Competition Law and Artificial Intelligence Dependency Risks

1. Introduction

Artificial Intelligence (AI) dependency risk arises when businesses, consumers, public authorities, or entire markets become substantially dependent upon a small number of AI providers, models, computing infrastructures, data ecosystems, cloud platforms, application programming interfaces (APIs), or distribution channels.

From a competition-law perspective, AI dependency becomes problematic where control over an indispensable or strategically important AI input enables an undertaking to:

exclude competitors;

raise rivals' costs;

foreclose downstream markets;

discriminate between customers;

impose exclusivity;

bundle AI with other products;

restrict interoperability;

deny access to essential data or computing resources;

engage in self-preferencing;

acquire emerging competitors; or

use contractual or technical restrictions to lock customers into its ecosystem.

AI dependency is therefore not itself unlawful. The competition-law issue is how market power is obtained or exercised and whether that conduct harms competition.

2. Why AI Creates New Dependency Risks

AI markets have several characteristics capable of producing dependency.

A. Concentration of computing resources

Training sophisticated foundation models can require enormous quantities of:

GPUs;

specialised processors;

cloud infrastructure;

electricity;

data-storage capacity;

networking infrastructure.

If only a limited number of firms can provide these inputs, AI developers may become dependent on them.

B. Concentration of foundation models

Downstream businesses may rely upon a small number of foundation-model providers for:

text generation;

reasoning;

image generation;

coding;

speech;

embeddings;

AI agents.

A downstream developer may therefore face substantial switching costs if its application is designed around one model provider.

C. Data dependency

AI systems require data for:

training;

fine-tuning;

evaluation;

retrieval;

personalisation.

Control over commercially valuable datasets can therefore create competitive advantages that are difficult for rivals to reproduce.

D. Cloud dependency

AI applications frequently depend on cloud platforms for:

model hosting;

inference;

storage;

networking;

security;

deployment.

Vertical integration between cloud providers and AI developers can therefore create competition concerns.

3. Relevant Competition-Law Theories

AI dependency can potentially engage several areas of competition law.

3.1 Abuse of Dominance

Where an AI provider possesses a dominant position, competition authorities may examine whether it abuses that position through:

refusal to supply;

discriminatory access;

tying;

bundling;

exclusivity;

loyalty rebates;

self-preferencing;

interoperability restrictions;

excessive contractual restrictions;

predatory pricing;

foreclosure of downstream competitors.

In the EU, Article 102 TFEU is particularly relevant.

4. Essential-Facility-Type Concerns

One of the most important questions is whether access to an AI input is indispensable.

Potential examples include:

highly specialised AI infrastructure;

unique datasets;

critical APIs;

dominant foundation models;

AI compute;

technical interfaces.

However, not every important AI resource is an essential facility.

Competition law generally requires stringent conditions before imposing a duty on a dominant undertaking to deal with competitors.

The classic European approach examines factors such as:

indispensability;

elimination of effective competition;

absence of objective justification;

whether access is genuinely necessary.

5. Tying and Bundling

AI dependency can arise through bundling.

For example, a dominant technology provider could potentially combine:

operating system + cloud + AI assistant + productivity software + AI model

and make access to one product conditional upon purchasing another.

The competition-law question would be whether the arrangement forecloses competitors in the tied market.

AI therefore creates a particularly important modern application of traditional tying doctrine.

6. Exclusive AI Arrangements

An AI provider could potentially require customers to:

use only its model;

use only its cloud infrastructure;

purchase minimum quantities;

avoid competing AI providers;

route all AI queries through its platform.

Such arrangements can produce dependency by making customers unable or unwilling to multi-home.

The competition analysis would consider:

duration;

market coverage;

market power;

switching costs;

availability of alternatives;

foreclosure effects;

efficiencies.

7. Switching Costs and Lock-In

AI dependency can also arise without an express exclusivity clause.

A customer may become locked into an AI provider because it has invested heavily in:

proprietary APIs;

prompts;

fine-tuning;

model-specific workflows;

vector databases;

agent architectures;

employee training;

proprietary integrations.

Switching providers may then involve substantial technical and financial costs.

This creates an important competition-law distinction:

technical dependency ≠ automatically unlawful conduct.

Competition law becomes relevant where a powerful undertaking creates or exploits such dependency in a manner capable of restricting competition.

8. Interoperability and API Restrictions

AI applications increasingly operate through APIs.

A dominant AI platform could potentially disadvantage rivals by:

withholding API access;

limiting functionality;

imposing discriminatory access terms;

restricting interoperability;

degrading compatibility;

preventing integration with competing services.

Such behaviour could raise refusal-to-deal or discriminatory-access concerns under dominance rules.

9. Self-Preferencing

AI platforms may operate simultaneously as:

infrastructure providers;

model providers;

application providers;

marketplaces.

This creates opportunities for self-preferencing.

For example, an AI platform could potentially:

rank its own AI application above competitors;

provide its own applications with superior API access;

use proprietary data unavailable to competitors;

give its own AI products preferential computing resources.

The competition question is whether such conduct gives the integrated undertaking an exclusionary advantage.

10. Vertical Foreclosure

AI dependency can occur through vertical integration.

Consider:

Cloud provider → AI infrastructure → foundation model → AI application → distribution platform

If one undertaking controls multiple layers, it could potentially disadvantage rivals at another level.

For example:

Cloud provider
↓
Own foundation model
↓
Own productivity application

The provider could theoretically use control over cloud infrastructure to disadvantage competing AI developers.

This is a classic vertical foreclosure problem adapted to AI markets.

11. Data Dependency

Data may function as an important competitive input.

A dominant AI company may control:

search data;

consumer interaction data;

advertising data;

shopping data;

location data;

professional information;

proprietary industrial data.

Competitors may face difficulty obtaining comparable datasets.

Competition law may therefore examine whether:

access to data is artificially restricted;

data is used discriminatorily;

competitors are denied interoperability;

exclusive data arrangements foreclose rivals.

12. Acquisitions and Killer Acquisitions

AI dependency can also emerge through mergers and acquisitions.

A dominant technology company might acquire:

promising AI startups;

model developers;

AI infrastructure providers;

specialised datasets;

AI application companies.

The acquisition of a small company may nevertheless have substantial competitive significance if the target represents an emerging competitive constraint.

Competition authorities may therefore examine:

potential competition;

innovation competition;

access to data;

access to talent;

interoperability;

vertical integration;

ecosystem effects.

13. Important Case Law

The following cases provide the principal doctrinal foundations for analysing AI dependency.

1. Bronner v Mediaprint

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

Principle

The CJEU established a strict framework for refusal-to-deal claims.

A facility is not automatically subject to mandatory access merely because competitors would benefit from access.

The facility must generally be indispensable, meaning that there is no actual or potential substitute.

AI relevance

The case is highly relevant to:

AI APIs;

specialised datasets;

model access;

AI compute;

cloud infrastructure.

A competitor would need to demonstrate more than inconvenience or higher costs.

14. Magill

Joined Cases C-241/91 P and C-242/91 P, RTE and ITP v Commission

Principle

The CJEU recognised circumstances in which refusal to license intellectual property could constitute abusive conduct.

The case is important because intellectual-property rights are not automatically immune from Article 102.

AI relevance

AI dependency can involve:

model weights;

training datasets;

proprietary AI interfaces;

copyrighted databases;

technical standards.

Where a dominant undertaking controls a critical intellectual-property resource, Magill provides part of the doctrinal framework for analysing exceptional compulsory-access situations.

15. IMS Health

Case C-418/01, IMS Health GmbH & Co OHG v NDC Health GmbH & Co KG

Principle

The Court developed the exceptional circumstances doctrine concerning refusal to license intellectual property.

The analysis included:

indispensability;

elimination of competition;

prevention of a new product for which consumer demand exists;

absence of objective justification.

AI relevance

An AI platform might control an exceptionally important proprietary resource.

However, IMS Health demonstrates that competition law does not impose routine compulsory licensing.

This is particularly important when considering demands that dominant AI companies provide competitors with access to models or datasets.

16. Microsoft v Commission

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

Principle

The General Court upheld important findings concerning:

refusal to provide interoperability information;

tying;

leveraging dominance from one market into another.

Microsoft's conduct demonstrated how control over an important technological interface can be used to disadvantage competing products.

AI relevance

The case is particularly valuable for analysing:

AI API access;

interoperability;

operating-system AI integration;

cloud/AI bundling;

AI assistants integrated into dominant platforms.

The technological environment has changed, but the underlying competition principles remain relevant.

17. Google Shopping

Case T-612/17, Google and Alphabet v Commission

Principle

The EU Courts examined Google's treatment of competing comparison-shopping services.

The case illustrates how a dominant platform can potentially use control over an important distribution infrastructure to favour its own downstream service.

AI relevance

The reasoning can inform analysis of AI platforms that simultaneously operate:

the infrastructure;

the AI model;

the application marketplace;

competing downstream services.

Potential concerns include:

self-preferencing;

discriminatory ranking;

preferential access;

leveraging.

18. Google Android

Case T-604/18, Google and Alphabet v Commission

Principle

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

The General Court examined how contractual restrictions imposed by a dominant ecosystem operator could protect and reinforce market power.

AI relevance

The case provides useful analogies for AI ecosystems involving:

AI assistants;

operating systems;

app stores;

default settings;

model distribution;

mandatory AI integration.

An AI provider controlling an ecosystem may potentially use contractual arrangements to make competing AI systems less accessible.

19. Intel

Case C-413/14 P, Intel Corporation Inc. v Commission

Principle

The CJEU clarified the assessment of exclusivity rebates by dominant undertakings.

The Court emphasised that where the dominant undertaking provides evidence that its conduct is not capable of restricting competition, the authority must examine the relevant circumstances and effects.

AI relevance

AI providers could potentially offer:

cloud discounts;

model-access rebates;

volume discounts;

preferential pricing;

infrastructure credits.

If such arrangements effectively encourage customers to avoid competing AI providers, Intel provides an important framework for analysis.

20. Qualcomm

Case T-235/18, Qualcomm v Commission

Principle

The case concerned exclusivity payments and the assessment of exclusionary effects.

Although the specific technological market involved chipsets rather than AI, the case is relevant to technology markets characterised by:

high R&D costs;

network effects;

technological dependency;

substantial switching costs.

AI relevance

Similar analytical issues can arise where an AI infrastructure provider provides economic incentives for customers to remain exclusively within its ecosystem.

21. Broadcom

Commission Decision AT.40608 – Broadcom

The European Commission examined contractual arrangements concerning chipset supply and exclusivity.

Principle

The case demonstrates the continuing competition-law concern with contractual restrictions that can reinforce dependence upon a dominant technology supplier.

AI relevance

The analogy is particularly relevant to:

AI accelerator supply;

cloud infrastructure;

model-hosting agreements;

long-term AI compute contracts.

AI infrastructure markets may involve similar dependency concerns where customers have limited alternatives.

22. Google Search (AdSense)

Case T-334/19, Google and Alphabet v Commission

The case concerned contractual restrictions imposed by Google in online advertising.

Principle

The case demonstrates the possibility of leveraging dominance through contractual arrangements that restrict competitors' access to important distribution opportunities.

AI relevance

Similar concerns could theoretically arise if an AI platform uses contracts to prevent customers from:

integrating rival AI models;

displaying competing AI services;

using competing AI APIs;

distributing competing AI assistants.

23. AI Dependency and Competition Law: Analytical Framework

A competition authority examining AI dependency would generally need to consider several stages.

Stage 1 — Define the relevant market

Possible markets include:

AI foundation models;

generative AI services;

AI inference;

AI cloud infrastructure;

GPU/AI accelerator supply;

AI development tools;

AI application marketplaces.

Market definition is fact-specific.

Stage 2 — Determine market power

Relevant indicators may include:

market share;

switching costs;

barriers to entry;

access to compute;

data advantages;

network effects;

ecosystem integration;

customer dependence.

Stage 3 — Identify the dependency

The authority must determine what customers actually depend upon.

For example:

"Customers depend on AI provider X."

is insufficiently precise.

The relevant dependency might instead be:

"Customers depend on Provider X's specialised inference API because migrating their existing applications would require substantial redevelopment."

Stage 4 — Identify the conduct

Possible conduct includes:

refusal to supply;

discrimination;

exclusivity;

tying;

bundling;

self-preferencing;

predatory pricing;

loyalty rebates;

interoperability restrictions.

Stage 5 — Assess foreclosure

The central question becomes whether the conduct can substantially reduce competitors' ability to compete.

Relevant factors include:

number of affected customers;

duration;

market coverage;

availability of alternatives;

switching costs;

entry barriers;

customer dependence.

Stage 6 — Examine objective justification

A dominant AI provider may have legitimate reasons for particular restrictions.

For example:

cybersecurity;

privacy;

intellectual-property protection;

technical reliability;

prevention of misuse;

capacity management.

Competition law therefore requires the justification and proportionality of the restriction to be examined rather than assuming that every access restriction is unlawful.

24. AI Dependency and Network Effects

AI markets can generate powerful network effects.

More users can produce:

more data;

more feedback;

more integrations;

more developers;

more applications;

more complementary services.

This can produce a feedback loop:

More users → more data → better AI → more developers → more applications → more users

If a dominant undertaking controls the ecosystem, these effects can reinforce its position.

However, network effects themselves are not unlawful. The competition issue arises where they are reinforced through exclusionary conduct.

25. AI Dependency and Multi-Homing

Competition is generally more resilient where customers can use multiple AI providers.

For example:

Company → Model A
Company → Model B
Company → Model C

is structurally different from:

Company → Model A only

Multi-homing can reduce dependency because customers retain credible alternatives.

Consequently, competition authorities may examine whether contractual or technical arrangements artificially prevent multi-homing.

26. AI Dependency and Cloud Computing

Cloud infrastructure deserves special attention.

A cloud provider may simultaneously provide:

computing resources;

AI development platforms;

foundation models;

AI applications.

This creates opportunities for vertical leveraging.

A competition concern could arise where the provider uses dominance in cloud infrastructure to disadvantage independent AI developers.

Relevant theories include:

tying;

bundling;

discriminatory access;

refusal to supply;

self-preferencing;

exclusivity.

27. AI Dependency and Labour/Talent

AI dependency can also concern specialised human resources.

A small number of firms may attract large proportions of:

AI researchers;

engineers;

model specialists;

safety researchers.

Acquisitions or restrictive employment arrangements can potentially affect competition for AI talent.

Competition law may therefore intersect with:

labour-market competition;

non-compete restrictions;

no-poach arrangements;

acquisitions of AI startups.

28. AI Dependency and Interoperability

Interoperability can reduce dependency.

Examples include:

portable model configurations;

standard APIs;

common data formats;

transferable embeddings;

interoperable agent protocols.

Where switching is technically feasible, dependency may decrease.

Conversely, proprietary interfaces can increase switching costs.

Competition law may therefore become concerned where a dominant undertaking deliberately limits interoperability to exclude competing systems.

29. Difference Between Legitimate Dependency and Anticompetitive Dependency

This distinction is fundamental.

Legitimate dependency

A company becomes dependent because:

a provider developed a superior technology;

customers voluntarily chose it;

alternatives remain available;

the provider does not impose exclusionary restrictions.

This is not automatically a competition violation.

Potentially problematic dependency

The situation becomes more concerning where:

the provider is dominant;

alternatives are effectively eliminated;

customers are contractually locked in;

competitors cannot obtain necessary access;

interoperability is deliberately restricted;

the dominant firm discriminates against rivals;

the conduct forecloses downstream competition.

30. Competition-Law Compliance Measures for AI Businesses

Companies operating AI systems should consider:

Contractual safeguards

Avoid unnecessarily broad:

exclusivity;

non-compete;

non-use;

minimum-purchase requirements.

Interoperability

Where commercially and technically feasible, support:

API portability;

data portability;

migration tools.

Information governance

Competitors should not exchange competitively sensitive information through AI systems.

M&A compliance

AI acquisitions should receive early competition-law review where they involve:

emerging competitors;

critical datasets;

foundation models;

AI infrastructure.

Distribution

AI platforms should review:

ranking;

recommendations;

defaults;

access terms;

marketplace rules.

31. Key Case-Law Principles at a Glance

32. Conclusion

AI dependency is fundamentally a competition-structure issue rather than an independent offence.

The emergence of powerful AI ecosystems can create dependency at several levels:

Compute → Cloud → Foundation Model → API → Application → Distribution

Competition law becomes particularly important when a firm controlling one layer uses that position to restrict competition at another layer.

The traditional jurisprudence remains highly relevant. Bronner, Magill and IMS Health provide the foundation for analysing access and indispensability; Microsoft provides important guidance on interoperability and tying; Google Shopping and Google Android illuminate platform leveraging and ecosystem restrictions; while Intel, Qualcomm and Broadcom demonstrate how exclusivity and contractual arrangements can reinforce dependency in technology markets.

The central legal question is therefore not simply whether businesses depend upon AI, but whether a firm with substantial market power creates, exploits, or reinforces that dependency through conduct capable of foreclosing effective competition without sufficient objective justification or efficiency justification.