Competition Law and Artificial Intelligence Market Entry Barriers

1. Introduction

Artificial Intelligence (AI) market entry barriers are the economic, technological, legal, financial, data-related and strategic obstacles that make it difficult for new or smaller undertakings to enter AI markets and compete effectively with established firms.

AI markets can exhibit unusually strong entry barriers because successful AI development may require simultaneous access to:

enormous quantities of data;

advanced computing infrastructure;

GPUs and specialized AI chips;

cloud-computing capacity;

foundation models;

highly skilled researchers and engineers;

proprietary software ecosystems;

distribution channels;

application programming interfaces (APIs);

capital-intensive research and development;

technical standards; and

established user networks.

Competition law becomes relevant where an incumbent with substantial market power creates, maintains or strengthens barriers to entry through exclusionary conduct, rather than merely benefiting from legitimate technological advantages.

2. Meaning of AI Market Entry Barriers

An entry barrier is a factor that makes entry more difficult, costly or risky for potential competitors than it is for an established undertaking.

In AI markets, barriers can be divided into several categories.

A. Technological barriers

Advanced AI systems may require:

large-scale computing;

specialized hardware;

distributed data centres;

sophisticated model architectures;

expensive training infrastructure; and

continuous model development.

These requirements can make entry exceptionally expensive.

B. Data barriers

AI systems may improve through access to large and diverse datasets.

An incumbent may therefore possess an advantage because it has accumulated:

search data;

consumer interaction data;

transaction data;

behavioural information;

proprietary enterprise data; and

user-generated content.

C. Computing barriers

Training advanced foundation models can require enormous computing resources.

If access to appropriate computing capacity is concentrated among a small number of cloud or infrastructure providers, competitors may become dependent upon those providers.

D. Human-capital barriers

Highly specialized AI researchers, engineers and technical personnel may be scarce.

A large undertaking may attract or acquire substantial portions of available talent, potentially increasing competitors' costs.

E. Distribution barriers

Even where a competitor successfully develops an AI model, reaching customers can be difficult if incumbent platforms control:

operating systems;

app stores;

search engines;

enterprise software;

cloud platforms;

browsers;

advertising networks; or

digital assistants.

3. Competition-Law Significance

The existence of a high entry barrier does not itself constitute an antitrust violation.

Competition law generally asks whether:

an undertaking possesses substantial market power or dominance;

the market contains significant barriers to entry;

the undertaking has engaged in conduct capable of excluding competitors;

the conduct lacks adequate legitimate justification; and

competition or consumers are likely to suffer harm.

Thus, a distinction must be maintained between:

Natural or technological entry barriers

and

Strategically created exclusionary barriers.

For example, the cost of developing a sophisticated AI model may be a legitimate technological barrier. However, deliberately denying competitors access to an indispensable input while using dominance to prevent effective competition may raise abuse-of-dominance concerns.

4. Relevant Markets in AI

AI cannot necessarily be treated as one single relevant market.

Potentially distinct markets may include:

AI chips;

cloud computing;

AI training infrastructure;

foundation models;

generative AI services;

AI APIs;

AI-powered search;

AI assistants;

enterprise AI software;

AI cybersecurity;

AI data services;

AI application distribution.

The relevant market must be determined according to substitutability, competitive conditions and the particular circumstances of the case.

5. Economies of Scale

AI development can produce significant economies of scale.

A larger firm may spread enormous fixed costs across millions of users.

For example:

High fixed AI-development costs + enormous user base = lower average cost per user

A new entrant with only a small customer base may therefore face substantially higher average costs.

This may discourage entry even without deliberate exclusionary conduct.

6. Economies of Scope

Large technology firms may also have economies of scope.

An incumbent may combine:

search;

advertising;

cloud computing;

operating systems;

consumer devices;

productivity software;

social networks;

enterprise software; and

AI.

AI can therefore be integrated into an existing ecosystem.

A new AI company may need to establish an entire distribution ecosystem from scratch.

7. Network Effects

Network effects can strengthen AI entry barriers.

Consider an AI assistant:

More users → more interactions → more data/feedback → improved service → more users.

A successful AI platform can therefore become increasingly difficult to challenge.

However, network effects are not inherently anticompetitive.

The legal issue arises where a dominant undertaking uses network effects together with exclusionary conduct to prevent rivals from attaining sufficient scale.

8. Data Advantages

Data can create a competitive feedback loop:

Users → data → improved AI → better service → more users → more data.

A dominant undertaking may therefore possess an important competitive advantage.

Competition-law questions can arise where the incumbent:

restricts competitors' access to essential data;

combines datasets in an exclusionary manner;

prevents data portability;

uses customer data from dependent firms to compete against them; or

acquires potential competitors primarily to eliminate emerging competitive threats.

Again, possession of valuable data alone is not necessarily abusive.

9. Cloud-Computing Dependence

Advanced AI developers often require substantial cloud resources.

This creates a potential vertical relationship:

Cloud provider → AI developer → AI application

If the cloud provider also develops competing AI systems, potential concerns may include:

discriminatory access;

preferential pricing;

capacity restrictions;

tying;

exclusive arrangements;

discriminatory technical performance; and

use of commercially sensitive information.

The competition analysis must distinguish ordinary vertical integration from conduct capable of excluding competing AI developers.

10. GPU and AI-Chip Bottlenecks

Specialized AI computing hardware can constitute another potential entry barrier.

If access to suitable GPUs or AI accelerators is constrained, new AI companies may face:

higher training costs;

delayed model development;

reduced capacity;

difficulty scaling;

increased dependence upon cloud providers.

Where a firm possesses substantial market power in a critical upstream input, discriminatory or exclusionary access practices may potentially attract competition-law scrutiny.

11. Intellectual Property

AI companies may rely on:

patents;

copyrights;

trade secrets;

model weights;

proprietary datasets;

software;

algorithms; and

specialized architectures.

Intellectual-property protection legitimately rewards innovation.

Competition concerns may nevertheless arise if intellectual-property rights are used strategically by a dominant undertaking to exclude competitors beyond what is reasonably necessary to protect legitimate innovation.

12. Exclusive AI Partnerships

An important modern issue is exclusive cooperation between:

AI developers and cloud providers;

AI companies and hardware manufacturers;

AI companies and operating-system providers;

AI companies and major distributors.

An exclusive arrangement can produce legitimate efficiencies.

But where a dominant undertaking uses exclusivity to deny rivals access to an indispensable distribution or infrastructure channel, it may create foreclosure concerns.

13. Acquisitions and Killer Acquisitions

AI markets may also experience acquisitions of promising startups by established technology companies.

A transaction can potentially eliminate an emerging competitive constraint before the startup becomes a significant rival.

Competition authorities may therefore examine:

the target's existing competitive position;

innovation potential;

pipeline products;

access to important technology;

data assets;

employees;

intellectual property;

likely future competition.

This makes merger control particularly important in rapidly developing AI markets.

14. Case Law

1. Microsoft v Commission — Case T-201/04

General Court of the European Union

Microsoft is an important authority concerning technological ecosystems, interoperability and exclusionary conduct.

The case concerned Microsoft's refusal to provide interoperability information and its tying of Windows with Windows Media Player.

Importance for AI

The case provides an analytical framework for situations where a technologically dominant undertaking controls an important platform and uses that control in a manner that may disadvantage competing products.

An analogous AI scenario could involve:

dominant operating system/cloud ecosystem → proprietary AI service → restrictions on competing AI systems.

The case demonstrates why technological architecture can become an important component of competition analysis.

15. Bronner v Mediaprint — Case C-7/97

Bronner is a leading authority concerning refusal to provide access to infrastructure.

The Court applied a demanding standard before requiring a dominant undertaking to provide competitors with access to infrastructure.

AI relevance

The principles can be relevant to:

AI cloud infrastructure;

computing capacity;

proprietary AI interfaces;

essential technical infrastructure;

specialized data systems.

A competitor cannot automatically demand access to every resource controlled by another undertaking.

The question is whether the infrastructure satisfies the stringent requirements for intervention under refusal-to-deal principles.

16. Commercial Solvents v Commission — Joined Cases 6/73 and 7/73

Commercial Solvents concerned the refusal by a dominant undertaking to supply an input to a downstream competitor.

The Court recognized that a dominant undertaking controlling an important upstream input could not use that position to eliminate downstream competition in circumstances constituting abuse.

AI relevance

The case is relevant to vertically integrated AI markets.

For example:

AI chip/cloud infrastructure → foundation model → AI application

If a dominant upstream provider deliberately restricts access to an indispensable input to disadvantage competing downstream AI providers, Commercial Solvents provides an important doctrinal reference.

17. United Brands v Commission — Case 27/76

United Brands remains a foundational Article 102 case concerning dominance and abusive conduct.

The judgment examined market power, barriers and the commercial dependence of customers.

AI relevance

AI markets can create substantial dependency because businesses may rely upon a particular:

AI API;

cloud platform;

model;

operating system;

marketplace.

The case illustrates the importance of examining the actual economic relationship between a powerful supplier and dependent customers.

18. Hoffmann-La Roche v Commission — Case 85/76

Hoffmann-La Roche is a major authority on exclusionary practices by dominant undertakings.

The Court emphasized the special responsibility of a dominant firm not to use practices that distort genuine competition.

AI relevance

Potentially relevant AI practices could include:

exclusive contracts;

loyalty-inducing arrangements;

conditional discounts;

preferential access;

platform incentives.

Where such mechanisms substantially foreclose rivals, the Hoffmann-La Roche framework becomes relevant.

19. Google Shopping — Case T-612/17

The Google Shopping litigation is particularly relevant to AI because it concerns a digital platform's control over visibility and ranking.

The Commission found that Google systematically favoured its own comparison-shopping service within its general search results, and the General Court substantially upheld the decision.

AI relevance

An AI platform could potentially control:

chatbot responses;

search results;

recommendation systems;

application visibility;

AI-generated referrals.

If the platform gives its own downstream services preferential treatment while competitors depend upon the platform for customer access, self-preferencing concerns may arise.

20. Google Android — Case T-604/18

The Android case concerned Google's practices relating to its mobile operating-system ecosystem, including contractual arrangements concerning application distribution, search and browsers.

AI relevance

The case demonstrates how a firm can leverage control over one technological ecosystem into related markets.

An analogous AI structure could involve:

Operating system → AI assistant → search → applications → advertising

If access to the ecosystem is conditioned upon adoption of particular complementary services, competition authorities may examine tying, bundling and exclusionary effects.

21. Intel v Commission — Case C-413/14 P

Intel is important for the analysis of exclusivity-inducing rebates and foreclosure.

The Court clarified the importance of assessing whether conduct is capable of restricting competition, particularly where evidence is presented concerning its actual or potential exclusionary effects.

AI relevance

The principle may be relevant where AI infrastructure providers offer:

exclusive cloud discounts;

preferential computing rates;

capacity incentives;

rebates tied to exclusivity;

long-term commitments.

The economic effect of such arrangements may be more important than their formal contractual terminology.

22. Qualcomm v Commission — Case T-235/18

Qualcomm concerned payments and arrangements relating to baseband chipsets.

The case is relevant to technology markets where contractual arrangements can affect access to important components and downstream competition.

AI relevance

Similar analytical issues may arise where AI developers depend upon:

chips;

cloud resources;

mobile distribution;

technical standards.

Contractual incentives can potentially become an entry barrier where they substantially restrict rivals' access to an important technological ecosystem.

23. Amazon Marketplace Competition Proceedings

European competition authorities have examined Amazon's use of non-public marketplace seller information and the relationship between its marketplace role and its own retail activities.

AI relevance

The analogy is significant for AI ecosystems.

A platform may possess information concerning third-party AI developers while also competing with them through its own AI products.

Potential concerns could include:

use of developer data;

preferential ranking;

discriminatory access;

preferential API terms;

self-preferencing.

The underlying competition question is whether control of the platform is being used to disadvantage independent competitors.

24. Types of AI Entry Barriers

25. Natural Versus Artificial Entry Barriers

This distinction is fundamental.

Natural barriers

These arise from the economics of AI itself:

computational expense;

economies of scale;

R&D expenditure;

scarcity of specialized expertise.

These are not automatically unlawful.

Artificial barriers

These may result from deliberate conduct:

exclusionary contracts;

discriminatory API access;

refusal to supply;

interoperability restrictions;

strategic tying;

self-preferencing;

discriminatory cloud access;

exclusionary acquisitions.

Competition authorities are more likely to focus on the second category when it involves an undertaking with substantial market power.

26. AI Entry Barriers and Abuse of Dominance

Under an Article 102-type framework, a competition authority would typically examine:

Step 1 — Dominance

Does the undertaking possess substantial market power?

Step 2 — Entry conditions

Are there substantial barriers preventing competitors from entering?

Step 3 — Conduct

Has the dominant undertaking engaged in conduct that reinforces those barriers?

Step 4 — Foreclosure

Can the conduct substantially reduce competitors' ability to compete?

Step 5 — Effects

What consequences arise for:

price;

quality;

innovation;

consumer choice;

privacy;

technological development?

Step 6 — Justification

Does the undertaking have legitimate objective justifications or efficiencies?

27. AI and Refusal to Deal

A dominant AI company may possess infrastructure that competitors cannot realistically reproduce.

Examples could include:

unique datasets;

specialized AI interfaces;

critical cloud capacity;

interoperability infrastructure.

But competition law generally does not impose a universal obligation to assist competitors.

The strict approach developed in Bronner is therefore important.

Compulsory access normally requires a strong factual and legal basis.

28. AI and Self-Preferencing

Suppose a dominant AI marketplace permits independent developers to publish AI applications while also offering its own applications.

The platform controls:

rankings;

recommendations;

search;

user reviews;

advertising;

default positioning.

If the platform systematically promotes its own applications, competitors may face an artificial disadvantage.

The Google Shopping case provides an important analytical reference.

29. AI and Tying

A dominant undertaking could potentially require customers purchasing one AI service to purchase another.

Examples might include:

cloud computing + proprietary AI model;

operating system + AI assistant;

enterprise software + proprietary AI service;

AI model + proprietary payment infrastructure.

Competition authorities would examine whether:

the products are distinct;

the undertaking is dominant in the tying market;

customers are effectively compelled to obtain the tied product;

competitors are foreclosed; and

the practice has legitimate justification.

30. AI and Interoperability

Interoperability can significantly affect entry.

If users can easily move between AI systems, switching costs decline.

If platforms prevent interoperability, users may become locked into one ecosystem.

Relevant issues include:

model portability;

API compatibility;

data portability;

identity portability;

application interoperability;

cross-platform functionality.

The Microsoft case provides a useful precedent for understanding the relationship between interoperability and competitive foreclosure.

31. AI Market Entry and Merger Control

Merger control can be particularly significant because AI markets develop rapidly.

A startup with:

few current revenues;

highly skilled employees;

valuable research;

promising technology;

important intellectual property

may nevertheless represent a significant future competitive constraint.

Authorities may therefore need to examine not merely current market shares but also:

innovation competition;

pipeline products;

technological capabilities;

potential entry;

access to important inputs.

32. Remedies

Where competition authorities establish unlawful exclusionary conduct, possible remedies may include:

Behavioural remedies

non-discriminatory access;

interoperability;

data portability;

prohibition of exclusive arrangements;

transparency requirements;

restrictions on self-preferencing.

Structural remedies

In exceptional circumstances:

divestiture;

separation of infrastructure and downstream operations;

business-unit separation.

Merger remedies

Possible remedies may include:

asset divestiture;

licensing;

access commitments;

interoperability commitments;

restrictions on exclusive agreements.

33. Competition-Law Policy Challenge

AI creates an unusual competition-law tension.

Competition law should not treat every technological advantage as an antitrust problem.

A company may legitimately obtain an advantage because it:

innovated first;

invested heavily;

developed superior technology;

created a better product;

achieved economies of scale.

Intervention becomes more relevant when an undertaking uses market power to artificially prevent rivals from competing on the merits.

The distinction between:

competition through innovation

and

exclusion through control of essential competitive inputs

is therefore central.

34. Conclusion

AI markets can contain unusually strong entry barriers because of the combination of data, computing power, specialized hardware, intellectual property, human capital, network effects, capital requirements and distribution ecosystems.

The existence of these barriers is not itself unlawful. The principal competition-law concern arises when a firm possessing substantial market power deliberately uses exclusionary strategies to make those barriers greater or to prevent otherwise viable competitors from entering or expanding.

The leading cases of United Brands, Hoffmann-La Roche, Commercial Solvents, Bronner, Microsoft, Intel, Google Shopping and Google Android provide useful doctrinal foundations.

In practical terms, AI competition analysis should therefore examine the entire value chain:

AI chips → cloud infrastructure → data → foundation models → APIs → applications → distribution → consumers.

The central legal question is whether control at one or more levels of that chain merely reflects legitimate technological success, or whether it is being used in a manner that unlawfully restricts the ability of competing AI undertakings to enter, expand and compete.