Competition Law and Institutional Redesign of Competition Authorities for AI Economies

1. Introduction

Artificial intelligence is changing the conditions under which competition authorities operate. Traditional competition law generally assumes identifiable firms, relatively stable relevant markets, observable prices, and conduct that can be investigated through conventional evidence. AI economies complicate each of these assumptions.

AI markets may involve:

foundation-model developers;

cloud-computing providers;

semiconductor and accelerator suppliers;

data providers;

model marketplaces;

AI application developers;

AI agents;

API providers;

app stores;

enterprise software platforms;

digital advertising systems;

autonomous procurement and trading systems.

The resulting competition problems may therefore arise across an interconnected technological stack rather than within a single conventional product market.

“Institutional redesign” does not mean replacing competition law with an AI-specific antitrust code. It means adapting competition authorities' expertise, investigative powers, organisational structures, market-monitoring systems, merger review, technical capabilities, remedies, and cooperation mechanisms so that existing competition principles can operate effectively in AI-intensive markets.

The central question is:

How should competition authorities be structured so that they can detect, investigate and remedy anticompetitive conduct when market power increasingly depends upon computing capacity, data, algorithms, interoperability, ecosystems, intellectual property, cloud infrastructure and AI-enabled coordination?

2. Why AI Economies Require Institutional Redesign

Traditional competition authorities were largely designed around markets where the authority could identify:

the relevant market;

the relevant firms;

the relevant conduct;

the economic effects; and

an appropriate remedy.

AI markets can make each step more difficult.

Example

A dominant AI ecosystem might simultaneously control:

chips → cloud computing → foundation model → API → application store → distribution → user data → enterprise integration.

An exclusionary strategy at one layer may therefore strengthen market power at several other layers.

A competition authority examining only the downstream AI application market may miss the real source of competitive advantage.

3. Institutional Redesign: The Main Components

A modern competition authority dealing with AI economies may require at least ten institutional capabilities.

3.1 AI and computational expertise

Authorities need personnel capable of understanding:

machine learning;

neural networks;

foundation models;

model training;

inference economics;

GPUs and AI accelerators;

cloud infrastructure;

APIs;

algorithmic pricing;

synthetic data;

model fine-tuning;

reinforcement learning;

AI agents.

Economic lawyers alone may not be able to determine whether a particular technical restriction genuinely creates interoperability problems or whether an alleged technical justification is credible.

3.2 Continuous market monitoring

Traditional investigations are frequently triggered by:

complaints;

merger notifications;

leniency applications;

market studies;

information received from competitors.

AI markets may require more continuous monitoring.

Authorities could establish dedicated AI market observatories monitoring:

model concentration;

compute concentration;

cloud dependency;

API access;

switching costs;

model licensing;

data access;

acquisitions;

interoperability;

developer dependence;

pricing algorithms.

This would move competition enforcement partly from a reactive model toward a continuous market-intelligence model.

4. Redesigning Relevant-Market Analysis

AI markets can produce several complications for conventional market definition.

A single AI service may be supplied through:

a subscription;

API access;

enterprise licensing;

bundled cloud services;

advertising;

hardware;

software ecosystems.

Moreover, consumers may receive an AI service for zero monetary price.

Therefore, authorities may need to consider competition in terms of:

quality;

privacy;

innovation;

computational capacity;

latency;

accuracy;

model performance;

interoperability;

data access;

switching costs.

Example

Suppose an AI assistant is supplied free of charge but collects extensive behavioural data.

A conventional price-based analysis may incorrectly treat the service as highly competitive merely because consumers pay nothing.

The authority may instead investigate whether the firm possesses substantial power over:

attention;

data;

distribution;

developer access;

advertising;

AI-agent transactions.

5. Case Law: United States v Microsoft

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

This is one of the most important precedents for institutional redesign in digital markets.

Microsoft possessed substantial power in PC operating systems and used that position in ways that affected browser competition.

The case demonstrated that competition authorities must understand:

technological interfaces;

software architecture;

platform economics;

network effects;

distribution arrangements;

technical restrictions.

Relevance to AI

AI ecosystems similarly contain technological bottlenecks.

A foundation-model provider could potentially use control over:

APIs;

technical interfaces;

model access;

developer tools;

cloud infrastructure;

to disadvantage competing applications.

The institutional lesson is that competition authorities require technical investigation capability, not merely conventional economic analysis.

6. Case Law: Microsoft Corp. v Commission

Microsoft Corp. v Commission, Case T-201/04, General Court

The European Microsoft case concerned, among other matters, interoperability information and Microsoft's position in software markets.

The case illustrates the importance of technical interoperability where a dominant firm's infrastructure becomes important for competitors.

AI relevance

AI ecosystems increasingly depend on interoperability among:

models;

applications;

APIs;

cloud services;

identity systems;

data systems;

agent protocols.

A competition authority therefore needs specialists capable of distinguishing:

legitimate intellectual-property protection

from

technical restrictions that unnecessarily prevent competitive interoperability.

7. Case Law: Google Android

Google Android, Case AT.40099, European Commission

The European Commission examined contractual restrictions imposed within Google's Android ecosystem, including arrangements concerning search, applications and distribution.

The case illustrates how dominance can operate through an ecosystem of complementary products rather than through one isolated product.

Institutional significance

AI competition may similarly involve ecosystems consisting of:

operating system + cloud + model + app store + assistant + advertising + data.

Authorities therefore need organisational structures capable of examining cross-market leveraging.

A traditional product-by-product enforcement structure can overlook these interactions.

8. Case Law: Google Shopping

Google Search (Shopping), Case AT.39740

The European Commission found that Google had given favourable positioning to its comparison-shopping service within its general search results.

The case is particularly relevant to institutional design because the authority had to understand:

search algorithms;

ranking;

traffic allocation;

platform economics;

data feedback loops.

AI relevance

Generative AI and AI-agent ecosystems may create analogous issues involving:

model ranking;

recommendation;

agent routing;

API prioritisation;

application visibility;

AI-generated referrals.

An AI competition authority therefore needs the ability to audit algorithmic systems and their effects on market access.

9. Case Law: Google AdSense

Google AdSense, Case AT.40411, European Commission

The Commission examined contractual restrictions associated with Google's advertising intermediation services.

The case demonstrates how contractual arrangements can reinforce platform power across an ecosystem.

AI relevance

AI platforms may impose contractual restrictions on:

model access;

cloud portability;

API use;

model deployment;

data extraction;

application distribution;

interoperability.

Institutionally, authorities therefore require teams combining:

competition lawyers + economists + data scientists + engineers + sector specialists.

10. Case Law: Qualcomm

FTC v Qualcomm

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

The litigation examined Qualcomm's licensing and business practices in relation to cellular technology.

The case illustrates an important institutional problem: complex technology markets often require authorities to understand the interaction between:

intellectual property;

licensing;

component markets;

standards;

device markets;

bargaining power.

AI relevance

AI markets similarly combine:

patents;

model weights;

training infrastructure;

software;

cloud capacity;

standards;

APIs.

Competition authorities therefore need personnel capable of analysing technology-stack interactions.

11. Case Law: FTC v Meta / Facebook

The U.S. litigation involving Facebook/Meta's acquisitions of Instagram and WhatsApp illustrates the importance of examining innovation and potential competition in rapidly evolving digital markets.

The institutional lesson is particularly relevant to AI:

A small AI company may not have a large current market share but may possess:

valuable technology;

specialised researchers;

proprietary datasets;

promising models;

distribution relationships;

technological capabilities.

Consequently, competition authorities need the ability to identify future competitive constraints, rather than relying exclusively on present market shares.

12. Case Law: United States v AT&T

United States v AT&T, 552 F. Supp. 131 (D.D.C. 1982)

The AT&T litigation illustrates the importance of structural analysis where control over infrastructure can affect downstream competition.

AI relevance

Computational infrastructure can become a competitive bottleneck.

Consider:

AI accelerator → cloud → training → model → application.

If one firm controls several layers, competitors may become dependent upon it.

This means competition authorities must develop expertise in vertical infrastructure ecosystems.

13. Case Law: Aspen Skiing

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

The U.S. Supreme Court considered circumstances surrounding a dominant firm's termination of a profitable cooperative arrangement.

Although Aspen Skiing is a narrow refusal-to-deal precedent, it provides an important conceptual framework.

AI relevance

A competition authority may need to investigate whether an AI platform:

suddenly withdraws interoperability;

terminates previously available API access;

blocks data portability;

removes a competing application;

changes technical access conditions.

But accumulated cooperation alone would not automatically establish an antitrust violation. The legal test remains conduct- and market-specific.

14. Case Law: MCI Communications Corp. v AT&T

MCI Communications Corp. v AT&T Co., 708 F.2d 1081 (7th Cir. 1983)

The case is important to essential-facilities and access analysis.

Its broader institutional significance is that competition authorities must understand when control over infrastructure creates substantial competitive dependence.

AI application

Potential bottlenecks may include:

specialised chips;

cloud computing;

model-access infrastructure;

critical datasets;

AI distribution channels;

technical standards.

Authorities therefore need infrastructure economists and engineers capable of evaluating whether an alleged bottleneck is genuinely indispensable or whether alternatives exist.

15. Case Law: IMS Health

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

The CJEU addressed refusal to license an intellectual-property-protected system under stringent circumstances.

AI significance

AI markets increasingly combine:

copyright;

database rights;

trade secrets;

patents;

proprietary datasets;

model weights.

Competition authorities must therefore develop expertise in the interaction between:

IP protection and competition.

Institutional redesign should not treat every refusal to license as anticompetitive.

Instead, authorities should determine:

whether the input is indispensable;

whether competition can realistically occur without access;

whether the refusal eliminates effective competition;

whether exceptional circumstances justify intervention.

16. Case Law: Allied Tube & Conduit v Indian Head

Allied Tube & Conduit Corp. v Indian Head, Inc., 486 U.S. 492 (1988)

The U.S. Supreme Court examined conduct surrounding private standard-setting.

The case demonstrates that seemingly private technical institutions can affect competitive conditions.

AI relevance

AI economies increasingly depend on:

technical standards;

interoperability standards;

safety standards;

certification;

model evaluation standards;

agent protocols.

Competition authorities therefore need capacity to investigate standard-setting ecosystems, especially where competitors participate in creating technical rules that determine market access.

17. Institutional Problem: Traditional Organisational Silos

A conventional authority may have departments for:

mergers;

cartels;

abuse of dominance;

economics;

legal services.

AI markets cut across all these categories.

For example, an AI-cloud acquisition might simultaneously raise:

merger concerns;

vertical foreclosure;

data concerns;

compute concentration;

innovation concerns;

interoperability issues;

labour-market effects;

IP issues.

Possible institutional redesign

Authorities could establish a dedicated:

AI and Digital Markets Directorate

containing integrated teams of:

competition lawyers;

economists;

AI researchers;

software engineers;

data scientists;

cybersecurity specialists;

IP specialists;

procurement experts;

sector economists.

18. AI Market Observatory

A permanent AI market observatory could monitor indicators such as:

This does not mean that concentration itself is unlawful.

It means concentration becomes a monitoring signal requiring further investigation.

19. Redesigning Merger Control for AI

Traditional merger control often relies heavily on:

turnover;

market shares;

concentration ratios;

current competitive overlaps.

AI requires greater attention to:

19.1 Acqui-hiring

A transaction may acquire a small AI company principally for:

researchers;

engineering teams;

algorithms;

specialised expertise.

19.2 Data acquisition

A target may possess:

proprietary datasets;

behavioural information;

specialised training data.

19.3 Model acquisition

The target may own an emerging model with substantial future competitive potential.

19.4 Infrastructure acquisition

A cloud provider acquiring an AI model company could strengthen vertical integration.

19.5 Ecosystem foreclosure

A transaction may allow the acquiring firm to favour its own AI products across:

cloud;

search;

operating systems;

advertising;

app stores.

Therefore, institutional redesign should include forward-looking innovation analysis without relying exclusively on current market shares.

20. AI and Algorithmic Coordination

AI creates another institutional challenge: algorithms may facilitate coordination without conventional human communications.

Potential mechanisms include:

algorithmic pricing;

automated bidding;

autonomous agents;

real-time market monitoring;

machine-to-machine negotiations.

Competition authorities therefore need technical tools capable of distinguishing:

independent algorithmic adaptation;

conscious parallelism;

information exchange;

algorithm-assisted coordination;

explicit collusion.

The legal test still depends upon the applicable competition-law framework. The presence of an algorithm does not itself establish a cartel.

21. AI Agents and Competition Authorities

Autonomous AI agents could potentially:

negotiate purchases;

choose suppliers;

change prices;

allocate inventory;

buy advertising;

enter contracts;

optimise supply chains.

This creates an institutional question:

Who should be investigated when autonomous systems generate potentially anticompetitive outcomes?

The answer remains centred on the firms and persons legally responsible for designing, deploying and controlling the systems.

Authorities therefore need evidence concerning:

model instructions;

system architecture;

optimisation objectives;

training;

monitoring;

safeguards;

human intervention;

deployment conditions.

22. Evidence and Investigative Powers

AI competition investigations can generate enormous technical datasets.

Authorities may need access to:

source code;

model documentation;

API logs;

training records;

pricing histories;

system prompts;

contractual restrictions;

internal communications;

model evaluations;

experiments;

technical architecture;

compute allocation records.

Institutional redesign therefore requires modern digital forensic capability.

A competition authority that can subpoena documents but cannot technically analyse millions of lines of logs or model outputs may possess formal powers without effective investigative capacity.

23. Algorithmic Audit Units

A specialised algorithmic audit unit could examine:

Inputs

What information does the system receive?

Objectives

What is the algorithm optimising?

Constraints

What restrictions are imposed?

Outputs

How are competitors, consumers or suppliers treated?

Feedback loops

Does the system become more powerful as it receives additional data?

Market effects

Does the system systematically disadvantage competing firms?

This could be particularly important in:

search;

advertising;

marketplaces;

financial services;

procurement;

transportation;

AI-agent markets.

24. Remedies Must Also Be Redesigned

Traditional remedies include:

fines;

behavioural commitments;

structural separation;

divestiture.

AI may require more technically specific remedies.

Possible remedies

Interoperability

Require technical compatibility between systems.

Data portability

Permit users or businesses to transfer relevant data.

API access

Require non-discriminatory access under defined conditions.

Non-discrimination

Prevent a platform from systematically favouring its own downstream service.

Contractual restrictions

Prohibit particular exclusivity or tying arrangements.

Firewall requirements

Separate sensitive information between business units.

Structural remedies

In exceptional cases, address vertical integration or ownership concentration.

The remedy must correspond to the identified competitive harm.

25. Cooperation Between Competition Authorities

AI markets are inherently international.

A model may be:

developed in the United States;

trained using international data;

hosted on European cloud infrastructure;

deployed by an Indian company;

accessed by consumers globally.

Competition authorities therefore need stronger cooperation concerning:

evidence;

merger investigations;

market studies;

algorithmic expertise;

remedies;

technical standards.

International cooperation can reduce inconsistent remedies and duplicated investigations.

26. Competition Authorities and Other Regulators

AI markets also intersect with:

data-protection authorities;

telecommunications regulators;

financial regulators;

consumer-protection agencies;

intellectual-property offices;

cybersecurity authorities;

sectoral regulators.

Institutional redesign should therefore establish mechanisms for:

information sharing;

coordinated investigations;

jurisdictional allocation;

joint market studies;

compatible remedies.

However, competition authorities should retain an independent competition-law mandate rather than allowing general AI regulation to replace antitrust analysis.

27. Indian Competition-Law Framework

Under the Competition Act, 2002, the institutional redesign issue can be approached through existing powers rather than requiring every AI issue to be treated as a new legal category.

Section 3

Section 3 addresses agreements that cause or are likely to cause an appreciable adverse effect on competition.

AI-related examples could potentially include agreements involving:

coordinated pricing;

market allocation;

information exchange;

restrictive platform agreements;

exclusionary vertical arrangements.

Section 4

Section 4 prohibits abuse of dominant position.

Potential AI-related theories could involve:

denial of market access;

discriminatory access to APIs;

tying;

bundling;

leveraging;

unfair conditions;

exclusionary conduct;

restrictions on technical development.

The important point is:

Dominance itself is not prohibited; abuse of dominance is.

Sections 5 and 6

Merger control becomes particularly important where AI companies possess low present turnover but substantial:

technology;

data;

researchers;

models;

intellectual property;

competitive potential.

This makes institutional expertise in innovation and technology assessment particularly important.

28. Competition Commission of India and AI Expertise

For an authority such as the Competition Commission of India, institutional adaptation could include specialised capabilities in:

AI economics;

cloud markets;

semiconductor supply chains;

data markets;

algorithmic auditing;

digital forensics;

AI mergers;

platform interoperability;

computational competition analysis;

AI-agent markets.

A multidisciplinary investigation team could combine legal, economic and technical expertise from the beginning rather than adding technical expertise only after litigation begins.

29. From Market Definition to Ecosystem Mapping

A major institutional change is moving from purely market-centred analysis toward ecosystem mapping.

For an AI platform, an authority could map:

Compute

↓

Cloud

↓

Foundation model

↓

API

↓

Applications

↓

Distribution

↓

Users

↓

Data

↓

Improved model

This creates a feedback loop.

The authority can then ask:

Where is the bottleneck?

Who controls it?

Are competitors dependent upon it?

Can users switch?

Can suppliers multi-home?

Can rivals obtain equivalent inputs?

Does vertical integration reinforce market power?

30. Institutional Memory and AI Enforcement

Competition authorities themselves also need institutional memory.

AI investigations may involve:

years of technical development;

changing algorithms;

multiple acquisitions;

previous commitments;

evolving markets.

An authority therefore needs its own competition-intelligence infrastructure.

This could include databases containing:

prior merger decisions;

previous commitments;

market studies;

technical evidence;

algorithmic investigations;

industry structures;

recurring firms and ecosystems.

This is important because an authority may otherwise repeatedly rediscover the same technological facts.

31. Risks of Poor Institutional Design

Without institutional redesign, several problems can arise.

31.1 Under-enforcement

Authorities may fail to detect exclusionary conduct because they lack technical expertise.

31.2 Over-enforcement

Authorities may incorrectly treat technological innovation or integration as anticompetitive simply because it increases concentration.

31.3 Slow investigations

AI markets can evolve faster than conventional proceedings.

31.4 Remedy obsolescence

A remedy designed for one technological architecture may become ineffective after the platform changes its technology.

31.5 Fragmented enforcement

Different regulators may impose conflicting obligations.

31.6 Evidence asymmetry

The dominant AI firm may possess vastly more technical information than the regulator.

32. Proposed Institutional Model

A future competition authority could be organised around six interconnected units:

1. AI Competition Division

Specialised in AI markets and economics.

2. Algorithmic Investigation Laboratory

Technical testing and algorithmic auditing.

3. Digital Market Observatory

Continuous market monitoring.

4. AI Merger Review Unit

Assessment of acquisitions involving models, data, researchers and infrastructure.

5. Digital Forensics Unit

Collection and analysis of technical evidence.

6. International AI Competition Cooperation Office

Coordination with foreign competition authorities.

This model would supplement—not replace—the authority's existing cartel, dominance and merger functions.

33. Six Core Case-Law Lessons

Several of these cases are analogical rather than AI-specific precedents. They provide doctrinal tools for analysing problems that AI economies create; they do not establish a standalone legal doctrine of “AI institutional redesign.”

34. Key Principles for Institutional Redesign

A competition authority designed for AI economies should follow several principles.

Principle 1 — Technical neutrality

The authority should regulate competitive effects rather than favouring or disfavoring AI technology itself.

Principle 2 — Evidence-based intervention

Concentration should trigger investigation, not automatically constitute an infringement.

Principle 3 — Ecosystem awareness

Authorities should examine vertical and complementary relationships.

Principle 4 — Innovation sensitivity

Enforcement should account for both present competition and potential innovation.

Principle 5 — Technical competence

Legal and economic expertise must be supplemented by engineering and AI expertise.

Principle 6 — Continuous monitoring

Fast-moving AI markets may require ongoing market observation.

Principle 7 — Proportionate remedies

Remedies should address the specific competitive mechanism causing harm.

Principle 8 — Institutional independence

AI regulation and competition enforcement should remain analytically distinct even where they cooperate.

35. Conclusion

Institutional redesign of competition authorities is becoming an important component of competition policy for AI economies.

AI changes not only the products being traded but also the infrastructure through which competition occurs. Market power may arise from control over computing resources, data, algorithms, models, APIs, cloud infrastructure, distribution channels, technical standards and ecosystems.

The central institutional challenge is therefore not simply to create an “AI antitrust law.” It is to create competition authorities capable of understanding AI-driven sources of market power using existing competition-law principles.

The experience of Microsoft, Google, Qualcomm, IMS Health, MCI, Aspen Skiing, Allied Tube and Tetra Pak demonstrates that competition enforcement has repeatedly had to adapt when technology changes the structure of markets.

For AI economies, the next generation of competition authorities will likely need to combine competition law, economics, computer science, data science, digital forensics, merger analysis, infrastructure economics and algorithmic auditing.

The ultimate objective is to ensure that competition authorities can identify when AI-driven scale represents legitimate innovation and when control over technological bottlenecks is being used to exclude rivals, restrict market access, suppress innovation or extend dominance across interconnected markets.