1. Introduction

Cognitive infrastructure refers to the underlying technological infrastructure that enables computationally intensive intelligence and decision-making systems. In the modern digital economy, it can include:

• AI chips, GPUs and specialised accelerators;
• cloud-computing infrastructure;
• data centres and high-performance computing;
• AI training and inference infrastructure;
• foundational datasets and data-processing systems;
• model-development platforms;
• AI APIs and model-serving infrastructure;
• operating systems and developer environments;
• identity, payment and authentication infrastructure;
• networking and storage infrastructure; and
• specialised AI software stacks.

A cognitive infrastructure monopoly arises when one undertaking obtains substantial and durable market power over one or more of these indispensable inputs and uses that position to restrict competition in downstream markets.

Competition law therefore examines not merely whether a company is large, but how infrastructure control affects entry, interoperability, innovation, pricing, access, switching and downstream competition.

Recent enforcement developments demonstrate why this issue is becoming important. The U.S. FTC's study of major cloud-AI partnerships identified possible effects on access to computing resources, engineering talent, switching costs and access to commercially sensitive information. The UK CMA's cloud-services investigation likewise concluded in 2025 that competition concerns existed and recommended consideration of strategic-market-status investigations concerning Microsoft and AWS.

2. Meaning of Cognitive Infrastructure Monopoly

A conventional monopoly concerns control over a defined product or service market.

A cognitive infrastructure monopoly may instead involve control over a critical technological layer upon which several adjacent markets depend.

For example:

AI-chip infrastructure → cloud computing → foundation models → AI applications → downstream commercial services

If a company controls a critical upstream layer, it may have the ability to influence competition at several downstream levels.

Example

Suppose a cloud provider:

1. controls a large proportion of specialised AI computing capacity;
2. operates a foundation-model platform;
3. owns an AI application marketplace; and
4. imposes contractual restrictions preventing customers from efficiently moving AI workloads elsewhere.

The competition problem is not simply "large market share." It may involve vertical foreclosure, whereby infrastructure power is used to protect or extend dominance into related markets.

3. Relevant Competition-Law Concepts

A. Relevant Market Definition

The first question is whether the relevant infrastructure constitutes a distinct market.

Possible markets include:

• GPU/AI accelerator supply;
• AI cloud-computing services;
• model-training infrastructure;
• model-inference services;
• cloud storage;
• AI development platforms;
• AI API services;
• data-access services.

The market may be defined narrowly where customers cannot readily substitute conventional CPUs for specialised AI accelerators, or one cloud service for another.

However, market definition must remain evidence-based. Technological substitutability, switching costs, performance, price and customer behaviour are important.

4. Sources of Cognitive Infrastructure Market Power

4.1 Control of Scarce Computational Capacity

Advanced AI requires enormous computational resources.

A firm controlling scarce GPU or accelerator capacity may be able to:

• prioritise its own AI models;
• allocate capacity preferentially to affiliated businesses;
• impose restrictive contracts;
• increase prices to independent developers;
• restrict access during periods of shortage.

This creates a potential input foreclosure problem.

4.2 High Switching Costs

AI infrastructure can involve:

• proprietary APIs;
• specialised hardware;
• proprietary software;
• data-format dependencies;
• model-optimisation tools;
• contractual commitments;
• technical migration costs.

A customer that has trained or deployed models on one provider's infrastructure may face significant costs in migrating elsewhere.

The resulting lock-in can make existing market power more durable.

4.3 Vertical Integration

The same company may operate at multiple levels:

Hardware → Cloud → Foundation Model → AI Application → Marketplace

Vertical integration is not inherently unlawful.

The competition concern arises when an integrated undertaking uses upstream control to disadvantage competitors downstream.

5. Six Major Case Laws

1. United States v. Microsoft Corp.

253 F.3d 34 (D.C. Cir. 2001)

Facts

Microsoft possessed substantial power in the PC operating-system market. It used contractual and technological strategies involving computer manufacturers, Internet Explorer and software developers.

The government alleged that Microsoft had used its operating-system position to protect its dominance against emerging competitive threats.

Legal Principle

The case established important principles concerning:

• monopoly maintenance;
• exclusionary conduct;
• leveraging;
• technological tying;
• foreclosure of emerging competitors.

The court distinguished between conduct that reflects legitimate competition and conduct that unlawfully maintains monopoly power.

Relevance to Cognitive Infrastructure

The Microsoft principle is particularly relevant where an infrastructure provider controls a foundational technological layer.

For example:

AI cloud infrastructure → AI model platform → AI applications

If the infrastructure provider conditions access to the upstream infrastructure upon adoption of its downstream products, competition authorities may investigate whether the conduct protects monopoly power rather than merely producing efficiencies.

Key Lesson

Control over an essential technological platform cannot automatically justify exclusionary restrictions in adjacent markets.

2. United States v. Google LLC — Search and Search Advertising

U.S. District Court for the District of Columbia, 2024

Facts

The U.S. Department of Justice challenged Google's conduct concerning distribution agreements and mechanisms that helped maintain its position in general search.

The litigation examined how contractual arrangements and distribution advantages could reinforce durable market power.

Competition Principle

The case illustrates the importance of examining:

• network effects;
• default positions;
• distribution arrangements;
• barriers to entry;
• scale advantages;
• accumulation of data;
• reinforcement of market power.

Relevance to Cognitive Infrastructure

AI infrastructure can produce similar reinforcement mechanisms.

For example:

More customers → more computing demand → larger infrastructure → better optimisation → lower effective costs → more customers

A dominant infrastructure provider can therefore develop feedback loops that make entry increasingly difficult.

Key Lesson

Competition analysis must examine whether infrastructure advantages create a self-reinforcing competitive moat.

3. Google Android / Google Search (European Commission)

Case: Google Android

European Commission Decision, Case AT.40099 (2018)

Facts

The European Commission found that Google had engaged in several practices concerning Android, including arrangements involving:

• pre-installation;
• search applications;
• licensing conditions; and
• restrictions affecting competing mobile operating systems and applications.

Legal Principle

The case illustrates how dominance in one technological layer may be leveraged into neighbouring markets.

The Commission's analysis concerned:

• tying;
• contractual restrictions;
• foreclosure;
• leveraging;
• protection of dominant positions.

Relevance to Cognitive Infrastructure

Imagine a dominant AI-cloud provider requiring customers who purchase specialised AI infrastructure to use its:

• foundation model;
• AI API;
• identity system;
• marketplace;
• data-processing service.

Such arrangements may raise questions analogous to those examined in Android.

Key Lesson

Infrastructure dominance becomes more problematic when access to the dominant layer is used to reinforce dominance in complementary markets.

4. FTC v. Qualcomm Inc.

9th Cir. 2020

Facts

The litigation concerned Qualcomm's position in cellular modem-chip technology and licensing practices.

The FTC alleged that Qualcomm's licensing arrangements and business practices unlawfully maintained monopoly power.

The Ninth Circuit ultimately rejected the FTC's Sherman Act theory on the record before it.

Competition Principle

The case is particularly important because it demonstrates that possession of technological intellectual property or market power does not automatically establish an antitrust violation.

Competition authorities must demonstrate the relevant exclusionary conduct and its relationship to competition.

Relevance to Cognitive Infrastructure

AI-chip manufacturers may possess:

• patents;
• proprietary architectures;
• software ecosystems;
• compiler technology;
• optimisation libraries.

These advantages can legitimately reward innovation.

But competition analysis becomes relevant where licensing or contractual practices are allegedly designed to exclude competing infrastructure providers.

Key Lesson

A technological bottleneck alone does not establish unlawful monopolisation; the competitive effects of the challenged conduct must be demonstrated.

5. Intel v. European Commission

Case C-413/14 P, Intel Corporation Inc. v European Commission (2017)

Facts

The European Commission had imposed a major fine on Intel concerning rebates provided to computer manufacturers and a major retailer.

The Commission considered the rebates capable of restricting competition from AMD.

The Court of Justice subsequently required greater examination of whether the rebates were actually capable of producing exclusionary effects, particularly through the as-efficient-competitor (AEC) analysis where appropriate.

Relevance to Cognitive Infrastructure

Infrastructure providers could theoretically use:

• volume discounts;
• cloud credits;
• committed-spend discounts;
• preferential infrastructure pricing;
• bundled AI compute;
• long-term capacity agreements.

These arrangements may generate efficiencies.

But if such incentives effectively prevent customers from purchasing competing infrastructure, competition authorities may investigate whether they produce exclusionary effects.

Key Lesson

Discounts and rebates should be assessed according to their competitive effects rather than being treated automatically as unlawful or lawful.

6. Bronner v. Mediaprint

Case C-7/97, European Court of Justice (1998)

Facts

Bronner concerned access to a newspaper distribution system controlled by another newspaper group.

The claimant sought access to the established distribution infrastructure.

Legal Principle

The Court adopted a stringent test for imposing compulsory access to infrastructure under the abuse-of-dominance rules.

Among the relevant considerations were whether:

1. the facility was indispensable;
2. duplication was practically or economically impossible;
3. refusal would eliminate effective competition; and
4. there was no objective justification.

Relevance to Cognitive Infrastructure

This is highly relevant to:

• AI computing infrastructure;
• cloud infrastructure;
• specialised accelerators;
• data centres;
• AI model-serving platforms;
• proprietary AI APIs.

A competitor cannot necessarily demand access merely because another company controls an important infrastructure asset.

The indispensability threshold remains significant.

Key Lesson

Essential-facility theories require more than demonstrating that the infrastructure is useful or commercially attractive.

6. IMS Health v. NDC Health

Case C-418/01, IMS Health GmbH & Co. OHG v NDC Health (2004)

Facts

IMS Health controlled a data structure used by pharmaceutical companies.

A competitor sought access to the system.

Legal Principle

The Court developed stringent conditions concerning compulsory licensing/access to intellectual property where refusal could constitute an abuse of dominance.

The framework focused on factors including:

• indispensability;
• elimination of competition;
• absence of objective justification;
• emergence of a new product or service.

Relevance to Cognitive Infrastructure

The case has significant implications for:

• proprietary AI datasets;
• specialised data architectures;
• model-training resources;
• AI APIs;
• proprietary technical standards.

If an infrastructure input is genuinely indispensable, refusal to provide access can become a competition-law issue.

Key Lesson

Control over intellectual-property-based infrastructure does not automatically create an obligation to license, but exceptional circumstances can justify intervention.

7. United States v. Terminal Railroad Association

224 U.S. 383 (1912)

This classic U.S. Supreme Court case is important for the essential-facilities/access dimension of infrastructure competition.

Facts

A group controlled critical railroad terminal facilities in St. Louis.

Competitors faced difficulties obtaining comparable access.

Principle

The Supreme Court required arrangements that prevented the infrastructure from being used to exclude competitors.

Cognitive-Infrastructure Relevance

The analogy is particularly useful for:

• AI data centres;
• cloud interconnection;
• specialised computing clusters;
• infrastructure gateways;
• network infrastructure.

Where duplication is extraordinarily difficult and the infrastructure represents a bottleneck, exclusionary access practices can attract antitrust scrutiny.

8. How the Cases Apply Collectively

9. Main Forms of Cognitive Infrastructure Monopolisation

A. Refusal to Deal

A dominant infrastructure provider may refuse access to a critical:

• AI accelerator;
• cloud platform;
• dataset;
• API;
• model-serving system.

The legality depends heavily on whether the infrastructure satisfies the stringent conditions associated with essential-facility doctrine.

B. Self-Preferencing

A vertically integrated provider might rank or allocate infrastructure resources in favour of its own AI products.

Example:

Independent AI model → standard computing allocation
Affiliated AI model → preferential capacity allocation

The competition question is whether this conduct disadvantages equally efficient competitors.

C. Tying

A provider could make access to:

AI computing → conditional upon use of its AI model

or:

cloud infrastructure → conditional upon use of its proprietary AI security system.

Such conduct may raise tying concerns where the relevant market conditions and effects are established.

D. Exclusive Dealing

Long-term contracts may require customers to purchase most or all of their AI-computing requirements from one provider.

Exclusive purchasing can create:

• entry barriers;
• foreclosure;
• reduced contestability;
• capacity pre-emption.

The analysis must nevertheless consider contractual duration, market coverage, efficiencies and actual competitive effects.

10. Cloud-AI Partnerships and Infrastructure Concentration

The issue is particularly important because AI developers increasingly depend upon cloud providers.

The FTC's 2025 staff report examined Microsoft–OpenAI, Amazon–Anthropic and Google–Anthropic relationships. It identified potential concerns involving access to computing resources, switching costs, engineering talent and access to sensitive information.

This creates a possible infrastructure–model feedback loop:

Cloud provider

↓

Investment in AI developer

↓

AI developer commits cloud expenditure

↓

Cloud provider receives strategic information/control rights

↓

Competitors face greater difficulty obtaining equivalent infrastructure

↓

Infrastructure position becomes more entrenched

This does not, by itself, establish an antitrust violation. The legal question is whether particular agreements or conduct produce unlawful foreclosure or other harm to competition.

11. Switching Costs as a Competition Concern

Cloud and AI infrastructure can generate unusually high switching costs.

Technical switching costs

• APIs;
• proprietary model formats;
• hardware optimisation;
• software libraries;
• retraining;
• migration of datasets.

Commercial switching costs

• committed-spend contracts;
• termination fees;
• minimum-purchase obligations;
• volume discounts.

Organisational switching costs

• employee expertise;
• DevOps systems;
• security certification;
• compliance architecture.

The cumulative effect can make a nominally competitive market less contestable in practice.

The CMA's cloud investigation specifically examined issues including technical barriers, licensing practices, committed-spend arrangements and egress fees before concluding that competition concerns existed.

12. Data as Cognitive Infrastructure

Data may function as infrastructure where it is:

• difficult to reproduce;
• continuously updated;
• essential for training;
• protected by network effects;
• generated by a large installed user base.

A dominant firm may obtain a competitive advantage through a cycle:

Users

→ Data

→ Better models

→ Better products

→ More users

→ More data

This creates a potential data-network-effect barrier to entry.

However, data accumulation alone is not necessarily unlawful. Competition law must establish market power and exclusionary conduct.

13. AI Chips as Cognitive Infrastructure

Specialised AI accelerators may become bottleneck inputs because high-end AI models require:

• massive parallel processing;
• specialised memory;
• high-bandwidth interconnection;
• optimised software;
• specialised compilers.

A competition investigation may therefore examine:

1. market share;
2. technical substitutability;
3. capacity constraints;
4. interoperability;
5. switching costs;
6. software compatibility;
7. exclusive supply agreements;
8. rebates;
9. bundling;
10. licensing restrictions.

The central question is whether technological superiority results from legitimate innovation or is reinforced through exclusionary conduct.

14. Interoperability and Portability

Competition law may increasingly intersect with:

• cloud portability;
• AI-model portability;
• API interoperability;
• data portability;
• containerisation;
• open standards.

Interoperability can reduce switching costs and make infrastructure markets more contestable.

Possible remedies include:

• interoperability obligations;
• API access;
• data portability;
• prohibition of discriminatory access;
• reduction of contractual lock-in;
• transparency obligations.

But mandatory access must be carefully designed because excessive intervention can reduce incentives to invest in infrastructure.

15. Conglomerate Effects

Cognitive infrastructure monopolies can produce conglomerate leverage.

For example:

Cloud dominance
↓
AI model dominance
↓
AI application marketplace dominance

The same company can potentially use its position at one level to reinforce another.

Competition authorities therefore increasingly need to examine ecosystems rather than isolated products.

16. Competition Concerns in India

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

• Section 3 – anti-competitive agreements;
• Section 4 – abuse of dominant position;
• Section 5 – combinations;
• Section 19 – inquiry into agreements and dominance;
• Section 26 – investigation procedure;
• Section 27 – orders following findings of contravention.

Cognitive-infrastructure conduct could potentially raise Section 4 questions involving:

Section 4(2)(a)

Unfair or discriminatory conditions or prices.

Section 4(2)(b)

Limiting or restricting:

• production;
• technical development;
• markets.

Section 4(2)(c)

Denial of market access.

Section 4(2)(d)

Tying one condition to another unrelated obligation.

Section 4(2)(e)

Using dominance in one relevant market to enter into or protect another market.

The last provision is particularly relevant to vertically integrated AI ecosystems.

17. Possible Competition-Law Remedies

Where unlawful conduct is established, authorities could potentially consider:

Structural remedies

• divestiture;
• separation of infrastructure and downstream operations;
• restrictions on acquisitions.

Behavioural remedies

• non-discriminatory access;
• interoperability;
• prohibition of exclusivity;
• restrictions on tying;
• transparency requirements.

Contractual remedies

• limiting long-term exclusivity;
• reducing switching penalties;
• modifying committed-spend agreements;
• restricting discriminatory rebates.

Data remedies

• portability;
• interoperability;
• access under appropriate conditions.

18. Key Distinction: Monopoly vs Abuse of Monopoly

A critical examination point is:

Competition law generally does not condemn market power simply because a company possesses it.

A company may become dominant because it has:

• superior technology;
• better infrastructure;
• greater efficiency;
• successful innovation;
• economies of scale.

The competition-law problem arises when dominance is maintained or extended through exclusionary or exploitative conduct prohibited by the applicable legal system.

This distinction is particularly important for AI infrastructure because substantial economies of scale may be economically legitimate.

19. Emerging Regulatory Direction

The regulatory landscape is moving toward closer examination of cloud and AI infrastructure.

The European Commission has investigated whether AWS and Microsoft Azure should be treated as gatekeepers under the Digital Markets Act framework, with particular attention to their entrenched positions, switching costs, ecosystems and AI-related infrastructure.

Academic competition-law analysis similarly identifies control over computational resources and data as potential sources of AI entry barriers, particularly where infrastructure control permits vertical integration and foreclosure.

20. Important Legal Tests

A competition authority examining a cognitive-infrastructure monopoly would generally need to address:

Step 1 — Define the relevant market

What infrastructure product or service is being supplied?

Step 2 — Establish market power

Does the undertaking possess substantial and durable market power?

Step 3 — Identify the conduct

Is the issue:

• refusal to deal;
• tying;
• exclusive dealing;
• discriminatory access;
• predatory pricing;
• rebates;
• self-preferencing;
• interoperability restrictions;
• acquisition;
• licensing restriction?

Step 4 — Establish foreclosure

Does the conduct materially restrict competitors?

Step 5 — Examine effects

What is the effect on:

• prices;
• quality;
• innovation;
• consumer choice;
• entry;
• technical development?

Step 6 — Consider efficiencies

Could the conduct produce:

• security benefits;
• reliability;
• lower costs;
• better performance;
• innovation;
• economies of scale?

Step 7 — Select proportionate remedies

The remedy should address the identified competition problem without unnecessarily destroying legitimate infrastructure efficiencies.

21. Case-Law Synthesis

The combined lessons of the major cases can be stated as follows:

1. United States v Microsoft — technological dominance cannot automatically justify exclusionary strategies designed to protect monopoly power.
2. Google Android — dominance in one technological layer can potentially be leveraged into adjacent markets through tying and contractual restrictions.
3. Google Search — distribution advantages and ecosystem effects can reinforce durable platform power.
4. FTC v Qualcomm — technological and intellectual-property dominance alone does not establish an antitrust violation; exclusionary conduct must be demonstrated.
5. Intel — rebates and discounts must be assessed according to their capacity and effects in the relevant circumstances.
6. Bronner — compulsory access to infrastructure requires a demanding indispensability analysis.
7. IMS Health — proprietary information infrastructure may attract exceptional access obligations only when stringent conditions are satisfied.
8. Terminal Railroad — control of an infrastructure bottleneck can generate serious competition concerns where access is used to exclude rivals.

22. Conclusion

Cognitive infrastructure monopolies represent a new application of established competition-law principles rather than an entirely separate category of antitrust law.

The central competition issue is the ability of a firm controlling a critical technological input to use that control to foreclose competitors, increase switching costs, restrict access, leverage dominance into adjacent markets or reinforce an ecosystem position.

The most important legal concepts are therefore:

market definition + dominance + foreclosure + essential facilities + tying + exclusive dealing + vertical leveraging + interoperability + switching costs + innovation effects.

The key challenge for competition authorities is to distinguish legitimate technological economies of scale and innovation from strategies that transform control over cognitive infrastructure into durable exclusionary market power.