AI Model Marketplace Competition Issues

1. Introduction

An AI model marketplace is a platform through which developers, enterprises, public bodies, or consumers can discover, compare, access, license, fine-tune, or deploy AI models supplied by multiple developers. Examples of marketplace functions include model discovery, API access, model hosting, benchmarking, pricing, authentication, billing, safety screening, and deployment infrastructure.

Competition concerns arise when a marketplace operator controls the gateway between AI-model suppliers and users. The operator may simultaneously act as marketplace intermediary, model provider, cloud provider, API provider, and application developer. This creates opportunities for self-preferencing, exclusionary ranking, tying, discriminatory access, data advantages, interoperability restrictions, excessive commissions, and acquisition of potential competitors.

The principal legal questions include:

1. What constitutes the relevant product and geographic market?
2. When does an AI marketplace operator possess market power or dominance?
3. Can ranking algorithms amount to discriminatory treatment?
4. Can the platform favour its own AI models?
5. Can access to APIs, compute, data, or deployment infrastructure be restricted?
6. Can marketplace fees or commissions exclude smaller model developers?
7. Does tying marketplace access to cloud or inference services create an antitrust problem?
8. How should competition authorities treat acquisitions of emerging AI-model competitors?

2. Relevant Market

AI marketplaces can involve several overlapping markets rather than one conventional market.

A. Model Marketplace Services

The relevant market may encompass platforms facilitating access to third-party AI models.

Possible competitive parameters include:

• number of available models;
• model quality;
• price;
• latency;
• reliability;
• interoperability;
• licensing conditions;
• developer tools;
• security;
• enterprise support.

B. AI Model Markets

Separate markets may exist for:

• foundation models;
• large language models;
• image-generation models;
• speech models;
• coding models;
• specialised industry models;
• open-weight models;
• multimodal models.

C. AI Inference and API Services

A marketplace may also compete with:

• direct model APIs;
• cloud-hosted inference;
• self-hosted models;
• specialised inference providers.

D. Cloud and Compute Infrastructure

Where marketplace access is linked to cloud infrastructure, competition authorities may examine whether the operator possesses leverage in:

• GPU computing;
• model hosting;
• inference;
• storage;
• networking;
• deployment infrastructure.

The relevant-market inquiry should therefore examine substitutability and competitive constraints, rather than assuming that every AI-related service forms one market.

3. Sources of Competition Problems

A. Self-Preferencing

The marketplace may rank its own models above competing models.

For example, a platform could:

• place its own model first;
• give its model preferential search visibility;
• recommend its own model more frequently;
• provide competitors with less favourable API access;
• use undisclosed ranking parameters favouring affiliated models.

This can become particularly important where users rarely inspect models beyond the first few search results.

The competition concern is stronger when the marketplace is both:

the intermediary controlling access + a competing AI-model supplier.

4. Search and Ranking Manipulation

AI marketplaces depend heavily on algorithms.

A marketplace may determine:

• which models appear first;
• which models receive recommendations;
• which models receive badges;
• which models qualify for enterprise listings;
• which models are included in default workflows.

Ranking discrimination can therefore function as an algorithmic form of exclusion.

Relevant evidence could include:

• ranking changes following competitive entry;
• internal communications;
• unexplained algorithmic adjustments;
• differential exposure;
• conversion rates;
• recommendation data;
• treatment of affiliated and unaffiliated models.

5. Access Discrimination

A marketplace can potentially discriminate between competing model providers through:

• API access;
• compute allocation;
• technical documentation;
• model certification;
• security reviews;
• deployment permissions;
• latency;
• data access;
• customer support.

For example, an independent model might technically remain available but receive substantially inferior infrastructure performance.

Such conduct can make competition difficult without an explicit prohibition on marketplace participation.

6. Data Advantages

An AI marketplace can generate valuable information concerning:

• user searches;
• model selection;
• prompts;
• model performance;
• conversion rates;
• pricing;
• enterprise demand;
• abandonment rates;
• model failures.

If the marketplace also operates an AI model, it could potentially use marketplace information to improve its competing model.

This creates a data feedback loop:

Marketplace users → marketplace data → proprietary model improvement → better model → more marketplace demand → additional data

The competition issue is not simply possession of data. The relevant question is whether access to marketplace-generated information gives the integrated operator a competitive advantage that rivals cannot reasonably replicate.

7. Tying and Bundling

An AI marketplace operator may bundle marketplace access with:

• cloud services;
• inference APIs;
• storage;
• cybersecurity;
• enterprise software;
• identity services;
• payment systems.

For example:

Access to premium marketplace distribution may be conditioned upon purchasing inference capacity from the platform operator.

This may foreclose independent inference providers.

Conversely, a model developer could require marketplace users to purchase related cloud services.

The analysis normally requires examination of:

• market power;
• separate products;
• coercion or conditionality;
• foreclosure;
• efficiencies;
• consumer effects.

8. Exclusive Dealing

An AI marketplace could enter arrangements under which model developers agree:

“Do not list this model on competing AI marketplaces.”

Alternatively, enterprise customers could be given discounts conditional upon exclusive use of the marketplace.

Potential consequences include:

• reduction of multi-homing;
• increased entry barriers;
• foreclosure of rival marketplaces;
• reduced access for competing model providers.

The importance of multi-homing is particularly high because AI developers and users may otherwise be able to distribute demand across several platforms.

9. Marketplace Commission and Fee Issues

AI marketplaces may charge:

• listing fees;
• API commissions;
• transaction fees;
• hosting fees;
• inference fees;
• subscription percentages;
• enterprise certification fees.

A dominant marketplace could theoretically impose conditions that disproportionately burden smaller model developers.

Possible theories include:

Excessive pricing

Where marketplace access is indispensable and the operator possesses substantial market power.

Margin squeeze

Where the marketplace operator competes downstream and sets wholesale access conditions that make efficient competition difficult.

Predatory pricing

Where marketplace services are supplied below an appropriate cost benchmark as part of an exclusionary strategy.

10. Interoperability and Portability

Competition can be weakened when users cannot easily move between AI marketplaces.

Potential barriers include:

• proprietary APIs;
• incompatible model formats;
• non-portable fine-tuning;
• proprietary embeddings;
• restricted prompt libraries;
• proprietary evaluation frameworks;
• customer lock-in.

The resulting effect may be:

Marketplace → model → application → data → workflow → switching costs

Once an enterprise has built its AI stack around one marketplace, switching can become expensive.

11. Tying Model Access to Applications

A marketplace operator may favour its own model by making it the default within:

• office software;
• search engines;
• coding environments;
• customer-service platforms;
• enterprise applications.

The competition question becomes whether control of a downstream application market allows the operator to extend its power into AI-model distribution.

This is particularly important where the marketplace operator controls a high-frequency distribution channel.

12. Network Effects

AI marketplaces may exhibit several network effects.

Direct network effects

More users attract more model developers.

Indirect network effects

More models attract more users, while more users attract more models.

Data network effects

More usage generates performance information that can improve marketplace recommendations and proprietary models.

Ecosystem effects

More developers create tools, integrations, benchmarks, and applications.

Consequently, an incumbent marketplace may become difficult to challenge even where its underlying technology is technically replicable.

13. Switching Costs and Lock-In

Users may face substantial switching costs because of:

• API redesign;
• model-specific prompt engineering;
• fine-tuning investments;
• proprietary evaluation systems;
• enterprise security approvals;
• data migration;
• workflow redesign;
• employee training.

These costs can create artificial persistence of market power.

Competition authorities may therefore examine not merely whether alternatives exist, but whether customers can practically switch to them.

14. Most-Favoured-Nation Clauses

AI marketplaces may impose contractual clauses requiring suppliers to provide:

• identical prices;
• identical commercial conditions;
• identical availability;

across competing channels.

Such provisions can potentially restrict price competition between AI marketplaces.

The analysis depends heavily on market structure and the actual effects of the clause.

15. Exclusive Model Distribution

An important emerging concern is an arrangement where a marketplace obtains exclusive rights to distribute a significant AI model.

This can produce:

Model developer → exclusive marketplace → users

If competing marketplaces cannot access an important model, their ability to attract users may be weakened.

The analysis should consider:

• duration;
• scope;
• alternatives;
• market coverage;
• model importance;
• switching possibilities;
• technical justification.

16. Merger and Acquisition Concerns

AI marketplace operators may acquire:

• model developers;
• inference providers;
• evaluation companies;
• AI application companies;
• model-routing platforms;
• data providers.

A transaction may raise concerns if the marketplace acquires a company that could otherwise become a competitive constraint.

Potential theories include:

• horizontal concentration;
• vertical foreclosure;
• elimination of potential competition;
• access to competitively sensitive data;
• ecosystem consolidation;
• reinforcement of network effects.

17. Relevant Case Laws

Because AI marketplaces are a relatively new institutional form, there are not yet many reported judgments specifically involving AI-model marketplaces. Traditional digital-platform, essential-facility, tying, refusal-to-deal, self-preferencing, and data-driven competition cases therefore provide the principal legal analogies.

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

Principle

Microsoft involved exclusionary conduct designed to protect and extend monopoly power in operating systems.

The court examined Microsoft's use of contractual and technical restrictions affecting competing browser distribution.

Relevance to AI marketplaces

The case provides an important analogy where an AI marketplace controls a critical distribution channel.

Potentially comparable conduct includes:

• preventing rival model distribution;
• restricting interoperability;
• making competing models difficult to access;
• using technical architecture to disadvantage competitors.

The central lesson is that competition law can address technical and contractual mechanisms of exclusion, not merely explicit refusals to compete.

18. Ohio v. American Express Co., 585 U.S. 529 (2018)

Principle

The Supreme Court treated the credit-card platform as a two-sided transaction platform, requiring competitive analysis to account for both sides of the platform.

Relevance to AI marketplaces

AI marketplaces can also connect two or more groups:

Model developers ↔ marketplace ↔ users

In some circumstances there may also be:

Developers ↔ enterprises ↔ cloud providers ↔ application developers

The case is useful for understanding why market definition and competitive effects in platform markets cannot automatically be analysed like a conventional one-sided product market.

19. FTC v. Qualcomm Inc., 969 F.3d 974 (9th Cir. 2020)

Principle

Qualcomm concerned licensing practices and the relationship between intellectual-property licensing, component markets, and competition.

The Ninth Circuit ultimately rejected the FTC's principal antitrust theory on the evidence presented.

Relevance to AI marketplaces

The case illustrates the importance of carefully distinguishing:

• legitimate IP licensing;
• contractual leverage;
• refusal to license;
• exclusionary conduct;
• harm to competition.

For AI marketplaces, similar questions could arise concerning:

• model licences;
• proprietary weights;
• API licences;
• inference rights;
• model redistribution rights.

20. Aspen Skiing Co. v. Aspen Highlands Skiing Corp., 472 U.S. 585 (1985)

Principle

The Supreme Court found liability where a monopolist terminated a profitable cooperative arrangement with a rival and thereby harmed competition.

The case is an important refusal-to-deal precedent.

Relevance to AI marketplaces

Suppose an established AI marketplace previously provided meaningful access to an independent model provider and subsequently terminated that relationship despite the absence of a legitimate business justification.

Relevant evidence could include:

• previous profitable cooperation;
• changed treatment following competitive entry;
• discriminatory access;
• absence of ordinary commercial justification.

The analogy is strongest where the marketplace's prior conduct demonstrates that interoperability or cooperation was commercially viable.

21. Verizon Communications Inc. v. Law Offices of Curtis V. Trinko, LLP, 540 U.S. 398 (2004)

Principle

Trinko established that antitrust law generally does not impose a broad obligation upon firms to assist competitors.

The Court was cautious about turning antitrust law into a general regulatory regime governing access.

Relevance to AI marketplaces

This is particularly important for AI infrastructure.

The fact that a marketplace controls:

• APIs;
• computing resources;
• proprietary models;
• data;
• infrastructure;

does not automatically mean competitors have an antitrust right to access them.

An AI access claim would therefore require careful examination of the applicable doctrine and the specific conduct.

22. United Brands Co. v. Commission, Case 27/76 (1978)

Principle

The Court of Justice of the European Communities examined dominance, market definition, discriminatory conduct, and refusal to supply.

The case remains an important authority concerning abusive conduct by dominant undertakings.

Relevance to AI marketplaces

The reasoning is relevant where a dominant marketplace potentially:

• discriminates among trading partners;
• restricts supplies;
• imposes unfair conditions;
• exploits dependence.

An AI marketplace could therefore face scrutiny where independent model developers are commercially dependent upon access to its distribution infrastructure.

23. Google Shopping, Case AT.39740, European Commission / General Court litigation

Principle

The Google Shopping proceedings concerned Google's treatment of its own comparison-shopping service within general search results.

The European competition-law framework examined whether Google's conduct gave preferential positioning to its own service while disadvantaging competing comparison-shopping services.

Relevance to AI marketplaces

This is one of the strongest conceptual analogies for AI marketplace self-preferencing.

An AI marketplace may similarly:

• rank its own models preferentially;
• give them greater visibility;
• integrate them into default recommendations;
• disadvantage competing models through ranking mechanisms.

The important distinction is that an AI marketplace's competitive effects would need to be established on the particular evidence and market structure.

24. Android / Google Search Cases

European Union proceedings concerning Google's Android ecosystem examined contractual arrangements, defaults, tying, and the use of ecosystem power.

Relevance to AI marketplaces

AI marketplaces may similarly combine:

Operating system / cloud / application / marketplace / model

When one firm controls several layers, it may have opportunities to leverage market power from one layer into another.

Potential competition concerns include:

• default-model placement;
• mandatory API use;
• tying AI models to cloud services;
• restrictions on alternative AI assistants;
• contractual limitations on competing marketplaces.

25. Amazon Marketplace Competition Proceedings

Competition authorities have examined Amazon's marketplace practices concerning the relationship between Amazon as marketplace operator and Amazon as a seller.

Relevance to AI marketplaces

The structural concern is highly comparable:

Platform operator + participant on the platform

An AI marketplace operator may simultaneously be:

1. marketplace owner;
2. AI-model developer;
3. cloud provider;
4. inference provider;
5. application provider.

This creates potential conflicts concerning:

• ranking;
• seller/model data;
• recommendation systems;
• access conditions;
• marketplace fees;
• self-preferencing.

26. Essential Facility Doctrine

The essential-facility concept becomes relevant when a marketplace or infrastructure service is genuinely indispensable to effective competition.

However, mere commercial importance is not enough.

Competition-law analysis ordinarily asks questions such as:

1. Is the facility genuinely indispensable?
2. Can competitors reasonably replicate it?
3. Is refusal capable of eliminating effective competition?
4. Is access technically feasible?
5. Is there an objective justification for refusal?
6. Would compelled access undermine investment incentives?

For AI marketplaces, possible candidates could include:

• uniquely important model distribution;
• indispensable interoperability infrastructure;
• technically unavoidable model-routing systems.

The doctrine must nevertheless be applied cautiously.

27. Competition Risks by Conduct

28. Competition Assessment Framework

A regulator assessing an AI marketplace could proceed through the following sequence:

Step 1 — Define the market

Identify whether the relevant market concerns:

• AI-model distribution;
• model APIs;
• model hosting;
• inference;
• cloud services;
• applications;
• or a combination.

Step 2 — Identify market power

Consider:

• market share;
• user base;
• model availability;
• network effects;
• switching costs;
• data advantages;
• technical barriers;
• ecosystem integration.

Step 3 — Identify the conduct

Determine whether the platform is:

• self-preferencing;
• tying;
• refusing access;
• discriminating;
• imposing exclusivity;
• using MFNs;
• exploiting data;
• restricting interoperability.

Step 4 — Determine foreclosure

Examine whether rival model developers or rival marketplaces actually lose:

• customers;
• visibility;
• distribution;
• infrastructure access;
• data;
• interoperability.

Step 5 — Evaluate efficiencies

Potential justifications may include:

• security;
• safety testing;
• fraud prevention;
• quality control;
• privacy;
• latency optimisation;
• technical compatibility;
• protection of intellectual property.

Step 6 — Consider remedies

Possible remedies could include:

• non-discriminatory access;
• transparency requirements;
• interoperability;
• data-use separation;
• ranking controls;
• contractual restrictions;
• portability;
• behavioural commitments;
• structural remedies in exceptional circumstances.

29. Special Problem: Marketplace Operator's Own AI Model

The most significant structural concern arises when one company controls both sides:

AI Marketplace

↓

Third-party AI models

Own AI model

This creates a potential incentive to disadvantage competing models.

For example:

A marketplace could advertise itself as an open platform while algorithmically steering users toward its own model.

The competition assessment should distinguish legitimate product integration from exclusionary conduct. Integration itself is not necessarily anticompetitive.

The critical question is whether the integration protects or extends market power by restricting effective competition.

30. AI-Specific Evidence

Traditional antitrust evidence may be insufficient for AI marketplaces.

Authorities may need to examine:

• ranking algorithms;
• recommendation logs;
• API latency;
• model-selection data;
• prompt-routing systems;
• A/B tests;
• model evaluation methodologies;
• internal model-comparison documents;
• developer access logs;
• cloud allocation;
• training-data usage;
• customer switching behaviour.

Algorithmic evidence can be particularly important because discrimination may not appear in contractual terms.

31. Potential Remedies

A. Non-discrimination

Require comparable models to receive comparable access conditions.

B. Ranking transparency

Require disclosure of significant ranking criteria where appropriate.

C. Data separation

Prevent the marketplace's proprietary AI model from receiving competitively sensitive information unavailable to rivals.

D. Interoperability

Facilitate portability between models and marketplaces.

E. Multi-homing

Prevent contractual restrictions that unnecessarily prevent model providers or users from using competing marketplaces.

F. API access

Ensure that access restrictions have objectively justified technical or security grounds.

G. Structural separation

In exceptional circumstances, competition authorities could consider separation between marketplace operations and competing AI-model activities.

32. Indian Competition-Law Perspective

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

• Section 3 — anti-competitive agreements;
• Section 4 — abuse of dominant position;
• Section 5 — combinations;
• Section 6 — regulation of combinations.

An AI marketplace could potentially raise Section 4 issues involving:

• unfair or discriminatory conditions;
• unfair or discriminatory prices;
• denial of market access;
• leveraging;
• tying or bundling;
• exclusionary conduct.

Section 3 may become relevant to:

• exclusive arrangements;
• horizontal coordination among AI providers;
• information exchange;
• platform-imposed vertical restrictions.

For combinations, AI-marketplace acquisitions should be examined not only through traditional market shares but also through:

• data;
• network effects;
• innovation;
• potential competition;
• ecosystem integration;
• control of AI infrastructure.

33. Overall Legal Position

AI model marketplaces create a hybrid competition environment combining characteristics of:

• digital platforms;
• app stores;
• cloud markets;
• software marketplaces;
• intellectual-property licensing;
• data markets;
• AI infrastructure.

The most significant competition questions are therefore likely to concern control of distribution, self-preferencing, interoperability, data advantages, access discrimination, tying, exclusivity, and ecosystem expansion.

The existing case law does not provide a single AI-marketplace doctrine. Instead, principles from platform competition, monopolisation, abuse of dominance, tying, refusal to deal, essential facilities, vertical restraints, and digital self-preferencing must be applied to the technical and economic characteristics of AI markets.

Key Case-Law Takeaway

Conclusion

The central competition-law problem in an AI model marketplace is not simply that a platform is large. It is the possibility that control over AI-model distribution becomes a strategic bottleneck through which the platform can influence which models are visible, accessible, interoperable, commercially viable, and ultimately capable of competing.