Competition Law and Competition Policy in AI-Dominated Industries

Introduction

Artificial intelligence is increasingly becoming a general-purpose technology affecting search, cloud computing, semiconductors, software, advertising, finance, healthcare, manufacturing, transportation, retail and professional services. In an AI-dominated industry, competitive conditions may therefore depend not merely on the number of firms selling AI products, but on control over compute, chips, cloud infrastructure, data, foundation models, talent, distribution channels, application ecosystems and user interfaces.

Competition law must consequently address a vertically connected AI value chain:

Semiconductors → Cloud/Compute → Data → Foundation Models → AI APIs → Applications → Distribution → Consumers

The central competition-law question is whether control at one layer can be used to foreclose competitors at another layer.

Recent enforcement activity demonstrates that authorities are already examining AI partnerships, acquisitions, acqui-hiring, access restrictions and platform conduct. For example, the U.S. FTC examined Microsoft–OpenAI, Amazon–Anthropic and Google–Anthropic relationships, while the UK CMA examined Microsoft–OpenAI, Microsoft–Inflection, Amazon–Anthropic and other AI arrangements.

I. Meaning of an AI-Dominated Industry

An AI-dominated industry is one in which AI systems or AI-related infrastructure substantially determine:

1. production;
2. distribution;
3. pricing;
4. product differentiation;
5. access to consumers;
6. innovation;
7. allocation of resources; or
8. competitive entry.

Examples include:

• generative AI;
• AI search;
• AI-powered cloud computing;
• autonomous vehicles;
• AI chips;
• AI cybersecurity;
• AI healthcare;
• AI advertising;
• algorithmic financial services;
• AI robotics;
• AI-powered enterprise software;
• AI-enabled e-commerce.

The distinctive feature is that AI competitive advantage is cumulative. A firm possessing superior compute may train better models; better models attract users; more users generate data and revenue; revenue finances additional compute and talent.

This can create a feedback loop:

Compute → Model Quality → Users → Data/Revenue → Investment → More Compute → Stronger Model

Competition policy must determine when this feedback represents legitimate innovation and when it becomes an exclusionary mechanism.

II. Principal Competition-Law Problems

1. Concentration of Compute

Advanced AI requires enormous quantities of GPUs, specialized accelerators, networking equipment and data-centre capacity.

If a small number of firms control critical compute resources, competitors may face:

• higher costs;
• delayed access;
• capacity shortages;
• discriminatory access;
• contractual restrictions;
• dependency on a dominant cloud provider.

The NVIDIA/Run:ai transaction illustrates why competition authorities may examine control over complementary AI infrastructure. Run:ai developed software for scheduling workloads on GPU clusters, while NVIDIA supplied accelerated computing platforms. The European Commission ultimately cleared the acquisition under Article 6(1)(b), but the case illustrates the relevance of GPU infrastructure and software ecosystems to AI competition.

III. Data as a Competitive Input

AI models depend heavily on:

• training data;
• user interaction data;
• proprietary datasets;
• behavioural data;
• search queries;
• transaction data;
• scientific data;
• enterprise data.

A dominant digital platform may therefore possess an important data advantage.

Competition concerns can arise where a dominant firm:

• refuses access to essential data;
• combines datasets unavailable to competitors;
• imposes discriminatory data-access conditions;
• uses competitors' data to improve its own AI products;
• prevents portability;
• imposes contractual restrictions on data sharing.

Data therefore becomes comparable to a strategic input or infrastructure asset.

IV. Foundation-Model Dominance

Foundation models can serve as an intermediate layer between infrastructure and applications.

A dominant foundation-model provider may potentially control:

Model → API → Developers → Applications → Consumers

Possible exclusionary strategies include:

• discriminatory API pricing;
• exclusive distribution;
• tying cloud services to AI models;
• technical interoperability restrictions;
• preferential access for affiliated applications;
• restrictions on model portability;
• preferential treatment of the firm's own applications.

The UK CMA's foundation-model work specifically identified the importance of maintaining competition both among AI developers and in downstream deployment of foundation models.

V. Cloud–AI Vertical Integration

One of the most important structural issues is the combination of:

Cloud provider + AI developer + AI model + enterprise distribution.

For example, a cloud provider may simultaneously:

1. supply computing capacity;
2. finance an AI developer;
3. obtain commercial or governance rights;
4. distribute the developer's model;
5. operate competing AI models;
6. possess information concerning the AI developer.

The FTC's 2025 study of major AI partnerships identified potential effects involving compute access, engineering talent, switching costs and access to commercially sensitive information.

This makes traditional vertical-integration analysis particularly important.

VI. Six Major Case Laws / Enforcement Cases

Note: Because AI competition law is a rapidly developing field, several of the most important AI-specific precedents are administrative merger investigations or antitrust proceedings rather than final judicial judgments. They are nevertheless highly relevant to competition-law analysis.

1. Microsoft Corporation / OpenAI — UK CMA

Background

Microsoft entered into a multi-billion-dollar relationship with OpenAI involving investment, AI research and infrastructure collaboration. Microsoft also became OpenAI's exclusive cloud provider under the relevant arrangements.

The CMA examined whether the relationship could constitute a relevant merger situation under the Enterprise Act 2002 and whether it could substantially lessen competition.

The CMA ultimately closed the matter in March 2025 after determining that the partnership did not qualify for investigation under the UK's merger provisions.

Competition Issues

The case is important because it demonstrates that competition authorities cannot examine AI investments solely as ordinary financial investments.

Relevant questions include:

• Does an investor obtain material influence?
• Does the partnership create de facto control?
• Is there exclusive cloud supply?
• Can competitors access equivalent compute?
• Does the partnership increase switching costs?
• Can the cloud provider disadvantage rival AI developers?

Legal Principle

AI partnerships may have merger-control significance even without a conventional acquisition of shares or assets.

2. Microsoft / Inflection AI — UK CMA

This is especially important for AI talent and acqui-hiring.

Microsoft hired significant numbers of employees from Inflection AI and entered associated arrangements concerning Inflection's technology.

The CMA investigated whether these arrangements amounted to a relevant merger situation. It ultimately cleared the transaction at Phase 1.

Competition Significance

AI companies are unusually dependent upon:

• machine-learning researchers;
• engineers;
• model architects;
• specialised technical teams;
• accumulated know-how.

Consequently, acquiring a team rather than the corporate entity can potentially have an economic effect similar to acquiring an AI business.

The CMA's analysis recognized the economic continuity associated with the transfer of Inflection's AI capabilities.

Principle

Competition law may need to examine acqui-hiring as a possible means of acquiring AI capabilities without formally purchasing the company.

This has major implications for merger-control thresholds.

3. NVIDIA / Run:ai — European Commission

Facts

NVIDIA proposed to acquire Run:ai, an Israeli company providing software that schedules AI workloads across GPU clusters.

The transaction was referred to the European Commission under Article 22 of the EU Merger Regulation.

The Commission ultimately approved the transaction without opposition under Article 6(1)(b).

Competition Significance

The case illustrates the increasing importance of AI infrastructure markets.

Competition is not limited to foundation models.

It can exist at:

• GPU level;
• accelerator level;
• networking;
• workload-management software;
• cloud infrastructure;
• AI orchestration;
• model deployment.

Principle

Competition authorities should examine complementary AI infrastructure because control over one technological layer may affect competitive conditions at another.

4. FTC AI Partnerships and Investments Study — Microsoft/OpenAI, Amazon/Anthropic and Google/Anthropic

In 2024, the U.S. FTC issued compulsory orders to Alphabet, Amazon, Anthropic, Microsoft and OpenAI concerning their AI partnerships and investments.

The subsequent FTC staff report examined partnerships involving:

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

The FTC identified potential competition concerns concerning:

• access to computing resources;
• access to engineering talent;
• switching costs;
• exclusivity;
• governance rights;
• access to sensitive technical and business information. 

Competition Principle

The case demonstrates the importance of analysing minority investments and strategic partnerships, not merely traditional mergers.

AI markets may produce competitive harm through:

Investment + infrastructure + exclusivity + information rights + distribution

even where formal ownership remains separate.

5. Meta AI / WhatsApp — European Commission

This is one of the most directly relevant contemporary AI competition proceedings.

Meta changed the terms governing access to WhatsApp in a manner that excluded third-party general-purpose AI assistants while allowing Meta AI to remain available.

The European Commission opened an Article 102 TFEU investigation concerning the exclusion of AI competitors from WhatsApp. By June 2026, the Commission had adopted interim measures requiring restoration and maintenance of free access for third-party general-purpose AI assistants pending the final outcome.

Competition Issues

This raises classic essential-access and self-preferencing questions:

• Is WhatsApp an important distribution channel for AI assistants?
• Does Meta have dominance in the relevant market?
• Is restricting third-party AI access exclusionary?
• Does preferential treatment of Meta AI disadvantage competing assistants?
• Can interoperability restrictions protect an adjacent AI market?

Principle

A dominant digital platform may not necessarily be permitted to use control over a major distribution ecosystem to exclude competing AI services from reaching users.

The case also demonstrates the importance of interoperability as a competition remedy.

6. Google AI and Data-Related Practices — European Commission

The European Commission's competition case register identifies AT.40983 – Google AI and Data-related practices, with a decision dated 9 December 2025.

The proceeding demonstrates the expansion of competition scrutiny toward the relationship between:

• AI;
• data;
• digital platforms;
• downstream services.

Competition Significance

Google's position across multiple digital markets potentially gives it access to large quantities of data and important distribution channels.

The competition-law questions in such cases include:

1. whether data constitutes a strategically important input;
2. whether data advantages can reinforce dominance;
3. whether AI services receive preferential access;
4. whether rivals can obtain commercially relevant data;
5. whether AI development reinforces an existing digital monopoly.

Principle

Competition policy increasingly has to evaluate data accumulation and AI capability together, rather than treating them as independent markets.

VII. Important Traditional Case Laws Applicable to AI

AI-specific precedents are still developing. Therefore, established competition-law jurisprudence remains extremely important.

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

The Microsoft case concerned exclusionary conduct involving the operating-system monopoly and browser competition.

Relevance to AI

The case provides a framework for analysing:

• technological tying;
• platform leverage;
• exclusionary interoperability decisions;
• protection of an emerging technology;
• conduct designed to prevent competitive threats.

An AI platform with a dominant position could potentially use its control over an established ecosystem to disadvantage emerging AI competitors.

8. Google Search — United States

The modern Google search litigation is important because it examines how a dominant digital platform can maintain market power through distribution arrangements and default positioning.

AI relevance

The same analytical concerns can arise where an AI provider controls:

• search interfaces;
• browsers;
• operating systems;
• mobile assistants;
• default AI applications.

The central question is whether default placement or distribution arrangements foreclose rival AI systems.

9. Google Shopping — European Commission

The EU Google Shopping decision established the importance of analysing self-preferencing by a dominant platform.

AI relevance

An AI platform might simultaneously operate:

• a general AI assistant;
• a marketplace;
• search;
• advertising;
• cloud services;
• third-party AI applications.

It may therefore have incentives to rank its own AI service more favourably.

The traditional self-preferencing analysis provides an important foundation for future AI cases.

VIII. AI and Abuse of Dominance

Traditional Article 102 TFEU / equivalent national provisions can apply to:

A. Refusal to supply

A dominant AI infrastructure provider may refuse access to:

• compute;
• APIs;
• data;
• model interfaces;
• distribution channels.

B. Discriminatory access

Different AI developers could receive different:

• prices;
• processing capacity;
• API limits;
• latency;
• technical functionality.

C. Tying

Examples:

Cloud service + proprietary AI model

or

Operating system + mandatory AI assistant.

D. Bundling

A dominant cloud provider could bundle:

• storage;
• compute;
• AI models;
• cybersecurity;
• enterprise software.

E. Self-preferencing

A platform may rank its own AI assistant above competing assistants.

F. Exclusive dealing

A cloud provider could potentially require an AI developer to use only its infrastructure.

G. Margin squeeze

A vertically integrated AI firm could theoretically charge rivals high upstream prices while competing downstream at artificially low prices.

IX. AI and Cartel Law

AI introduces new forms of coordination.

1. Algorithmic Pricing

If competing firms use similar AI pricing systems, algorithms could potentially facilitate:

• price coordination;
• rapid retaliation;
• monitoring;
• parallel price adjustments.

The legal challenge is determining whether coordination is:

independent algorithmic behaviour or human-mediated concerted practice.

2. Algorithmic Collusion

AI systems can potentially:

• observe competitors;
• predict responses;
• optimise prices;
• communicate through market signals.

Competition policy should therefore distinguish:

Legitimate parallel conduct

from

Agreement or concerted practice.

An algorithm itself should not automatically transform lawful independent conduct into a cartel.

X. AI and Merger Control

Traditional merger thresholds may be inadequate for AI because startups frequently have:

• low turnover;
• valuable intellectual property;
• highly skilled employees;
• proprietary datasets;
• important algorithms;
• strategic technological potential.

Therefore, an AI startup may be competitively important despite having relatively little current revenue.

Relevant transaction types

Competition authorities should examine:

1. acquisitions;
2. minority investments;
3. joint ventures;
4. licensing arrangements;
5. exclusive cloud contracts;
6. acqui-hiring;
7. long-term compute agreements;
8. strategic partnerships.

The Microsoft/Inflection inquiry illustrates why acquisition of people and know-how can raise merger-control questions.

XI. Killer Acquisitions in AI

A dominant technology company may acquire a promising AI startup before the startup becomes a significant competitor.

Possible theories include:

Potential competition theory

and

Innovation competition theory

Authorities should therefore consider:

• pipeline products;
• research capabilities;
• engineering talent;
• patents;
• proprietary datasets;
• model architecture;
• customer relationships;
• likelihood of independent expansion.

This is particularly important because AI markets can change rapidly.

XII. Access to AI Chips

AI competition can be affected by concentration in accelerator hardware.

Important inputs include:

• GPUs;
• AI accelerators;
• high-bandwidth memory;
• networking;
• interconnect technology;
• data-centre capacity.

A competition authority could examine whether a dominant chip supplier:

• refuses supply;
• discriminates between customers;
• bundles hardware and software;
• restricts compatibility;
• imposes exclusivity;
• prevents competing accelerator ecosystems.

The NVIDIA/Run:ai case illustrates the increasing competition-policy relevance of the infrastructure surrounding GPUs.

XIII. AI and Cloud Competition

Cloud computing can function as an essential competitive input for advanced AI.

Potential problems include:

1. Cloud lock-in

AI developers may face substantial costs when moving:

Model + data + APIs + applications

from one cloud provider to another.

2. Exclusive arrangements

Long-term agreements may prevent AI developers from using rival infrastructure.

3. Preferential treatment

A cloud provider developing its own AI models may have incentives to disadvantage competing models.

4. Access to information

A cloud provider may obtain valuable information about:

• model performance;
• customers;
• workloads;
• pricing;
• technical development.

The FTC specifically identified access to sensitive technical and business information as one potential competitive issue in major AI partnerships.

XIV. AI and Labour-Market Competition

AI competition also has a labour-market dimension.

Highly specialised AI researchers are scarce.

Competition authorities may therefore examine:

• no-poach agreements;
• wage coordination;
• restrictive employment agreements;
• acquisition of entire technical teams;
• exclusionary talent practices.

The Microsoft/Inflection matter demonstrates how the movement of AI personnel can intersect with merger control.

XV. AI and Consumer Choice

Competition law should protect not merely low prices but also:

• innovation;
• quality;
• privacy;
• security;
• interoperability;
• model diversity;
• consumer choice.

In AI markets, the "price" of a service may be zero while users provide:

• personal data;
• behavioural data;
• feedback;
• model-training information.

Therefore:

Zero monetary price ≠ absence of competition concerns.

XVI. AI and Innovation Competition

Innovation is particularly important because AI markets can change quickly.

Competition policy should ask:

• Will the conduct reduce future innovation?
• Will competitors continue developing alternative models?
• Will startups have access to compute?
• Will independent research survive?
• Will dominant firms acquire emerging competitors?
• Will interoperability be maintained?

A market with several firms today can still become less competitive if a dominant ecosystem controls the most promising future technologies.

XVII. Competition Policy Framework for AI-Dominated Industries

A comprehensive competition policy should contain the following components.

1. Infrastructure access

Ensure competitive access to:

• compute;
• cloud;
• chips;
• data centres;
• networking.

2. Data portability

Users and businesses should be able to move relevant data between competing services where legally appropriate.

3. Interoperability

AI ecosystems should avoid unnecessary technical barriers to competing models and applications.

4. Merger scrutiny

Review:

• minority investments;
• acquisitions;
• acqui-hiring;
• strategic partnerships;
• joint ventures.

5. Ecosystem neutrality

Dominant platforms should not automatically favour their own AI services.

6. Transparency

Competition authorities need sufficient information concerning:

• AI pricing;
• APIs;
• contractual restrictions;
• access conditions;
• technical interoperability.

7. Algorithmic compliance

Businesses should implement competition-law controls around:

• automated pricing;
• bidding algorithms;
• recommendation systems;
• competitor-data use.

XVIII. Proposed Legal Test for AI Dominance

A useful analytical framework is:

Step 1 — Define the market

Identify whether the relevant market concerns:

• AI models;
• AI applications;
• cloud compute;
• GPUs;
• AI APIs;
• AI distribution;
• data;
• downstream services.

Step 2 — Identify strategic inputs

Determine whether the firm controls:

• data;
• compute;
• talent;
• models;
• APIs;
• distribution.

Step 3 — Determine market power

Consider:

• market shares;
• switching costs;
• network effects;
• economies of scale;
• data advantages;
• entry barriers.

Step 4 — Identify conduct

Examine:

• tying;
• bundling;
• exclusivity;
• refusal to supply;
• discrimination;
• self-preferencing;
• predatory conduct;
• acquisitions.

Step 5 — Determine foreclosure

Ask:

Can the conduct materially reduce rivals' ability or incentive to compete?

Step 6 — Assess efficiencies

Possible legitimate justifications include:

• security;
• privacy;
• technical efficiency;
• interoperability protection;
• safety;
• innovation.

Step 7 — Design remedy

Potential remedies include:

• access obligations;
• interoperability;
• non-discrimination;
• data portability;
• behavioural commitments;
• divestiture;
• licensing;
• prohibition of exclusivity.

XIX. Indian Competition-Law Perspective

In India, AI competition would principally engage the Competition Act, 2002, particularly:

• Section 3 — anti-competitive agreements;
• Section 4 — abuse of dominant position;
• Section 5 — combinations;
• Section 6 — regulation of combinations;
• Sections 19–27 — investigation and enforcement framework.

Potential Indian AI markets include:

• AI cloud services;
• AI-enabled fintech;
• AI healthcare;
• AI search;
• generative AI;
• AI-enabled e-commerce;
• autonomous mobility;
• AI semiconductor supply;
• enterprise AI software.

Section 4 could become particularly important where a dominant digital platform controls an important AI input, interface or distribution channel.

XX. Challenges for Competition Authorities

1. Rapid technological change

Market definition can become obsolete quickly.

2. Difficult valuation

AI startups may have enormous competitive significance despite low revenues.

3. Intangible assets

Traditional turnover tests may undervalue:

• data;
• algorithms;
• models;
• talent.

4. Multi-sided markets

An AI platform may simultaneously serve:

• consumers;
• developers;
• advertisers;
• enterprises.

5. Dynamic competition

A firm that appears dominant today may face substantial innovation tomorrow.

6. Technical complexity

Authorities require expertise in:

• machine learning;
• cloud architecture;
• semiconductor technology;
• data economics;
• algorithms.

XXI. Key Principles Emerging from the Cases

Conclusion

Competition law in AI-dominated industries is moving from a traditional firm-versus-firm model toward an ecosystem-based model.

The relevant competitive question is increasingly not simply:

“How large is the AI company?”

but:

“Which strategic layers of the AI ecosystem does the company control, and can that control be used to restrict competition at another layer?”

The Microsoft–OpenAI, Microsoft–Inflection, NVIDIA–Run:ai, FTC AI partnerships, Meta–WhatsApp and Google AI/data matters demonstrate different dimensions of this transformation.

The emerging competition-policy framework therefore needs to combine antitrust, merger control, interoperability, access regulation, data portability, infrastructure competition and innovation policy.

The ultimate objective is not to prevent successful AI firms from becoming large. Rather, competition law must distinguish competitive success based on innovation from the use of market power to prevent rivals from obtaining the inputs, distribution, talent, data or technological opportunities necessary to compete.