Competition Law and Governance of Reasoning Infrastructures

1. Introduction

“Reasoning infrastructures” refers to the technological and economic infrastructure through which digital systems collect information, process data, generate predictions or recommendations, rank alternatives, make automated decisions, and increasingly provide AI-generated reasoning or recommendations.

Examples include:

• AI foundation models and inference platforms;
• search and recommendation engines;
• algorithmic ranking systems;
• decision-support systems;
• cloud and computing infrastructure;
• data and identity infrastructures;
• AI assistants and agentic systems;
• automated pricing and allocation systems;
• application-programming interfaces (APIs);
• app stores and operating systems that determine access to AI functionality;
• platforms controlling the data necessary to train or improve competing systems.

Competition law becomes relevant when control over such infrastructure gives an undertaking the ability to exclude rivals, discriminate between downstream competitors, restrict interoperability, exploit data advantages, self-preference its own services, tie complementary products, or make entry dependent upon access to a bottleneck infrastructure.

The concept is not itself a statutory competition-law category. Rather, it is a useful analytical framework for applying established doctrines concerning dominance, monopolisation, exclusionary conduct, essential facilities, tying, leveraging, self-preferencing, discriminatory access, data advantages, interoperability and mergers.

Recent European enforcement illustrates the development of this area. In July 2026, the European Commission adopted measures requiring Google to provide competing AI assistants with effective access to relevant Android functionality and to make certain search data available to competing search services.

2. Meaning of Reasoning Infrastructure

Traditional infrastructure consists of physical or technical systems such as:

roads → electricity grids → telecommunications networks → payment systems.

Reasoning infrastructure increasingly consists of:

data → computing → models → algorithms → inference → ranking → recommendation → decision.

The important competition-law question is therefore not merely:

“Who owns the AI model?”

It may instead be:

“Who controls the infrastructure through which economically significant reasoning is produced and delivered?”

For example, a company may control:

1. the operating system;
2. the app marketplace;
3. cloud computing;
4. search data;
5. user behaviour data;
6. the AI model;
7. the distribution interface;
8. the ranking algorithm; and
9. the payment mechanism.

Control over several of these layers can create vertical and ecosystem-based competitive advantages.

3. Components of Reasoning Infrastructure

A. Data infrastructure

AI systems require enormous quantities of:

• search queries;
• click data;
• behavioural data;
• transaction data;
• location information;
• product information;
• user-generated content;
• training datasets;
• feedback data.

Data can become a competitive input where competitors cannot reasonably reproduce equivalent datasets.

B. Computing infrastructure

Reasoning requires substantial computational resources.

Relevant infrastructure includes:

• GPUs;
• AI accelerators;
• cloud computing;
• distributed computing;
• model-training infrastructure;
• inference servers;
• storage;
• networking.

A dominant cloud provider could potentially influence competition between AI companies if competitors depend upon its infrastructure.

The European Commission has, for example, indicated that AWS and Microsoft Azure are important gateways between businesses and customers and examined their entrenched positions, switching costs, ecosystems and AI-related capabilities under the DMA framework.

C. Model infrastructure

Foundation models may become infrastructure where many downstream products depend upon:

• an LLM;
• an embedding model;
• a multimodal model;
• a speech model;
• an AI reasoning engine;
• an API.

The competitive concern becomes particularly significant when the model provider also competes with businesses using its model.

D. Interface infrastructure

Interfaces determine how consumers reach competing services.

Examples include:

• search engines;
• app stores;
• browsers;
• operating systems;
• voice assistants;
• AI assistants;
• smart-device interfaces.

Control over the interface can permit preferential placement or exclusion of competitors.

E. Decision infrastructure

Algorithms increasingly determine:

• which product appears first;
• which seller receives visibility;
• which advertisement is displayed;
• which loan is approved;
• which consumer receives an offer;
• which route is recommended;
• which content is promoted.

Consequently, the algorithm itself can become a competitive bottleneck.

4. Competition-Law Concerns

4.1 Abuse of Dominance

The first question is whether the undertaking possesses substantial market power.

Traditional factors remain relevant:

• market share;
• barriers to entry;
• network effects;
• switching costs;
• economies of scale;
• access to data;
• vertical integration;
• technological advantages;
• ecosystem effects.

But digital reasoning infrastructures require broader analysis.

A company may have moderate market share in one market but possess considerable ecosystem power because it controls complementary infrastructure.

Germany's Section 19a regime illustrates this development by allowing intervention against companies of paramount significance for competition across markets; the Bundeskartellamt has applied the framework to Alphabet/Google, Amazon, Apple, Meta and Microsoft.

5. Self-Preferencing

Self-preferencing occurs where an infrastructure provider gives its own downstream service preferential treatment.

Examples:

Search engine → own AI answer → rival AI services pushed downward.

or:

App store → own AI assistant → rival assistants receive inferior access.

or:

Marketplace → platform's own products → rival sellers receive inferior ranking.

This is particularly important because the infrastructure operator simultaneously acts as:

infrastructure provider + competitor.

The European Commission's 2026 DMA enforcement against Google illustrates this concern: the Commission found Google non-compliant in relation to preferential treatment of its own services in Google Search and imposed a €460 million fine concerning self-preferencing.

6. Data Access and Data Advantage

A reasoning infrastructure may become difficult to challenge because the incumbent has access to uniquely valuable data.

This produces a possible feedback loop:

More users
↓
More data
↓
Better model/recommendation
↓
Better service
↓
More users
↓
More data.

This is sometimes described as a data-driven network effect.

Competition authorities may therefore examine:

• whether rivals can obtain equivalent data;
• whether data is portable;
• whether access is technically feasible;
• whether access is offered on discriminatory terms;
• whether data is unnecessarily tied to another service;
• whether the incumbent uses data generated by rivals to compete against them.

The EU's 2026 Google measures specifically addressed access by competing search services to search data, including data relevant to optimisation of search services.

7. Interoperability

Interoperability is particularly important for reasoning infrastructures.

Suppose an AI assistant can interact with:

• email;
• maps;
• messaging;
• calendars;
• shopping;
• payment;
• transportation.

If the platform owner gives its own AI assistant full technical access but restricts competing AI systems, the platform may create an artificial competitive advantage.

This is why the EU's 2026 Android measures sought equal access for competing AI assistants to Android features used by Google's own AI services.

8. Tying and Bundling

A reasoning infrastructure provider might condition access to one service upon acceptance of another.

Examples:

• AI assistant + operating system;
• search + browser;
• cloud + AI model;
• app store + payment service;
• AI API + cloud infrastructure.

The competition-law concern is whether the bundle forecloses competing providers.

9. Essential-Facilities-Type Problems

A particularly difficult question is:

When does AI or reasoning infrastructure become sufficiently indispensable that refusal of access raises competition-law concerns?

Traditional essential-facilities doctrine generally requires careful examination of:

1. indispensability;
2. lack of realistic alternatives;
3. elimination of effective competition;
4. technical and economic feasibility;
5. objective justification.

The doctrine should not automatically apply merely because an infrastructure is technologically important.

10. Algorithmic Discrimination

Reasoning infrastructures can discriminate between competitors through apparently neutral algorithms.

Examples:

• lower ranking;
• reduced recommendation;
• inferior API performance;
• delayed access;
• higher computational costs;
• reduced visibility;
• differential data access;
• restricted functionality.

The difficulty is that discriminatory conduct can be embedded in:

code rather than contracts.

Competition authorities therefore increasingly need to examine algorithmic architecture, not merely written agreements.

11. Algorithmic Collusion

Reasoning infrastructures may also facilitate coordination.

Multiple firms could employ algorithms that:

• monitor rivals;
• instantly react to price changes;
• learn from market behaviour;
• converge upon parallel pricing;
• reduce incentives for independent competitive decisions.

The legal challenge is distinguishing:

lawful independent algorithmic adaptation

from

concerted conduct or coordination prohibited by competition law.

Mere use of similar algorithms does not automatically establish an unlawful agreement.

12. Merger Control

Reasoning infrastructure also changes merger analysis.

A transaction may involve acquisition of:

• an AI model;
• a specialised dataset;
• a cloud provider;
• an AI chip company;
• an AI distribution platform;
• an algorithmic recommendation system;
• an AI application with a large user base.

The concern may arise even where the target's present revenue is relatively small.

Authorities may examine:

• nascent competition;
• innovation competition;
• access to data;
• interoperability;
• vertical foreclosure;
• input foreclosure;
• customer foreclosure;
• ecosystem expansion;
• control over future AI markets.

13. Six Major Case Laws

13.1 Google Search (Shopping) — European Commission / General Court

Google Search (Shopping) is one of the most important cases concerning algorithmic infrastructure.

Google operated a general search engine while also offering its own comparison-shopping service.

The Commission concluded that Google had abused its dominant position by systematically giving prominent placement to its own comparison-shopping service while applying less favourable positioning and display to competing comparison-shopping services.

Competition-law significance

The case demonstrates that:

control over ranking infrastructure can become a competitive advantage when the infrastructure operator also competes downstream.

The underlying issue was not simply price discrimination. It concerned the architecture of visibility.

Relevance to reasoning infrastructure

An AI search or recommendation system could present a similar problem where:

AI-generated recommendations → preferentially select the platform's own services.

The principle is therefore highly relevant to AI-mediated markets.

14. Google Android — Google LLC and Alphabet Inc. v European Commission

The Android litigation concerned several contractual practices involving Google's mobile ecosystem.

The Court of Justice's 2026 judgment in Case C-738/22 P addressed, among other issues:

• tying;
• exclusionary effects;
• payments connected with pre-installation;
• Android forks;
• competitive effects;
• the relevance of context and counterfactual analysis.

Competition-law significance

The case illustrates how control over one infrastructure layer can influence adjacent markets.

Android was not merely an operating system. It was part of an ecosystem involving:

operating system → search → app store → applications → users.

Relevance to reasoning infrastructure

The same structure can arise with:

operating system → AI assistant → applications → AI-generated actions.

If an operating-system provider controls technical access to applications and simultaneously operates its own AI assistant, interoperability restrictions can have exclusionary effects.

15. Microsoft — Internet Explorer

The Microsoft browser litigation remains an important precedent for understanding tying and technological integration.

Microsoft's position in PC operating systems was used as the basis for analysing its integration and distribution of Internet Explorer.

Competition-law significance

The case demonstrates that competition law may scrutinise technological integration when a dominant undertaking uses control over one product to influence competition in another product.

Relevance today

A comparable question could arise where:

dominant operating system + integrated AI assistant

makes it difficult for independent AI assistants to obtain equivalent distribution or functionality.

The legal inquiry should nevertheless distinguish legitimate product integration from conduct having exclusionary effects.

16. Bronner v Mediaprint

Oscar Bronner GmbH & Co. KG v Mediaprint is a foundational European case concerning refusal of access and essential-facilities-type reasoning.

The Court imposed demanding conditions before a refusal to supply could amount to an abuse.

Core significance

A facility will not become legally indispensable simply because access would be commercially convenient.

The analysis focuses on whether:

• access is indispensable;
• there is no actual or potential substitute;
• refusal would eliminate effective competition;
• access is objectively necessary.

Application to AI infrastructure

Suppose a dominant AI provider controls a particular dataset or API.

The mere fact that competitors would benefit from access would not automatically create a competition-law duty to share it.

The Bronner principle cautions against converting competition law into a general obligation to assist competitors.

17. IMS Health v NDC Health

IMS Health GmbH & Co. OHG v NDC Health GmbH & Co. KG is another major authority concerning compulsory access to infrastructure or intellectual-property-related assets.

The case concerned a pharmaceutical-sales information system and the circumstances in which refusal to license intellectual property could constitute abuse.

Significance

The Court identified demanding conditions associated with compulsory licensing, including:

• indispensability;
• prevention of emergence of a new product;
• lack of objective justification.

Relevance to reasoning infrastructure

AI infrastructure frequently involves:

• proprietary datasets;
• model weights;
• APIs;
• software interfaces;
• specialised architectures.

IMS Health demonstrates that competition law must balance:

access and competition

against

innovation incentives and property rights.

18. Magill

RTE and ITP v Commission (Magill) is another foundational European authority.

The dispute concerned television programme information and copyright.

The case helped establish the exceptional circumstances under which refusal to license intellectual property can constitute abuse of dominance.

Relevance

The case is particularly useful when reasoning infrastructure depends on proprietary information.

For example:

proprietary database → AI training → downstream AI services.

The existence of intellectual-property rights does not automatically immunise conduct from competition law, but compulsory access remains exceptional.

19. Microsoft v Commission — Interoperability

The Microsoft interoperability litigation is particularly relevant to reasoning infrastructures.

The case concerned Microsoft's refusal to provide interoperability information necessary for competing work-group server products.

Competition-law significance

The case demonstrates the importance of interoperability where control over a technical interface can affect downstream competition.

Modern application

Consider:

AI operating system → API → third-party AI assistant.

If the platform operator provides its own AI assistant with privileged access while restricting competing AI systems, interoperability may become a central competition issue.

This logic is particularly relevant to the EU's current regulatory treatment of AI interoperability on Android.

20. Meta/WhatsApp — India

The Indian competition-law treatment of WhatsApp's 2021 privacy-policy changes is particularly important for understanding data as competitive infrastructure.

The Competition Commission of India found that WhatsApp's conduct raised concerns involving:

• unfair conditions;
• extensive data collection and sharing;
• leveraging of dominance;
• the relationship between OTT messaging and online advertising.

The matter subsequently proceeded to appellate litigation. The 2025 NCLAT proceedings record the CCI's findings concerning the use of WhatsApp's position in OTT messaging and the sharing of WhatsApp data within the Meta group.

Importance

The case illustrates a major development:

privacy-related data practices can have competition-law consequences when data is an important competitive input.

For reasoning infrastructures, this becomes even more significant because user data may improve:

• model performance;
• recommendation quality;
• advertising;
• personalisation;
• prediction;
• targeting.

21. Amazon Marketplace — Germany

The Bundeskartellamt's proceedings against Amazon demonstrate another form of infrastructure governance.

Amazon simultaneously functions as:

• marketplace;
• retailer;
• service provider;
• ecosystem operator.

The Bundeskartellamt investigated Amazon's marketplace practices, including price-parity arrangements and later issues concerning its marketplace position.

Relevance

An AI marketplace may similarly become infrastructure for third-party providers.

For example:

AI platform → third-party AI applications → consumers.

If the infrastructure provider also offers competing applications, questions concerning:

• self-preferencing;
• access;
• ranking;
• data use;
• discriminatory treatment

become important.

22. Comparative Case-Law Matrix

23. New Form of Market Power: Reasoning Power

Traditional competition analysis often asks:

Who controls production?

Digital competition increasingly asks:

Who controls distribution?

Reasoning infrastructures introduce another question:

Who controls the mechanism through which economic choices are generated?

This can be described as reasoning power.

A firm may possess reasoning power where it controls a system capable of:

1. observing market participants;
2. collecting information;
3. predicting behaviour;
4. ranking alternatives;
5. recommending outcomes;
6. allocating opportunities;
7. executing transactions.

The more stages controlled by a single undertaking, the greater the possibility of vertical foreclosure and ecosystem leverage.

24. Governance Mechanisms

Competition law can govern reasoning infrastructures through several mechanisms.

A. Non-discrimination

Infrastructure operators may be required to apply comparable conditions to competing services.

B. Interoperability

Competing AI systems may require technical access to operating-system functions, APIs or other interfaces.

C. Data portability

Users or businesses may be allowed to transfer data to competing providers.

D. Data access

Where legally justified, strategically important datasets may need to be made accessible under appropriate conditions.

E. Transparency

Platforms may be required to disclose sufficient information concerning ranking or access rules to prevent discriminatory manipulation.

F. Structural separation

In particularly serious circumstances, authorities may consider separating infrastructure provision from downstream competitive activities.

G. Behavioural remedies

Examples include:

• non-discrimination obligations;
• access commitments;
• interoperability;
• firewalls;
• data-use restrictions;
• monitoring;
• independent auditing.

25. Digital Markets Act Dimension

The EU Digital Markets Act provides a particularly important ex ante complement to traditional competition law.

The distinction can be represented as:

Traditional competition law
↓
Identify market power
↓
Identify abusive conduct
↓
Investigate
↓
Remedy

versus:

Digital regulation
↓
Identify systemic gatekeeper
↓
Establish behavioural obligations
↓
Monitor compliance
↓
Intervene earlier.

This distinction is particularly significant for reasoning infrastructure because waiting for a market to tip completely may make restoration of competition difficult.

The Commission's 2026 measures concerning Google's Android AI interoperability and search-data access demonstrate this preventive approach.

26. The Feedback-Loop Problem

Reasoning infrastructures can generate self-reinforcing competitive advantages:

Users

↓

Data

↓

Better prediction/reasoning

↓

Better recommendations

↓

Higher user engagement

↓

More data

↓

Better AI

This creates a data–intelligence–distribution feedback loop.

Competition authorities should therefore examine not merely current market share but also:

• data accumulation;
• switching costs;
• learning effects;
• network effects;
• interoperability;
• ecosystem expansion;
• control over distribution.

27. Risks of Over-Regulation

Competition governance must also recognise that intervention can have costs.

Excessive mandatory access may:

• reduce incentives to innovate;
• undermine investment;
• expose confidential information;
• create cybersecurity risks;
• compromise privacy;
• facilitate free-riding;
• reduce incentives to develop proprietary infrastructure.

Therefore, the objective should not simply be:

“Force every AI provider to share everything.”

Rather, competition law should distinguish between:

legitimate proprietary infrastructure

and

strategic exclusionary control over indispensable competitive bottlenecks.

The demanding access principles from Bronner, IMS Health and Magill remain important safeguards.

28. Future Competition-Law Issues

28.1 AI agent competition

AI agents may become intermediaries between consumers and businesses.

An agent might decide:

which airline to book → which bank to use → which retailer to purchase from.

Control over the agent could therefore become equivalent to control over a major distribution channel.

28.2 AI search

Traditional search results may increasingly be replaced by AI-generated answers.

The competition question becomes:

Who determines which businesses, products and information sources are included in the AI answer?

This transforms ranking power into reasoning power.

28.3 AI cloud concentration

AI companies may depend upon a small number of cloud and accelerator providers.

This raises possible:

• input foreclosure;
• discriminatory access;
• bundling;
• switching-cost;
• interoperability concerns.

28.4 Foundation-model ecosystems

A foundation-model provider may compete simultaneously at several levels:

model → API → cloud → application → distribution.

The greater the vertical integration, the more important foreclosure analysis becomes.

28.5 Algorithmic procurement

AI systems may increasingly select:

• suppliers;
• contractors;
• logistics providers;
• financial products.

Competition law will have to consider whether dominant reasoning systems systematically favour affiliated businesses.

29. Proposed Analytical Framework

A competition authority examining a reasoning infrastructure could ask:

Step 1 — Identify the infrastructure

What exactly is controlled?

• data?
• model?
• cloud?
• API?
• operating system?
• search engine?
• AI assistant?

Step 2 — Identify dependent markets

Which downstream businesses depend upon it?

Step 3 — Determine market power

Consider:

• share;
• network effects;
• switching costs;
• data;
• scale;
• entry barriers.

Step 4 — Identify conduct

Is there:

• self-preferencing?
• tying?
• refusal to deal?
• discriminatory access?
• data exploitation?
• interoperability restriction?
• algorithmic discrimination?

Step 5 — Establish competitive effects

Does the conduct:

• exclude rivals?
• raise their costs?
• reduce innovation?
• restrict choice?
• foreclose entry?
• reinforce network effects?

Step 6 — Consider justification

Possible justifications include:

• cybersecurity;
• privacy;
• intellectual property;
• technical integrity;
• legitimate product design;
• efficiency.

Step 7 — Select remedy

Possible remedies include:

access → interoperability → non-discrimination → data portability → monitoring → behavioural restrictions → structural remedies.

30. Conclusion

Governance of reasoning infrastructures represents an evolution of competition law from regulating conventional market conduct toward regulating the infrastructure through which economic decisions themselves are increasingly generated.

The central competitive problem is not simply that a company possesses an AI model. It is that an integrated undertaking may simultaneously control:

data + computing + models + interfaces + ranking + distribution + execution.

The principal competition-law doctrines remain relevant:

• abuse of dominance;
• essential facilities;
• refusal to deal;
• tying;
• leveraging;
• self-preferencing;
• interoperability;
• discriminatory access;
• data-related exclusion;
• merger control;
• algorithmic coordination.

The cases of Google Shopping, Google Android, Microsoft, Bronner, IMS Health, Magill, Microsoft Interoperability, Meta/WhatsApp and Amazon Marketplace demonstrate that many of the legal principles already exist. What is changing is the object to which those principles are applied.

The emerging regulatory model can therefore be expressed as:

Control of infrastructure → control of information → control of reasoning → control of access → potential control of competition.