Competition Law and Competition Governance in Model-Driven Economies

1. Introduction

A model-driven economy is an economy in which important commercial decisions are increasingly made, assisted, or optimized by computational models. These may include:

• artificial-intelligence and machine-learning models;
• algorithmic pricing models;
• recommendation and ranking models;
• credit-scoring and risk models;
• demand forecasting models;
• advertising and auction models;
• logistics and allocation models;
• search and matching models;
• foundation models and generative-AI systems; and
• predictive models used in finance, healthcare, energy, transport and infrastructure.

The central competition-law problem is that market power may increasingly arise not merely from ownership of physical assets, but from control over data, models, computing capacity, interfaces, distribution channels and feedback loops.

Traditional competition law remains applicable. What changes is the factual and economic environment in which concepts such as dominance, collusion, exclusion, tying, refusal of access, discriminatory conduct, mergers and essential facilities must be examined.

2. Meaning of a Model-Driven Economy

A model-driven economy can be represented as:

Data → Model → Prediction → Decision → Market Outcome → New Data → Improved Model

This creates a potentially self-reinforcing competitive advantage.

For example:

A platform obtains large amounts of transaction data → trains a model → provides better recommendations → attracts more users → generates more transactions → obtains more data → improves the model.

The competitive concern is not necessarily the existence of the model itself. The issue arises where the model is used to:

1. exclude competitors;
2. coordinate otherwise independent competitors;
3. discriminate against business users;
4. exploit a dominant data position;
5. foreclose access to essential inputs;
6. leverage dominance from one market into another; or
7. make switching or entry increasingly difficult.

3. Legal Framework

A. Abuse of Dominance

A dominant undertaking may violate competition law when it uses its position to exclude competitors or exploit customers.

In model-driven markets, potentially problematic conduct includes:

• preferential treatment of the dominant firm's own model;
• self-preferencing;
• discriminatory access to data;
• discriminatory access to APIs;
• refusal to provide interoperability;
• tying a model to another service;
• exclusive use of a model;
• predatory or exclusionary pricing;
• manipulation of rankings; and
• using proprietary data to disadvantage competing businesses.

B. Algorithmic Collusion

Algorithms can make coordination easier even without a traditional agreement expressly stating:

"We agree to fix prices."

A pricing algorithm may continuously observe market conditions and competitors' prices and adjust prices accordingly.

Competition authorities therefore increasingly examine:

• information exchange;
• common pricing algorithms;
• common software providers;
• algorithmic monitoring;
• automated retaliation;
• pricing recommendations;
• common parameters;
• common datasets; and
• human instructions embedded in automated systems.

The important legal principle is that automation does not itself eliminate antitrust liability.

The U.S. Department of Justice's RealPage litigation is an important contemporary example: the DOJ alleged that competing landlords supplied competitively sensitive information to a common pricing system and received pricing recommendations generated using that information.

4. Data as a Source of Market Power

Data may become competitively significant when it is:

• difficult to replicate;
• generated continuously;
• highly granular;
• proprietary;
• necessary to train an effective model; or
• linked to a large user base.

However, possession of large quantities of data is not automatically equivalent to dominance.

Competition analysis should ask:

1. Is the data commercially important?

2. Is comparable data available elsewhere?

3. Can competitors obtain the data through alternative means?

4. Is the dataset necessary for effective competition?

5. Does the dominant undertaking restrict access?

6. Does the data generate network or learning effects?

5. Model Effects and Network Effects

Traditional network effects operate approximately as:

More users → greater platform value → more users

Model-driven markets can add another layer:

More users → more data → better model → better service → more users

This can be called a data-model feedback loop.

Such feedback loops can create significant entry barriers.

A new competitor may therefore face a simultaneous disadvantage in:

• data;
• computing resources;
• training information;
• user feedback;
• distribution;
• model performance; and
• capital requirements.

Competition authorities may consequently have to examine dynamic competition, rather than merely current market shares.

6. Computing Infrastructure and Competition

Advanced models may require:

• GPUs;
• specialized chips;
• cloud infrastructure;
• data centres;
• high-bandwidth networks;
• energy;
• proprietary datasets; and
• specialized engineering talent.

Consequently, competition questions may arise at several layers:

Semiconductors → Cloud → Computing → Data → Foundation Model → Application → Distribution

Control at one layer can potentially be leveraged into another.

For example, a cloud provider that also develops AI models could theoretically have incentives to:

• prioritize its own model;
• provide preferential computing capacity;
• discriminate against competing model developers;
• bundle cloud services with AI services; or
• restrict interoperability.

These are factual questions requiring market-specific evidence rather than automatic conclusions.

7. Self-Preferencing by Models

A platform may operate both:

1. an infrastructure/model layer; and
2. downstream commercial services.

This creates a potential conflict.

For example:

A marketplace operates a recommendation model while also selling its own products.

The model could theoretically rank the platform's products more favourably.

The competition-law inquiry would examine:

• whether the undertaking is dominant;
• whether the ranking system actually disadvantages rivals;
• whether the conduct is capable of foreclosure;
• whether consumers are harmed;
• whether legitimate product-quality explanations exist; and
• whether less restrictive alternatives exist.

The Google Shopping litigation is an important precedent for understanding how preferential treatment by a dominant digital platform can be assessed. The European Commission found that Google had abused its dominant position by systematically giving prominent placement to its own comparison-shopping service.

8. Model Transparency and Competition Law

A competition authority may need to understand:

• training data;
• input variables;
• ranking criteria;
• pricing parameters;
• optimization objectives;
• feedback mechanisms;
• automated decisions;
• model updates;
• API restrictions; and
• human intervention.

This creates a new evidentiary problem.

Traditional documents may not reveal the competitive mechanism.

Evidence may instead include:

• source code;
• model documentation;
• audit logs;
• API records;
• training datasets;
• system prompts;
• technical specifications;
• model cards;
• internal communications;
• experiment results; and
• telemetry.

Thus, technical auditability becomes increasingly important to competition governance.

9. Six Major Case Laws

Case 1: United States v. Microsoft Corp.

Court: U.S. District Court for the District of Columbia / D.C. Circuit
Year: 2001 appellate decision

Principle

Microsoft demonstrated how control over an important technological platform can be leveraged to disadvantage emerging competitors.

The case concerned Microsoft's conduct concerning the Windows operating-system platform and competing browser technology.

Relevance to model-driven economies

The case provides a foundation for analysing:

• platform leverage;
• exclusionary conduct;
• interoperability;
• technological tying;
• control of distribution channels; and
• protection of emerging competitors.

A model-driven platform may similarly possess a strategic bottleneck through which competitors must pass.

Case 2: Google Search — United States v. Google LLC

Court: U.S. District Court for the District of Columbia
Major liability decision: 2024

The court concluded that Google unlawfully maintained monopolies in general search services and general search text advertising through exclusionary agreements.

The subsequent remedies proceedings have involved questions concerning distribution agreements, data access, search-index information and competing search services.

Relevance

This case illustrates how competition can be affected by:

• default settings;
• distribution agreements;
• scale;
• access to data;
• user-interaction information;
• network effects; and
• barriers to reaching sufficient scale.

For model-driven markets, the important lesson is that control over distribution can be as important as control over the underlying technology.

Case 3: Google Shopping

European Commission: Google Search (Shopping)
Decision: 2017

The European Commission concluded that Google abused its dominant position by giving systematic prominence to its own comparison-shopping service in search results.

The case is particularly relevant to model-driven economies because a ranking or recommendation mechanism can influence competition between downstream businesses.

Relevance

It illustrates the potential importance of:

• ranking algorithms;
• self-preferencing;
• vertical integration;
• visibility;
• search neutrality; and
• access to consumers.

A modern AI assistant or recommendation engine could raise analogous questions if it systematically favoured the operator's downstream services.

Case 4: Amazon Marketplace

European Commission: Amazon Marketplace investigation
Commitments: 2022

The Commission's investigation concerned Amazon's use of non-public seller data and its operation of the Buy Box and Prime-related mechanisms.

Amazon offered commitments addressing, among other matters:

• use of non-public seller data;
• Buy Box selection;
• equal treatment;
• logistics choice; and
• marketplace competition. 

Relevance

This is highly significant for model-driven markets because marketplace operators can potentially use:

Seller data → model → prediction → ranking → commercial advantage

The competitive issue is therefore not merely ownership of data, but how proprietary business-user data is converted into competitive intelligence.

Case 5: FTC v. Facebook / Meta

Court: U.S. District Court for the District of Columbia
Proceedings: continuing litigation

The FTC alleges that Facebook maintained a personal-social-networking monopoly through a course of conduct involving acquisitions and restrictions affecting developers.

Relevance

The case illustrates the significance of:

• network effects;
• acquisitions of potential competitors;
• platform ecosystems;
• developer access;
• interoperability; and
• emerging competitive threats.

For model-driven economies, acquisitions may become especially significant where a large platform acquires:

• an AI-model developer;
• a specialized dataset;
• an AI agent;
• an application using a competing model; or
• a company possessing strategically important model infrastructure.

Case 6: United States and States v. RealPage, Inc.

Court: U.S. District Court for the Middle District of North Carolina
Filed: 2024

This is one of the clearest contemporary examples of competition law directly confronting algorithmic pricing.

The DOJ alleged that competing landlords supplied RealPage with competitively sensitive information and that RealPage's software used this information to generate pricing recommendations. The complaint alleged violations of Sections 1 and 2 of the Sherman Act.

The litigation subsequently generated proposed settlements and remedies concerning algorithmic coordination and the use of competitors' sensitive information. In 2025, the DOJ proposed restrictions on RealPage's use of competitors' non-public information in runtime pricing and on the use of certain current data for model training.

Relevance

RealPage demonstrates that:

Using an algorithm as the mechanism of coordination does not necessarily change the underlying competition-law analysis.

It raises questions concerning:

• algorithmic price coordination;
• data pooling;
• common algorithms;
• information exchange;
• automated pricing;
• model training;
• monitoring;
• tacit coordination; and
• technological facilitation of collusion.

10. Additional Important Precedents

United States v. Google LLC — Ad-Tech Litigation

In 2025, the U.S. District Court for the Eastern District of Virginia held Google liable for monopolization in important open-web digital advertising markets.

In September 2026, the court imposed significant remedies involving interoperability, data sharing, anti-discrimination requirements and restrictions concerning self-preferencing.

Model-economy significance

Advertising markets increasingly rely on:

• prediction;
• automated auctions;
• audience models;
• recommendation systems;
• real-time bidding; and
• algorithmic optimisation.

Therefore, competition analysis must sometimes examine the architecture of the algorithmic ecosystem, rather than merely the price charged to consumers.

11. Competition Risks in Model-Driven Economies

12. Merger Control in Model-Driven Economies

Traditional merger thresholds based primarily on turnover may fail to capture some strategically important transactions.

A startup might have:

• low revenue;
• a valuable model;
• unique training data;
• important engineers;
• rapidly growing users; or
• strategically important technology.

Competition authorities may therefore examine:

A. Potential competition

Could the target become a significant competitor?

B. Innovation competition

Does the transaction remove an important source of innovation?

C. Data concentration

Would the transaction combine datasets that competitors cannot replicate?

D. Model concentration

Would a merger create excessive control over important models?

E. Infrastructure concentration

Would the merged entity control both model development and essential computing resources?

13. Vertical Integration

Model-driven economies are particularly susceptible to vertical integration.

Consider:

Cloud provider → AI model → application → marketplace

If the same undertaking controls all four levels, possible competition concerns include:

• discriminatory cloud pricing;
• preferential computing capacity;
• model bundling;
• exclusive access;
• self-preferencing;
• interoperability restrictions;
• tying;
• refusal to supply; and
• raising rivals' costs.

Vertical integration is not inherently unlawful. The competition question is whether the structure or conduct produces legally cognizable foreclosure or other anticompetitive effects.

14. Essential Facilities and Model Access

An AI model, dataset, computing facility or API could potentially become competitively indispensable in particular circumstances.

The analysis should examine:

1. whether the facility is genuinely indispensable;
2. whether duplication is economically or technically feasible;
3. whether access can be provided;
4. whether refusal excludes effective competition;
5. whether objective justification exists; and
6. whether access can be supplied without undermining legitimate security or intellectual-property interests.

The doctrine should therefore be applied carefully rather than treating every successful model as an essential facility.

15. Interoperability

Interoperability becomes particularly important when models operate within ecosystems.

Examples include:

• AI assistants;
• cloud services;
• operating systems;
• payment systems;
• enterprise software;
• healthcare systems;
• smart infrastructure; and
• autonomous systems.

A dominant undertaking may potentially restrict interoperability through:

• closed APIs;
• proprietary formats;
• technical barriers;
• discriminatory authentication;
• access fees;
• data portability restrictions; or
• contractual limitations.

The objective of competition governance is not necessarily to mandate universal openness, but to determine when technical restrictions produce unlawful competitive foreclosure.

16. Competition and the EU Digital Markets Act

The EU has developed a complementary ex-ante framework through the Digital Markets Act (DMA).

The EU's designated gatekeepers include Alphabet, Amazon, Apple, ByteDance, Meta, Microsoft and Booking, with various core platform services designated under the DMA.

The DMA is particularly relevant to model-driven markets because it addresses conduct involving:

• data use;
• interoperability;
• self-preferencing;
• combining data;
• platform access; and
• gatekeeper obligations.

This represents a movement from purely ex-post antitrust enforcement toward a combination of:

competition law + ex-ante digital regulation + technical compliance monitoring.

17. Competition Governance

Competition governance in a model-driven economy requires more than conventional antitrust litigation.

It involves several layers.

Layer 1 — Ex-ante regulation

Rules governing:

• access;
• interoperability;
• data use;
• platform obligations;
• merger notification; and
• algorithmic accountability.

Layer 2 — Antitrust enforcement

Investigation of:

• cartels;
• abuse of dominance;
• exclusionary conduct;
• discriminatory practices;
• tying;
• refusal to deal; and
• anticompetitive mergers.

Layer 3 — Technical governance

Authorities may require:

• algorithmic audits;
• model documentation;
• data provenance;
• access logs;
• testing;
• monitoring;
• independent verification; and
• compliance reporting.

Layer 4 — Institutional governance

Competition authorities increasingly need:

• economists;
• data scientists;
• software engineers;
• AI specialists;
• cybersecurity experts; and
• technical investigators.

18. Algorithmic Evidence

In conventional cartel cases, evidence may include:

• emails;
• meetings;
• contracts;
• telephone calls; and
• pricing agreements.

Model-driven cases may require examination of:

• model architecture;
• source code;
• training datasets;
• system instructions;
• API calls;
• model outputs;
• model-update histories;
• parameter changes;
• pricing logs;
• decision logs;
• internal dashboards; and
• communications between developers and commercial teams.

Consequently, digital forensic capability becomes part of competition enforcement capacity.

19. The Human Responsibility Principle

A central legal issue is whether companies can avoid liability by claiming:

"The algorithm made the decision."

Competition law generally focuses on the conduct and responsibility of undertakings rather than treating software as an independent legal actor.

The RealPage litigation strongly illustrates this approach: the DOJ expressly pursued the underlying information-sharing and pricing practices rather than accepting the algorithm itself as a legal shield.

Therefore:

Automation ≠ immunity from competition law.

20. Model Bias and Competitive Discrimination

A model may produce discriminatory outcomes without an explicit human instruction to discriminate.

Competition authorities may therefore ask:

• What data was used?
• What variables were selected?
• Was competitor data incorporated?
• Were competitors treated differently?
• Did the model systematically disadvantage particular businesses?
• Did the platform know about the outcome?
• Did it modify the model after discovering the effect?

This does not mean every biased algorithm constitutes an antitrust violation. The discrimination must be connected to a recognized competition-law theory.

21. Consumer Welfare and Innovation

Model-driven markets require competition analysis beyond immediate price effects.

Relevant dimensions include:

Price

Does the conduct increase prices?

Quality

Does it reduce service quality?

Innovation

Does it reduce incentives to develop superior models?

Choice

Are users prevented from selecting alternatives?

Privacy

Does competition on privacy deteriorate?

Data portability

Can consumers and businesses move their data?

Interoperability

Can rival systems interact effectively?

Entry

Can new model developers enter the market?

22. Remedies

Competition authorities may employ:

Structural remedies

• divestiture;
• separation of business units;
• prohibition of certain acquisitions.

Behavioural remedies

• non-discrimination;
• data-access obligations;
• interoperability;
• restrictions on exclusivity;
• restrictions on data use;
• transparency requirements.

Technical remedies

• API access;
• algorithmic auditing;
• data firewalls;
• model monitoring;
• independent technical compliance.

The Google proceedings illustrate the increasingly technical character of remedies: recent U.S. remedies have included data-access, interoperability and anti-discrimination measures rather than relying exclusively on monetary penalties.

23. Key Legal Challenges

1. Defining the relevant market

Is the relevant market:

• AI models?
• foundation models?
• cloud computing?
• AI applications?
• search?
• data?
• computing capacity?

The answer may differ depending on the conduct.

2. Measuring market power

Market share alone may be inadequate where:

• markets evolve rapidly;
• products are free;
• quality competition matters;
• models are open-source;
• data creates feedback effects.

3. Separating innovation from exclusion

A company may legitimately improve its own model.

The legal question is whether particular conduct unlawfully excludes competitors rather than simply making the dominant product better.

4. Causation

Authorities must establish a sufficiently reliable connection between the challenged conduct and competitive harm.

5. Technical complexity

Competition authorities must understand highly technical systems without turning competition proceedings into unlimited technology audits.

24. A Model Competition-Governance Framework

A useful analytical framework is:

Step 1 — Identify the model

↓

Step 2 — Identify its inputs

Data / compute / infrastructure / users

↓

Step 3 — Identify its outputs

Prices / rankings / recommendations / allocation

↓

Step 4 — Identify the market position

Dominant / emerging / competitive

↓

Step 5 — Identify the conduct

Tying / exclusion / coordination / discrimination / refusal / acquisition

↓

Step 6 — Identify competitive effects

Entry barriers / foreclosure / coordination / innovation effects

↓

Step 7 — Examine efficiencies and objective justification

↓

Step 8 — Select proportionate remedy

25. Important Principles Emerging from the Case Law

The cases collectively demonstrate several important principles:

Principle 1

Technology does not create an exemption from antitrust law.

Principle 2

Control over data can become an important source of competitive advantage.

Principle 3

Control over distribution may reinforce control over models.

Principle 4

Algorithms can facilitate coordination and therefore require competition scrutiny.

Principle 5

Self-preferencing can be relevant where a dominant platform controls an important ranking or distribution mechanism.

Principle 6

Interoperability and data access can become important remedial tools.

Principle 7

Merger control must consider innovation and potential competition, not merely current turnover.

Principle 8

Competition enforcement increasingly requires technical evidence and technical expertise.

26. Conclusion

Competition law in a model-driven economy represents the intersection of traditional antitrust principles with data, artificial intelligence, algorithms, computing infrastructure and automated decision-making.

The fundamental legal principles remain recognizable:

• prohibit cartels;
• prevent abusive exclusion;
• scrutinize problematic mergers;
• protect competitive access;
• prevent discriminatory foreclosure; and
• preserve innovation and consumer choice.

What changes is the mechanism through which market power is exercised.

In a traditional economy, a dominant undertaking might control a factory, railway, pipeline or distribution network. In a model-driven economy, competitive power may instead arise from control over:

data + compute + models + interfaces + distribution + feedback loops.

The Microsoft, Google Shopping, Google Search, Amazon Marketplace, FTC v. Facebook/Meta, and RealPage matters therefore provide important building blocks for analysing modern model-driven markets. The most direct contemporary development is the RealPage litigation, which demonstrates how competition authorities can apply established antitrust principles to algorithmic pricing and shared competitively sensitive information.