than considering each AI product in isolation.

2. Meaning of a Synthetic Intelligence Ecosystem

A synthetic intelligence ecosystem can contain several interconnected layers:

1. Computational infrastructure
   • GPUs and AI accelerators
   • data centres
   • cloud computing
   • high-performance computing
2. Data layer
   • training data
   • synthetic data
   • proprietary datasets
   • user-generated data
   • behavioural data
3. Foundation-model layer
   • large language models
   • multimodal models
   • generative AI models
   • autonomous-agent models
4. Developer layer
   • APIs
   • software-development kits
   • model marketplaces
   • AI development platforms
5. Application layer
   • search
   • healthcare
   • finance
   • education
   • productivity
   • autonomous systems
6. Distribution layer
   • app stores
   • operating systems
   • search engines
   • cloud platforms
   • digital assistants
7. Autonomous-agent layer
   • AI agents purchasing products
   • algorithmic negotiation
   • automated pricing
   • automated procurement
   • autonomous allocation of resources.

Competition law can potentially intervene at each of these layers.

3. Why Synthetic Intelligence Ecosystems Create Competition Concerns

The principal competition concern arises from ecosystem concentration.

An undertaking might simultaneously control:

chips → cloud → data → foundation model → API → operating system → application → distribution

Such vertical integration can generate significant advantages that competitors cannot easily replicate.

For example, a cloud provider possessing its own AI infrastructure and foundation model could potentially provide preferential access to its own AI applications while imposing less favourable conditions on competing applications.

The legal question is therefore not simply:

“Is the company large?”

It is:

“Does its control over one or more strategically important layers enable it to restrict effective competition in another layer?”

4. Relevant Competition-Law Framework

A. Market Definition

Traditional competition law begins with the definition of relevant markets.

Synthetic-intelligence ecosystems may require several overlapping markets, including:

• AI compute;
• cloud infrastructure;
• AI accelerator chips;
• foundation models;
• generative-AI services;
• AI APIs;
• AI application markets;
• AI search;
• AI assistants;
• synthetic-data services;
• AI training-data markets.

The traditional SSNIP test may become difficult where services are offered at zero monetary prices.

Authorities may therefore examine:

• quality;
• privacy;
• data access;
• innovation;
• switching costs;
• interoperability;
• compute requirements;
• developer dependence.

5. Market Power in AI Ecosystems

Market power may derive from several sources.

1. Data advantages

Large datasets may improve model performance.

2. Compute advantages

Advanced AI models require enormous computational resources.

3. Network effects

More users can produce more data, developers and applications, increasing ecosystem attractiveness.

4. Economies of scale

AI model development may involve substantial fixed costs.

5. Switching costs

Developers may become dependent upon:

• proprietary APIs;
• cloud infrastructure;
• model-specific architectures;
• proprietary tools;
• data formats.

6. Ecosystem lock-in

A developer using one company's:

cloud + model + API + database + developer tools

may face substantial costs when attempting to migrate to a competitor.

6. Abuse of Dominance

A dominant synthetic-intelligence ecosystem could potentially engage in several forms of exclusionary conduct.

A. Self-Preferencing

An ecosystem operator may rank its own AI service above competing AI applications.

Example:

A search platform operates its own AI assistant and gives that assistant preferential placement over rival AI assistants.

Competition concerns can arise where preferential treatment disadvantages equally capable rivals.

7. Tying and Bundling

A powerful platform might condition access to one service upon acceptance of another.

Examples include:

• AI assistant + operating system;
• foundation model + cloud service;
• AI application + productivity suite;
• AI model + proprietary database;
• AI API + payment service.

The principal question is whether bundling produces efficiencies or instead forecloses competing suppliers.

8. Exclusive Dealing

AI companies may attempt to secure exclusive arrangements involving:

• cloud computing;
• model distribution;
• AI chips;
• data;
• enterprise customers;
• application stores.

Long-term exclusivity can be particularly important where competitors cannot obtain equivalent access to scarce computational resources.

9. Refusal of Access

Competition issues may arise where a dominant ecosystem refuses access to:

• APIs;
• data;
• interoperability interfaces;
• model marketplaces;
• cloud infrastructure;
• essential technical standards.

However, refusal to deal is not automatically unlawful. Competition law normally requires an examination of dominance, indispensability, competitive foreclosure and legitimate business justification.

10. Interoperability and Switching

Synthetic-intelligence ecosystems can create powerful technical switching barriers.

A competitor may have to migrate:

• datasets;
• prompts;
• model weights;
• APIs;
• embeddings;
• applications;
• customer histories;
• security systems.

Competition authorities may therefore consider:

• data portability;
• API interoperability;
• open standards;
• technical compatibility;
• multi-homing.

11. Algorithmic Collusion

AI agents can make competition concerns more complex.

Suppose competing autonomous pricing systems independently learn that maintaining a particular price produces greater profits.

The systems might repeatedly reach similar prices without a traditional human agreement.

This raises the question:

Can autonomous algorithmic behaviour amount to coordination prohibited by competition law?

Traditional cartel law generally requires evidence satisfying the relevant legal standard for an agreement or concerted practice. Mere parallel pricing is not automatically sufficient.

However, deliberate programming, information exchange, algorithmic coordination mechanisms or human instructions can materially change the legal analysis.

12. AI-Assisted Cartels

AI systems may also facilitate traditional cartels.

For example, competing firms might use common software that:

• monitors rivals;
• recommends prices;
• exchanges commercially sensitive information;
• detects deviations from agreed prices.

The technology does not necessarily eliminate responsibility simply because an algorithm performs the final action.

13. Merger Control

Synthetic-intelligence ecosystems present significant merger-control concerns.

A transaction may involve:

• AI model developers;
• cloud providers;
• chip manufacturers;
• data companies;
• AI startups;
• cybersecurity firms;
• robotics companies.

A transaction may be competitively significant even where the target has relatively low current revenue.

Authorities may therefore examine:

• innovation competition;
• future competition;
• access to data;
• access to compute;
• nascent competitors;
• ecosystem effects;
• vertical foreclosure.

14. Killer Acquisitions and Nascent AI Competitors

A dominant ecosystem might acquire a promising AI startup before the startup becomes a substantial competitor.

Traditional turnover thresholds can fail to capture such transactions.

Competition authorities may consequently examine:

• acquisition value;
• strategic importance;
• technological capabilities;
• pipeline products;
• patents;
• talent;
• data assets;
• potential future competition.

15. Vertical Foreclosure

Vertical integration is particularly important in AI.

Consider:

AI chip manufacturer → cloud provider → foundation model → AI application.

If the same company controls all four levels, it might theoretically disadvantage competitors through:

• higher prices;
• degraded service;
• discriminatory access;
• capacity allocation;
• technical incompatibility;
• preferential API performance.

Vertical integration is not itself unlawful. The competition inquiry focuses on whether it produces anticompetitive foreclosure.

16. Predatory Pricing

A major AI provider with substantial financial resources could potentially subsidise AI services for extended periods.

Competition authorities may investigate whether:

• prices are below relevant cost measures;
• losses are strategically incurred;
• rivals are being excluded;
• recoupment is plausible where legally relevant;
• legitimate innovation or introductory pricing explains the conduct.

Low AI prices therefore should not automatically be characterised as predatory.

17. Exclusive Access to Compute

Compute may become a strategically important input.

If a small number of companies control access to advanced AI chips or cloud infrastructure, competitors may experience difficulty developing competing models.

Potential competition concerns include:

• capacity reservation;
• exclusivity;
• discriminatory cloud pricing;
• preferential allocation;
• long-term contracts;
• refusal to supply.

18. Synthetic Data and Competition

Synthetic data can reduce dependence upon scarce real-world datasets.

However, dominant firms may use their control over:

• model-generation systems;
• proprietary datasets;
• validation mechanisms;
• data-generation platforms

to create another layer of market power.

Potential concerns include:

• exclusive synthetic-data licensing;
• discriminatory access;
• bundling;
• interoperability restrictions;
• data-quality discrimination.

19. AI Ecosystem and Consumer Choice

Consumers may face reduced choice when an ecosystem automatically selects:

• the AI model;
• search results;
• recommendations;
• merchants;
• financial products;
• healthcare providers.

Competition authorities may therefore consider whether autonomous decision systems prevent meaningful multi-homing or make competing suppliers invisible.

20. Six Important Case Laws and Enforcement Precedents

Because synthetic intelligence is an emerging field, there are relatively few mature judicial decisions specifically labelled “synthetic intelligence ecosystem”. The most useful authorities therefore come from digital platforms, technology ecosystems, data, tying, self-preferencing, exclusionary conduct, and algorithmic competition.

Case 1: United States v. Microsoft Corp. (2001)

Facts

Microsoft was found to have unlawfully maintained its monopoly in the market for Intel-compatible PC operating systems through exclusionary conduct involving Internet browsers.

Competition principle

The case demonstrates the importance of examining how dominance in one technological layer can be used to protect dominance against emerging competition in another layer.

Relevance to synthetic intelligence

An AI ecosystem could similarly involve:

operating system → AI assistant → search → applications.

The Microsoft precedent therefore illustrates the risks of using control over a platform to disadvantage an emerging technological competitor.

21. Case 2: European Commission v. Google Shopping

Facts

The European Commission found that Google had abused a dominant position by favouring its own comparison-shopping service in general search results.

Competition principle

The case is highly relevant to self-preferencing.

AI relevance

An AI ecosystem may contain:

general search → AI-generated answers → AI recommendations → commercial services.

If an ecosystem operator systematically gives its own AI service preferential treatment over rival AI providers, the Google Shopping framework provides an important analytical reference.

22. Case 3: Google Android — European Commission

Facts

The European Commission found several practices involving Google's Android ecosystem to be abusive, including restrictions concerning device manufacturers and application distribution.

Competition principle

The case illustrates how dominance at one technological layer can be leveraged through contractual restrictions affecting adjacent markets.

AI relevance

Comparable concerns could arise where access to:

• an operating system,
• AI assistant,
• app store,
• AI API

is tied together through contractual restrictions.

23. Case 4: Google Search (AdSense) — European Commission

Facts

The Commission examined Google's contractual restrictions concerning search advertising and found that certain restrictions limited the ability of competing search-advertising providers to compete.

Competition principle

The case demonstrates the significance of contractual foreclosure in digital ecosystems.

AI relevance

Similar contractual restrictions could theoretically arise in AI ecosystems through:

• API exclusivity;
• cloud restrictions;
• model-distribution arrangements;
• advertising restrictions;
• developer contracts.

24. Case 5: United States v. Google — Search and Search Advertising Litigation

Facts

The U.S. litigation concerning Google's search business examined agreements and practices affecting distribution and default placement of search services.

Competition principle

Control over distribution can materially affect competition even where alternative technologies technically exist.

AI relevance

AI assistants may increasingly depend on:

• mobile defaults;
• browsers;
• operating systems;
• search interfaces;
• voice assistants.

Control over these distribution channels can therefore become strategically significant for AI competition.

25. Case 6: FTC v. Qualcomm

Facts

The Federal Trade Commission challenged Qualcomm's licensing practices involving cellular-standard-essential patents.

The litigation concerned the relationship between intellectual-property licensing and competition in downstream chip markets.

Competition principle

The case illustrates the complex interaction between:

• technology inputs;
• licensing;
• downstream competition;
• exclusionary conduct.

AI relevance

AI ecosystems may similarly depend upon proprietary:

• model architectures;
• patents;
• chips;
• software;
• interfaces.

Control over an upstream technological input can therefore have downstream competitive consequences.

26. Case 7: FTC v. Meta Platforms

Facts

The U.S. Federal Trade Commission challenged Meta's acquisition of Instagram and WhatsApp as part of its broader theory concerning the maintenance of monopoly power in personal social networking.

Competition principle

The case illustrates the importance of examining acquisitions of potential or nascent competitors rather than focusing exclusively on existing market shares.

AI relevance

The principle is particularly significant for AI because today's small AI developer may become tomorrow's major platform competitor.

27. Case 8: Illumina/GRAIL

Facts

European competition authorities examined Illumina's acquisition of GRAIL, a company developing early cancer-detection tests.

The transaction raised concerns concerning the relationship between an established technology supplier and an innovative downstream company.

Competition principle

The case illustrates the importance of examining vertical relationships and the possibility of foreclosure of innovative rivals.

AI relevance

Comparable issues may arise when a dominant infrastructure provider acquires a downstream AI company that could become a significant competitor.

28. Case 9: Amazon Marketplace

Competition authorities in several jurisdictions have examined Amazon's treatment of sellers and the relationship between Amazon's marketplace and its own retail operations.

Competition principle

The central issues include:

• platform neutrality;
• use of commercially sensitive data;
• self-preferencing;
• conflicts between platform and downstream activities.

AI relevance

The same structural concern can arise where an AI platform:

hosts competing AI applications while simultaneously offering its own competing application.

29. Case 10: Apple App Store / Epic Games Litigation

The Apple–Epic litigation concerned Apple's control over app distribution and payment arrangements within its ecosystem.

Competition principle

The dispute illustrates the importance of:

• platform access;
• app distribution;
• payment restrictions;
• commission structures;
• technical rules.

AI relevance

If AI applications depend upon an app store or operating system for distribution, similar issues may arise concerning:

• AI assistants;
• AI agents;
• model marketplaces;
• AI payment systems.

30. Competition Risks Across the Synthetic Intelligence Value Chain

31. Competition Between AI Ecosystems

Competition may increasingly occur between ecosystems rather than individual firms.

For example:

Ecosystem A

Cloud + chips + foundation model + productivity software + AI assistant.

Ecosystem B

Open-source model + independent cloud + AI applications + interoperable APIs.

The relevant competition question becomes whether Ecosystem A can prevent users and developers from participating in Ecosystem B.

32. Multi-Homing

Multi-homing is particularly important.

Developers should ideally be able to use:

• several AI models;
• several cloud providers;
• several APIs;
• multiple AI marketplaces.

Where technical or contractual barriers prevent multi-homing, ecosystem power may become stronger.

33. Open-Source AI and Competition

Open-source models can potentially reduce entry barriers by allowing developers to access model technology without relying entirely on a proprietary provider.

However, open-source AI can also raise competition issues involving:

• access to compute;
• licensing restrictions;
• distribution;
• model hosting;
• cloud dependence.

Thus, open-source availability does not automatically eliminate ecosystem concentration.

34. Data Portability

Effective data portability may reduce switching costs.

Users and developers could potentially transfer:

• prompts;
• embeddings;
• training datasets;
• user histories;
• application configurations;
• agent preferences.

Competition authorities may therefore regard interoperability as an important competitive parameter in certain AI markets.

35. Merger Remedies

Possible merger remedies include:

Structural remedies

• divestiture;
• separation of business units;
• sale of assets.

Behavioural remedies

• interoperability;
• non-discrimination;
• API access;
• data portability;
• licensing commitments;
• prohibition of exclusivity.

Monitoring remedies

• independent compliance monitoring;
• reporting obligations;
• periodic review.

The appropriate remedy depends on the competitive harm established by the relevant authority.

36. Role of Consumer Welfare and Innovation

AI competition cannot be assessed solely through current prices.

Relevant competitive parameters may include:

• innovation;
• model quality;
• privacy;
• security;
• reliability;
• choice;
• speed of technological development;
• access to developers;
• interoperability.

An ecosystem may therefore produce substantial efficiencies while simultaneously creating foreclosure risks. Competition analysis must consider both.

37. Regulatory Challenges

Synthetic-intelligence ecosystems create several challenges for competition authorities.

1. Rapid technological change

Market boundaries may change faster than investigations.

2. Lack of transparency

AI models may be technically complex and difficult to audit.

3. Dynamic competition

A small company may rapidly become a significant competitor.

4. Compute scarcity

Access to AI chips and cloud capacity may become a major entry barrier.

5. Data concentration

Large ecosystems may possess unique datasets.

6. Ecosystem integration

Competition problems may arise from combinations of conduct rather than one isolated practice.

38. Possible Competition-Law Framework

A practical framework can be expressed as:

Identify ecosystem

↓

Map technological layers

↓

Define relevant markets

↓

Identify control points

↓

Assess market power

↓

Examine entry barriers

↓

Analyse exclusionary conduct

↓

Assess efficiencies and innovation

↓

Measure competitive foreclosure

↓

Consider remedies

39. Key Doctrinal Issues

The most important competition-law questions for synthetic intelligence ecosystems are:

1. Who controls the critical AI inputs?
2. Can rivals obtain equivalent compute?
3. Can developers switch models easily?
4. Can users multi-home?
5. Is interoperability available?
6. Does an ecosystem favour its own AI services?
7. Are competitors tied or bundled with complementary services?
8. Are exclusive arrangements preventing entry?
9. Can algorithms facilitate coordination?
10. Are acquisitions eliminating potential competitors?
11. Is access to data being strategically restricted?
12. Are AI infrastructure providers vertically leveraging their market power?

40. Conclusion

Competition law in synthetic-intelligence ecosystems requires a multi-layered ecosystem approach. The most significant competition risks are likely to arise where one undertaking controls several strategically important layers—such as compute, cloud infrastructure, data, foundation models, APIs, applications and distribution—and uses that position to restrict competitors.

The major doctrinal tools remain familiar:

• abuse of dominance;
• exclusionary conduct;
• tying and bundling;
• self-preferencing;
• refusal to deal;
• discriminatory access;
• exclusive dealing;
• predatory conduct;
• cartel and concerted-practice rules;
• merger control.

However, their application must account for AI-specific characteristics such as data advantages, compute dependency, network effects, interoperability, model switching costs, autonomous agents and rapid innovation.