Competition Law and Evolving Theories of Dominance in Machine-Managed Markets

1. Introduction

Machine-managed markets are markets in which important competitive decisions—such as pricing, ranking, allocation, recommendations, access, advertising, inventory management, or matching—are substantially determined or assisted by algorithms, artificial intelligence, automated decision systems, or machine-learning models.

Traditional competition law generally assumes that human firms make strategic decisions. Machine-managed markets complicate that assumption because:

• prices may be automatically changed without human approval;
• algorithms may learn from competitors' conduct;
• a common software provider may influence competing firms;
• platforms may use algorithms to rank their own products;
• access to data can determine the effectiveness of competing algorithms;
• machine-learning systems can create barriers to entry;
• an algorithm can become an essential intermediary between suppliers and consumers.

The central question is therefore evolving from:

“Is the firm dominant?”

toward a more complex inquiry:

“Where does decision-making power reside in a market increasingly managed by machines, data, models and automated infrastructures?”

Competition authorities increasingly recognize that algorithms can both improve competition and create risks of coordination, exclusion and market power. The UK CMA, for example, identifies explicit coordination, hub-and-spoke information exchange and potential autonomous tacit coordination as distinct algorithmic risks.

2. Meaning of Machine-Managed Markets

A machine-managed market exists where automated systems materially determine competitive outcomes.

Examples

The important point is that automation itself is not unlawful. Algorithms can lower costs, improve matching, reduce waste and increase price responsiveness. The competition concern arises when the machine becomes a mechanism for exercising, transferring, reinforcing or concealing market power.

3. Traditional Dominance Theory

Traditional dominance analysis normally examines:

1. relevant product market;
2. relevant geographic market;
3. market share;
4. barriers to entry;
5. buyer power;
6. control over important inputs;
7. network effects;
8. economic strength;
9. ability to behave independently of competitors and consumers.

Under EU law, Article 102 TFEU prohibits abuse of a dominant position. In the United States, Sections 1 and 2 of the Sherman Act address agreements restraining trade and monopolization.

Machine-managed markets do not eliminate these principles. Instead, they require courts and authorities to reinterpret market power, barriers, exclusion and competitive dependence.

4. Evolution of the Theory of Dominance

A. From Market Share to Algorithmic Control

A firm may possess substantial power even when its conventional market share does not fully reveal its influence.

An algorithm can control:

• visibility;
• ranking;
• access;
• prices;
• consumer matching;
• advertising allocation;
• data flows.

Thus, dominance can increasingly involve control over the mechanism through which competition occurs.

5. Data as a Source of Machine-Enabled Dominance

Machine-learning systems require data.

A firm possessing:

• large datasets;
• exclusive transaction histories;
• consumer behavioural data;
• real-time market information;
• proprietary training data;

may be able to build substantially better models than competitors.

This creates a potential data-driven entry barrier.

However, possession of data should not automatically equal dominance. The relevant question is whether the data is:

• sufficiently valuable;
• difficult to replicate;
• sufficiently comprehensive;
• timely;
• necessary for effective competition;
• capable of being obtained from alternative sources.

6. Algorithmic Network Effects

Machine-managed markets can produce particularly powerful network effects.

For example:

More users → more data → better algorithm → better service → more users → more data

This creates a feedback loop.

A dominant platform can therefore strengthen its position without simply increasing conventional market share.

The competition-law inquiry may consequently focus on algorithmic feedback loops and whether the dominant firm has deliberately prevented rivals from reaching sufficient scale.

7. Algorithmic Self-Preferencing

A platform may operate both:

1. the market infrastructure; and
2. competing services on that infrastructure.

Its algorithm may then rank its own services more favourably.

The European Commission's Google Shopping decision provides a major illustration. The Commission concluded that Google held dominant positions in general search and abused those positions by positioning and displaying its comparison-shopping service more favourably than competing comparison-shopping services.

The modern version of the theory is therefore:

Algorithmic infrastructure + vertical integration + preferential ranking = potential exclusionary concern.

This is particularly important where consumers rarely inspect or understand the ranking mechanism.

8. Algorithmic Pricing and Dominance

Pricing algorithms can create several different competition-law situations.

Situation 1 — Independent algorithmic competition

Each firm independently chooses its own algorithm.

Generally, this is not problematic merely because algorithms are used.

Situation 2 — Explicit agreement implemented through algorithms

Competitors agree to maintain prices and use algorithms to enforce the agreement.

This is conventional cartel conduct executed technologically.

Situation 3 — Common algorithm

Several competitors use the same pricing intermediary.

This creates potential hub-and-spoke coordination.

Situation 4 — Autonomous coordination

Algorithms independently learn that maintaining high prices produces greater profits.

This raises the most difficult question:

Can competition law respond when machines coordinate without a conventional human agreement?

The CMA has specifically identified autonomous tacit collusion as a developing competition concern.

9. Six Major Case Laws and Enforcement Developments

Case 1 — Google Search (Shopping) — European Commission

Case: Google Search (Shopping), Commission Decision AT.39740 (2017)

Principle

The European Commission found that Google had a dominant position in general search services and abused that position by systematically giving more favourable positioning to its own comparison-shopping service.

Relevance to machine-managed markets

The importance of the case extends beyond search engines.

The competitive decision was embedded in an algorithmic ranking system.

This demonstrates that:

The exercise of market power can occur through algorithmic design rather than an explicit contractual restriction.

Legal significance

Competition authorities can therefore examine:

• ranking algorithms;
• visibility;
• default positioning;
• algorithmic discrimination;
• self-preferencing.

The decision concluded that Google had been dominant in national general-search markets across the EEA and identified barriers to expansion, limited multi-homing and brand effects among relevant factors.

Case 2 — Trod Ltd / GB eye — UK CMA

Case: Trod Limited and GB eye Limited, CMA, 2016

Two competing Amazon Marketplace sellers agreed not to undercut one another.

They used automated repricing software to implement the arrangement.

The CMA found an infringement of competition law and imposed a fine on Trod; GB eye received immunity under the CMA's leniency policy.

Principle

Technology does not change the underlying legal character of cartel conduct.

If humans agree to fix prices, implementing that agreement through software remains price fixing.

Importance

The case established an important early proposition:

An algorithm can be the instrument of cartel enforcement without becoming the legal substitute for the cartel agreement.

Case 3 — Eturas — CJEU

Case: Eturas UAB and Others v Lietuvos Respublikos konkurencijos taryba, C-74/14 (2016)

Eturas operated an online travel-booking system.

A system message imposed a limitation on discounts that travel agencies could offer.

The CJEU considered whether knowledge of the electronic system and subsequent participation could establish a concerted practice.

Significance

Eturas is particularly important for machine-managed markets because it demonstrates that electronic systems can constitute the medium through which coordination occurs.

The case is relevant to the distinction between:

• direct communication;
• electronic communication;
• participation in a common technological environment;
• subsequent market conduct.

Modern algorithmic cases may extend this reasoning to automated systems.

Case 4 — RealPage Algorithmic Pricing Litigation — United States

Case: United States v. RealPage, Inc., 2024–2025 developments

The DOJ alleged that RealPage's revenue-management software facilitated coordination among competing landlords by collecting competitively sensitive information and generating rental-price recommendations.

The DOJ's 2024 complaint alleged violations of Sections 1 and 2 of the Sherman Act.

The litigation subsequently expanded to claims involving major landlords, and in November 2025 the DOJ announced a proposed settlement concerning information sharing and alignment of pricing.

Importance

RealPage illustrates a new theory:

Delegating pricing decisions to a common algorithm does not necessarily preserve independent competition.

The important factual issue is not merely whether firms used software, but whether the system:

• receives competitively sensitive information;
• aggregates rival information;
• produces recommendations;
• influences actual prices;
• reduces independent decision-making.

Case 5 — Cornish-Adebiyi v Caesars Entertainment — United States

Case: Cornish-Adebiyi v. Caesars Entertainment, Inc. and related litigation

The FTC and DOJ filed a statement of interest concerning allegations involving algorithmic hotel pricing.

The agencies emphasized that businesses cannot use an algorithm to engage in conduct that would be unlawful if performed by human actors. They also addressed circumstances where a common algorithm provider can facilitate coordination among competitors.

Principle

The case illustrates an important emerging rule:

Algorithmic intermediation does not immunize coordinated pricing from antitrust scrutiny.

The legal inquiry may therefore extend beyond direct communications between competitors to the structure of the technological intermediary.

Case 6 — Google AdTech — European Union / United States Developments

Google's advertising technology ecosystem provides another important example of machine-managed competition.

Ad auctions involve automated systems determining:

• which advertisements are displayed;
• ranking;
• price;
• allocation of advertising opportunities;
• interactions between publishers, advertisers and exchanges.

In 2026, the European Commission also imposed DMA penalties on Google concerning self-preferencing in Search and restrictions on steering users toward alternative purchase channels.

Separately, U.S. litigation concerning Google's advertising technology has addressed alleged monopolization and the design of automated advertising markets.

Significance

The modern dominance inquiry increasingly examines not merely:

“What market share does Google possess?”

but:

“How does Google's technological architecture influence the rules under which other market participants compete?”

10. The Emerging Concept of Algorithmic Dominance

Algorithmic dominance can be understood through several dimensions.

1. Data dominance

Control over commercially indispensable datasets.

2. Computational dominance

Superior access to:

• computing capacity;
• AI infrastructure;
• model training;
• specialised hardware.

3. Interface dominance

Control over the interface through which consumers access suppliers.

4. Ranking dominance

Ability to determine visibility and ordering.

5. Prediction dominance

Superior ability to predict:

• demand;
• consumer behaviour;
• prices;
• inventory;
• market movements.

6. Decision dominance

Ability to automate decisions previously made independently by market participants.

11. Theories of Dominance in Machine-Managed Markets

A. Infrastructure Dominance

A company may control the infrastructure necessary for rivals to participate.

Examples:

• cloud infrastructure;
• app stores;
• payment networks;
• digital advertising exchanges;
• AI APIs;
• operating systems.

The competition concern is stronger where competitors cannot realistically bypass the infrastructure.

B. Algorithmic Gatekeeper Theory

A platform may function as a gatekeeper because its algorithm determines who receives access to consumers.

The algorithm becomes a commercial gatekeeper.

Potential conduct includes:

• discriminatory ranking;
• self-preferencing;
• demotion of competitors;
• preferential access to data;
• algorithmic exclusion.

C. Model-Based Dominance

Future AI markets may involve competition between foundation models.

A firm could obtain power through:

• superior training data;
• model performance;
• compute resources;
• developer ecosystems;
• proprietary APIs;
• distribution;
• switching costs.

Dominance may therefore exist at several interconnected layers:

Compute → Model → API → Platform → Application → Consumer

12. Algorithmic Switching Costs

Machine-managed markets may create unusually high switching costs.

For example, a business may depend on:

• historical data;
• model-specific APIs;
• automated workflows;
• proprietary prediction systems;
• trained recommendation models.

Switching providers can require rebuilding entire technological infrastructures.

Consequently, even when another supplier nominally exists, effective competitive constraints may be weak.

13. Common Algorithm Providers and Hub-and-Spoke Theory

A particularly important emerging theory is:

Competitor A → Common Algorithm Provider ← Competitor B

The algorithm provider becomes the hub, while competing firms become the spokes.

The hub may receive:

• prices;
• inventory;
• demand information;
• occupancy;
• discounts;
• future commercial strategies.

It then generates recommendations for each participant.

The CMA has expressly identified common pricing systems and intermediaries as potential mechanisms for information exchange and coordination.

14. Autonomous Tacit Coordination

This is arguably the most difficult future problem.

Suppose:

1. Firm A's algorithm observes Firm B;
2. Firm B's algorithm observes Firm A;
3. both algorithms seek to maximise long-term profit;
4. neither company explicitly communicates;
5. the algorithms discover that maintaining high prices is profitable.

The machines may converge upon a coordinated outcome.

The traditional legal concepts of:

• agreement;
• intention;
• communication;
• concerted practice;

become difficult to apply.

The CMA has acknowledged this as an emerging theoretical possibility while distinguishing it from more established forms of algorithm-assisted coordination.

15. Personalised Pricing and Algorithmic Discrimination

Machine-managed markets can also produce individualised prices.

Algorithms may use:

• browsing history;
• purchasing history;
• location;
• device characteristics;
• consumer behaviour;
• demand elasticity.

Competition law may need to distinguish:

Legitimate dynamic pricing

Prices change because supply and demand change.

Potentially problematic personalised pricing

Prices change because algorithms identify individual consumers' willingness to pay.

The competition issue becomes particularly significant where a dominant platform possesses data unavailable to competitors.

16. Algorithmic Predation

Traditional predatory pricing involves deliberately charging below an appropriate cost benchmark to eliminate competitors.

Machine-managed systems could theoretically conduct predation at a far greater scale by automatically:

• identifying vulnerable competitors;
• lowering prices in selected regions;
• monitoring competitor exits;
• raising prices after exit.

The legal difficulty lies in proving:

1. below-cost pricing;
2. exclusionary strategy;
3. likely competitive harm;
4. recoupment where legally required.

17. Algorithmic Tying and Bundling

A dominant platform could automatically combine products or services.

Examples include:

• operating system + search;
• cloud + AI model;
• payment service + marketplace;
• advertising exchange + ad server;
• hardware + proprietary AI assistant.

Machine-learning systems may make tying more difficult to detect because the bundle can be dynamically generated.

18. Algorithmic Access Discrimination

A dominant platform may technically provide access to competitors while its algorithm provides unequal treatment.

For example:

Competitor A → normal ranking

Dominant firm's product → enhanced ranking

The discriminatory conduct may therefore be embedded in code rather than expressed in contractual language.

This is an important shift from contractual exclusion toward computational exclusion.

19. Evidence Problems

Machine-managed markets create major evidentiary challenges.

Authorities may need access to:

• source code;
• model architecture;
• training datasets;
• logs;
• prompts;
• model outputs;
• ranking criteria;
• A/B testing;
• pricing histories;
• API calls;
• internal communications.

A particularly difficult issue is the black-box problem.

A company may know what an algorithm does statistically without being able to explain every individual decision.

Competition authorities therefore increasingly need technical expertise alongside conventional economic analysis. The CMA has described expanding technical capabilities for analysing algorithms and AI.

20. New Tests for Dominance

Traditional market-share analysis may need supplementation by indicators such as:

A. Algorithmic dependence

How many competitors depend upon the firm's algorithm?

B. Data advantage

How difficult is it for competitors to reproduce the firm's data advantage?

C. Model superiority

Does the firm's model materially outperform alternatives?

D. Switching costs

Can users realistically migrate to competing systems?

E. Network effects

Does additional participation improve the algorithm?

F. Computational barriers

Would entry require economically prohibitive compute resources?

G. Ecosystem control

Does the firm control several interconnected layers?

H. Algorithmic opacity

Can competitors determine why they are being ranked, priced or excluded?

21. Difference Between Algorithmic Dominance and Traditional Dominance

22. Competition Law Remedies

Remedies may also evolve.

Structural remedies

• divestiture;
• separation of business units;
• ownership restrictions.

Behavioural remedies

• non-discrimination;
• interoperability;
• data portability;
• access obligations;
• prohibition of self-preferencing.

Algorithmic remedies

• independent algorithmic auditing;
• transparency requirements;
• model documentation;
• restrictions on sensitive-data inputs;
• monitoring of pricing systems;
• algorithmic firewalls.

Data remedies

• data portability;
• data-sharing obligations where legally justified;
• restrictions on combining datasets;
• access to interoperability interfaces.

23. Key Legal Principles Emerging from the Case Law

The six principal authorities collectively support several propositions.

Principle 1

Technology does not legalise conduct that would otherwise violate competition law.

Trod/GB eye demonstrates this directly.

Principle 2

Electronic systems can constitute mechanisms through which concerted practices occur.

Eturas is particularly important.

Principle 3

A common algorithm can facilitate coordination among competitors.

This is central to RealPage and the hotel-pricing litigation.

Principle 4

Algorithmic ranking can become an instrument of exclusion.

Google Shopping illustrates this.

Principle 5

Market power can increasingly arise from control over technological infrastructure rather than merely physical assets.

This is especially significant in platform and AI ecosystems.

Principle 6

Competition authorities must increasingly investigate the technological architecture underlying market conduct.

The CMA's continuing algorithm and AI work reflects this transition.

24. Future Direction of Competition Law

The most significant conceptual change may be the movement:

Firm dominance → Platform dominance → Ecosystem dominance → Algorithmic dominance → Machine-managed market power

In future disputes, courts may have to determine whether competitive power is exercised by:

• the company;
• the platform;
• the data;
• the algorithm;
• the model;
• the ecosystem;
• or the interaction between all of them.

This suggests that dominance analysis will increasingly become multi-layered rather than purely firm-centred.

25. Conclusion

Competition law in machine-managed markets does not require abandoning traditional doctrines of dominance. Instead, those doctrines must be adapted to circumstances where machines increasingly perform the functions through which market power is exercised.

The major developments are:

1. data can become a competitive asset and entry barrier;
2. algorithms can control access and visibility;
3. AI systems can facilitate coordination;
4. common algorithm providers can become hubs between competitors;
5. algorithmic ranking can produce self-preferencing;
6. machine-learning feedback loops can reinforce market power;
7. AI infrastructure can create new forms of dependency;
8. algorithmic opacity creates new evidentiary problems.

The central legal challenge is therefore not simply whether machines are making decisions. It is whether control over automated decision-making gives an undertaking the ability to weaken competitive constraints, exclude rivals, coordinate conduct, or make consumers and business users dependent upon a technological ecosystem.

The emerging theory of dominance consequently moves from ownership of a market toward control of the computational mechanisms through which the market operates.