Competition Law and Global AI Administration Infrastructure Dominance

1. Introduction

Global AI administration infrastructure may be understood as the infrastructure through which AI systems are developed, deployed, accessed, monitored, governed and integrated into public and private decision-making. It includes:

• AI cloud-computing infrastructure;
• GPUs, AI accelerators and specialised chips;
• data centres and high-performance computing;
• foundation-model infrastructure;
• model hosting and inference platforms;
• AI APIs and interoperability layers;
• identity, authentication and access-management systems;
• AI safety, auditing and compliance infrastructure;
• government AI procurement and administrative platforms;
• data repositories and data-sharing infrastructure; and
• software ecosystems connecting AI models to applications.

Competition-law concerns arise when a firm controls a critical infrastructure layer and uses that position to extend market power into adjacent AI markets.

This is particularly significant because AI infrastructure is becoming vertically integrated: cloud providers may simultaneously provide computing resources, finance AI developers, develop foundation models and distribute AI applications. The FTC's investigation into Microsoft–OpenAI, Amazon–Anthropic and Google–Anthropic specifically examined whether these arrangements could affect access to computing resources, increase switching costs and provide cloud providers with competitively sensitive information.

2. Meaning of AI Administration Infrastructure Dominance

AI administration infrastructure dominance exists where an undertaking possesses substantial and durable market power over infrastructure necessary for other firms or public bodies to develop, deploy or administer AI systems.

A simplified structure is:

Chips → Data Centres → Cloud → Foundation Models → APIs → Applications → Administrative/Government Systems

Dominance at one level can potentially be leveraged into another.

For example:

Dominant cloud infrastructure → preferential access to computing → stronger foundation-model position → preferential AI distribution → greater application-level dependence.

The competition problem therefore extends beyond conventional market-share analysis.

3. Why AI Infrastructure Is Particularly Susceptible to Dominance

A. Extremely high capital requirements

AI infrastructure requires enormous expenditure on:

• GPUs;
• networking;
• data centres;
• electricity;
• cooling;
• specialised software;
• cloud capacity; and
• research personnel.

These requirements create substantial barriers to entry.

B. Scarcity of specialised computing resources

A shortage of advanced AI processors can give suppliers considerable bargaining power.

The FTC has previously identified specialised AI chips and computational resources as potential competition bottlenecks.

C. Economies of scale

Large providers can spread infrastructure costs over millions of users.

This may make it difficult for smaller competitors to achieve comparable unit economics.

D. Switching costs

AI customers may become dependent upon:

• proprietary APIs;
• cloud-specific machine-learning tools;
• proprietary databases;
• model-management systems;
• specialised hardware;
• data pipelines; and
• contractual commitments.

The FTC specifically identified increased switching costs as a potential competitive concern in major AI partnerships.

E. Network effects

More users generate:

• more data;
• more developers;
• more applications;
• more integrations; and
• greater ecosystem attractiveness.

This can reinforce an incumbent's position.

4. Relevant Competition-Law Framework

Different jurisdictions approach the problem through several overlapping doctrines.

A. Abuse of Dominance

A dominant AI infrastructure provider may violate competition law by engaging in:

• exclusionary pricing;
• discriminatory access;
• tying;
• bundling;
• refusal to supply;
• exclusive dealing;
• loyalty rebates;
• self-preferencing;
• interoperability restrictions; or
• discriminatory technical standards.

B. Essential-Facilities Principles

Where infrastructure is genuinely indispensable, competition authorities may consider whether the infrastructure provider must provide access to competitors.

However, mere importance is insufficient.

Typical considerations include:

1. indispensability;
2. absence of realistic alternatives;
3. elimination of effective competition;
4. technical feasibility of access; and
5. objective justification.

C. Vertical Foreclosure

AI infrastructure providers frequently operate at multiple levels.

For example:

Cloud provider + AI model developer + AI application

can create incentives to disadvantage independent AI developers.

The OECD has identified bundling and tying between cloud infrastructure and AI models as an important potential competition issue.

5. Six Important Case Laws

Case 1 — United Brands v Commission

Court of Justice of the European Union

Principle

United Brands established an important framework for determining dominance and examining whether a firm possesses the ability to behave independently of competitors and customers.

Relevance to AI

An AI infrastructure provider may potentially possess dominance where:

• customers have limited alternatives;
• infrastructure is difficult to replicate;
• switching is costly;
• supply is constrained; and
• the provider can impose commercial conditions without effective competitive constraint.

Application

For example, if an AI cloud infrastructure provider controls a sufficiently narrow market for specialised AI compute, conventional market-share analysis could be supplemented by examination of:

• customer dependency;
• capacity constraints;
• switching costs;
• alternative providers; and
• technological substitutability.

Lesson: Market power in AI infrastructure should be assessed through economic dependency as well as market share.

Case 2 — Commercial Solvents v Commission

Court of Justice of the European Union

Principle

Commercial Solvents concerned refusal by a vertically integrated undertaking to supply an input to downstream competitors.

The case is important for the principle that a dominant undertaking controlling an essential upstream input cannot necessarily use that position to eliminate competition downstream.

AI relevance

Consider:

GPU/cloud infrastructure → AI model developers

If an infrastructure provider supplies computing capacity to independent AI developers while simultaneously competing with them through its own AI models, discriminatory supply could become a competition issue.

Potential conduct includes:

• withholding capacity;
• discriminatory pricing;
• inferior service levels;
• preferential allocation;
• technical restrictions; and
• discriminatory access to new hardware.

Lesson

AI infrastructure dominance becomes especially sensitive where the infrastructure owner also competes with its infrastructure customers.

Case 3 — Magill / IMS Health

European Union

Principle

The IMS Health jurisprudence developed important principles concerning refusal to license or provide access to indispensable inputs.

The doctrine is particularly relevant where:

• an input is indispensable;
• refusal eliminates effective competition;
• access is necessary for a new product or service; and
• refusal lacks objective justification.

AI application

Potential examples include:

• essential AI datasets;
• model-access interfaces;
• interoperability information;
• critical AI APIs;
• infrastructure-management interfaces; and
• specialised administrative AI datasets.

A company should not automatically be required to disclose proprietary technology merely because competitors would benefit from access.

The indispensability threshold remains critical.

Lesson

Competition law must balance:

access to infrastructure

against

innovation incentives and intellectual-property protection.

Case 4 — Microsoft v Commission

European Commission / Court of Justice of the European Union

Principle

Microsoft's European competition litigation is highly relevant to AI infrastructure because it involved interoperability and leveraging dominance from one technological layer into another.

The case demonstrated that competition authorities may intervene where technical restrictions prevent competing products from effectively interoperating with a dominant platform.

AI application

Comparable problems could arise where a dominant AI infrastructure ecosystem restricts interoperability between:

• cloud services;
• AI models;
• AI agents;
• identity systems;
• data-management systems;
• application programming interfaces; and
• government AI systems.

For example, a public administration using one AI ecosystem could become effectively locked into that provider if migrating models, data and applications becomes technically prohibitive.

Lesson

Interoperability can be a competition parameter.

Case 5 — Intel v Commission

European Union

Principle

Intel concerned conditional rebates and the use of pricing arrangements by a dominant undertaking.

The case is particularly relevant after the CJEU's clarification that the competitive effects of rebates may need detailed economic examination rather than relying exclusively on formal classifications.

AI infrastructure application

An AI-chip or cloud provider could theoretically offer:

• volume rebates;
• capacity discounts;
• preferential pricing;
• infrastructure credits;
• bundled AI services; or
• conditional discounts tied to exclusivity.

A competition authority would need to investigate whether these arrangements actually foreclose equally efficient competitors.

Example

Suppose an AI cloud provider offers:

40% cheaper compute provided that a customer obtains all AI workloads exclusively from that provider.

The competitive analysis would consider:

• duration;
• coverage;
• foreclosure effects;
• alternatives;
• switching costs;
• scale;
• efficiencies; and
• actual or likely competitive effects.

Lesson

Low prices are not automatically pro-competitive if their structure creates exclusionary foreclosure.

Case 6 — Qualcomm

European Commission / EU competition jurisprudence

Principle

The Qualcomm litigation illustrates competition concerns surrounding dominant technology suppliers, pricing practices and exclusionary effects in technologically concentrated markets.

AI infrastructure application

The AI hardware market may produce similar issues where an infrastructure supplier has substantial market power over:

• accelerators;
• interconnect technology;
• specialised networking;
• AI memory;
• inference hardware; or
• related software ecosystems.

A dominant supplier could potentially use:

• exclusivity;
• rebates;
• conditional discounts;
• supply restrictions;
• contractual restrictions; or
• ecosystem compatibility

to disadvantage rival infrastructure technologies.

Lesson

Competition analysis should examine the entire technological ecosystem, not merely the individual hardware product.

6. Additional Highly Relevant Precedent — Bronner

CJEU

Bronner is particularly important to AI infrastructure because it establishes a demanding standard for mandatory access to infrastructure.

The central concern is whether the infrastructure is genuinely indispensable and whether there are realistic alternatives.

AI application

Suppose a government AI platform wants mandatory access to a particular cloud provider.

The fact that the provider is:

• cheaper;
• larger;
• more technologically advanced; or
• commercially attractive

would not automatically make it an essential facility.

A competition authority would need to determine whether alternative infrastructure realistically exists.

7. Emerging AI-Specific Competition Matters

AI-specific competition law is developing rapidly, so not every important AI matter is yet a final judicial precedent.

Microsoft–OpenAI

The FTC investigated the Microsoft–OpenAI relationship as part of its broader study of AI partnerships and investments. The investigation examined issues including computing resources, switching costs, governance/control rights and access to competitively sensitive information.

Amazon–Anthropic

The FTC's investigation similarly examined the Amazon–Anthropic partnership, particularly the relationship between cloud infrastructure and AI development.

Google–Anthropic

Google's investment and partnership with Anthropic was also examined in the FTC's Section 6(b) study.

These matters demonstrate an important development:

Competition authorities are increasingly examining AI infrastructure as an ecosystem rather than treating cloud, chips, models and applications as completely separate markets.

8. Cloud Infrastructure as the Central AI Bottleneck

Cloud infrastructure is particularly significant because many AI developers do not own sufficient computing infrastructure.

The major cloud providers therefore occupy an upstream position in the AI supply chain.

The EU has specifically identified cloud computing as critical to AI development and has investigated whether existing digital-market rules adequately address competition problems in cloud markets.

In June 2026, the European Commission announced a preliminary position that AWS and Microsoft Azure should be designated as DMA gatekeepers for cloud services, citing their scale, ecosystem effects, switching costs and importance as gateways. This was a preliminary position rather than a final adjudication.

9. AI Infrastructure and Vertical Integration

One of the most important competition questions is:

What happens when the infrastructure provider competes with the companies dependent upon that infrastructure?

Consider:

The greater the vertical integration, the greater the possibility of input foreclosure and customer foreclosure.

10. Self-Preferencing

A dominant AI infrastructure provider could potentially favour its own AI products through:

• faster computing allocation;
• preferential API access;
• superior technical integration;
• better placement;
• preferential pricing;
• exclusive features;
• privileged data access; or
• interoperability advantages.

The competition question is whether the conduct disadvantages rivals in a manner capable of weakening effective competition.

11. Data as AI Administrative Infrastructure

AI infrastructure is not merely physical.

Data can function as a critical competitive input.

Examples include:

• government administrative databases;
• health databases;
• geographic data;
• financial data;
• scientific datasets;
• public records;
• transaction data; and
• behavioural datasets.

Where one company controls a uniquely valuable dataset, competition concerns may involve:

• refusal of access;
• discriminatory access;
• exclusive licensing;
• tying;
• data portability restrictions;
• interoperability barriers; and
• data accumulation through acquisitions.

12. Government AI Infrastructure

The issue becomes particularly important when governments use private AI infrastructure.

Suppose a government administration relies upon a single provider for:

• identity verification;
• document processing;
• welfare administration;
• tax analytics;
• healthcare AI;
• public procurement;
• cybersecurity;
• judicial administration; or
• national AI computing.

Long-term dependence may create infrastructure lock-in.

Competition law may therefore intersect with:

• public procurement law;
• digital-market regulation;
• data protection;
• cybersecurity;
• national-security rules;
• intellectual-property law; and
• administrative law.

13. Essential-Facility Problem in Public AI

A particularly difficult question is whether an AI infrastructure provider should be required to provide access to government or private competitors.

A competition authority might examine:

Indispensability

Is there any realistic alternative?

Replicability

Can competitors build equivalent infrastructure?

Time

Would replication take months or decades?

Cost

Is duplication economically feasible?

Interoperability

Can the customer migrate to another provider?

Security

Would mandatory access compromise cybersecurity?

Innovation

Would compulsory access reduce incentives to invest?

14. Mergers and Acquisitions

AI infrastructure concentration can also arise through mergers.

Competition authorities may investigate acquisitions involving:

• GPU companies;
• cloud providers;
• AI startups;
• model developers;
• data providers;
• AI cybersecurity firms;
• AI middleware companies; and
• foundation-model developers.

The concern is not merely whether the acquired company is currently a competitor.

Authorities may also consider whether it represents:

• potential competition;
• an important future entrant;
• an innovation competitor; or
• an alternative infrastructure pathway.

15. Killer-Acquisition Concerns

An infrastructure incumbent may have incentives to acquire a promising AI startup before it becomes a meaningful competitor.

The competitive theory may involve:

Incumbent infrastructure + emerging AI firm

→ acquisition

→ removal of potential competitive constraint

→ greater ecosystem concentration.

This is especially significant where traditional turnover thresholds fail to capture the value of an AI startup.

16. Exclusive Cloud Partnerships

Long-term exclusive agreements may create competition concerns where AI developers are tied to one infrastructure provider.

Potential effects include:

• reduced multi-cloud adoption;
• higher switching costs;
• reduced access to rival cloud providers;
• foreclosure of competing AI infrastructure;
• information advantages for the incumbent; and
• increased ecosystem dependence.

However, exclusivity is not automatically unlawful. The actual competitive effects and commercial justification must be examined.

17. Interoperability as a Competition Remedy

Possible remedies include:

A. API interoperability

Allowing AI systems to communicate with competing infrastructure.

B. Data portability

Allowing customers to export:

• datasets;
• models;
• configurations;
• logs; and
• metadata.

C. Cloud switching

Reducing technical and contractual obstacles to migration.

D. Non-discrimination

Requiring infrastructure providers to treat competing AI developers on objectively equivalent terms.

E. Separation remedies

In extreme circumstances, structural separation between infrastructure and downstream AI operations may be considered.

18. Competition Risks by Infrastructure Layer

19. Global Regulatory Dimension

The problem is inherently cross-border.

An AI infrastructure provider may:

• develop models in the United States;
• manufacture chips in Asia;
• operate cloud infrastructure in Europe;
• process data in multiple jurisdictions; and
• supply governments globally.

Consequently, competition authorities may apply different legal approaches to the same conduct.

Important jurisdictions include:

United States

• Sherman Act;
• Clayton Act;
• FTC Act;
• DOJ Antitrust Division;
• Federal Trade Commission.

European Union

• Articles 101 and 102 TFEU;
• EU Merger Regulation;
• Digital Markets Act;
• sector-specific digital regulation.

United Kingdom

• Competition Act 1998;
• Enterprise Act 2002;
• Digital Markets, Competition and Consumers Act 2024.

China

• Anti-Monopoly Law;
• merger-control rules;
• digital-platform regulation.

Other jurisdictions

Competition authorities in Australia, Japan, India, Canada and South Korea are also relevant to multinational AI infrastructure markets.

20. The Three Major Theories of Harm

Theory 1 — Infrastructure foreclosure

A dominant infrastructure supplier prevents rivals from obtaining essential resources.

Example:

GPU/cloud capacity → preferentially allocated to affiliated AI model.

Theory 2 — Ecosystem leveraging

A company uses infrastructure dominance to expand into adjacent AI markets.

Example:

Cloud dominance → bundled foundation model → AI application dominance.

Theory 3 — Administrative lock-in

A government or major enterprise becomes dependent upon one AI infrastructure ecosystem.

Example:

Government data + proprietary APIs + proprietary models + proprietary cloud

→ extremely expensive migration.

This can create durable entry barriers even if the initial procurement process was competitive.

21. Defences Available to AI Infrastructure Providers

An infrastructure provider may argue:

Security

Restrictions are necessary to protect sensitive infrastructure.

Reliability

Preferential integration may improve system performance.

Investment incentives

Mandatory access could discourage infrastructure investment.

Efficiency

Bundling may reduce costs.

Innovation

Vertical integration can accelerate AI development.

Capacity constraints

Limited supply may objectively require allocation mechanisms.

These defences must be assessed against the actual competitive effects.

22. Key Case-Law Principles Applied to AI

23. Future Competition-Law Questions

The development of AI infrastructure will raise increasingly difficult questions:

1. Can a cloud provider simultaneously finance and compete with its AI customers?
2. When does cloud dependence become an essential-facility problem?
3. Should AI models be portable between clouds?
4. Can governments prohibit exclusive AI infrastructure arrangements?
5. Should AI compute capacity be treated as a strategic infrastructure resource?
6. Can a dominant AI-chip supplier impose ecosystem-wide compatibility conditions?
7. Should foundation models be subject to interoperability obligations?
8. How should competition authorities evaluate AI acquisitions below traditional merger thresholds?
9. Can access to government datasets constitute an essential competitive input?
10. When does AI safety collaboration become competitively restrictive?

24. Conclusion

Global AI administration infrastructure dominance represents a convergence of traditional infrastructure competition law and digital-platform antitrust.

The central competition problem is not simply that one company may have a large AI market share. The deeper concern is control over the infrastructure on which competing AI firms, applications and public administrations depend.

The most important legal concepts are:

• abuse of dominance;
• essential facilities;
• refusal to deal;
• vertical foreclosure;
• tying and bundling;
• discriminatory access;
• exclusive dealing;
• conditional rebates;
• interoperability;
• data access;
• switching costs;
• ecosystem leveraging; and
• merger control.

The existing jurisprudence of United Brands, Commercial Solvents, IMS Health, Bronner, Microsoft, Intel and Qualcomm provides the principal doctrinal foundations, while contemporary investigations into major cloud–AI partnerships demonstrate how these principles are being applied to the emerging AI ecosystem. The FTC's AI partnership study expressly identified computing resources, switching costs and competitively sensitive information as potential competition issues.

The regulatory trajectory is therefore moving from a narrow question of “Who controls the AI model?” toward the broader competition question:

“Who controls the infrastructure through which the global AI ecosystem can enter, operate, interoperate and compete?”

That shift makes AI compute, cloud infrastructure, data, APIs and administrative AI systems increasingly important subjects of global competition law.