Competition Law and Market Concentration in Autonomous Enterprise Networks

1. Introduction

Autonomous Enterprise Networks (AENs) are business environments in which software agents, AI systems, enterprise platforms, connected machines, procurement systems, logistics systems, cloud infrastructure, and other automated tools interact with limited human intervention. An autonomous enterprise network may, for example, allow AI agents to:

• negotiate procurement terms;
• select suppliers;
• allocate inventory;
• determine prices;
• route logistics;
• purchase cloud capacity;
• manage energy consumption;
• select advertising channels;
• exchange data with other enterprises;
• execute contracts automatically; and
• coordinate production and distribution.

The competition-law problem arises when these networks become highly concentrated. A small number of firms may control the operating system, cloud layer, data, identity layer, AI models, APIs, transaction infrastructure, or agent marketplace on which other autonomous enterprises depend.

The central issue is therefore not merely whether an enterprise is large, but whether concentration creates the ability to control access, data, interoperability, transactions, standards, or autonomous decision-making in a way that excludes rivals or entrenches market power.

Recent competition-policy work is increasingly examining precisely these questions. For example, the UK CMA has identified agentic AI and algorithmic decision-making as emerging competition issues, including the possibility of new forms of algorithmic coordination.

2. Meaning of Market Concentration in Autonomous Enterprise Networks

Traditional concentration analysis generally examines the market shares of firms supplying a defined product or service.

In autonomous enterprise networks, concentration can exist at multiple technological layers.

A. Infrastructure concentration

A small number of firms may control:

• cloud computing;
• GPUs and AI compute;
• data centres;
• enterprise operating systems;
• network infrastructure;
• identity services.

B. Intelligence-layer concentration

Concentration may arise where only a few firms provide:

• foundation models;
• enterprise AI models;
• autonomous agents;
• reasoning systems;
• specialised industrial AI.

C. Data concentration

A firm may possess uniquely valuable:

• transaction data;
• customer data;
• industrial data;
• logistics data;
• behavioural data;
• training data.

D. Interoperability concentration

A dominant firm may control the APIs, protocols or technical standards necessary for agents to communicate.

E. Transaction concentration

An enterprise agent may ultimately have to use a particular:

• marketplace;
• payment network;
• procurement platform;
• logistics exchange;
• advertising exchange;
• cloud marketplace.

F. Decision-making concentration

The most novel issue is control over autonomous decisions.

If millions of enterprise agents rely upon the same AI model or optimisation platform, the provider could potentially influence decisions involving:

price + supplier selection + inventory + advertising + logistics + financing.

This can produce substantial competitive effects even where the underlying markets appear fragmented.

3. Why Autonomous Networks Can Intensify Concentration

3.1 Network effects

Autonomous enterprise networks frequently exhibit strong network effects.

More participating enterprises produce:

more transactions → more data → better models → better autonomous decisions → more users → more data.

This can create a self-reinforcing feedback loop.

The resulting concentration may therefore be considerably stronger than ordinary economies of scale.

3.2 Data feedback loops

Suppose Platform A has 60% of enterprise transactions.

Its AI agent observes:

• millions of supplier quotations;
• procurement outcomes;
• delivery times;
• price changes;
• customer behaviour;
• inventory movements.

The resulting data improves its model.

A competing platform with only 5% of transactions receives substantially less data.

This produces:

market share → data advantage → AI improvement → greater market share.

Competition law may therefore need to consider dynamic concentration, rather than merely current market shares.

4. The Multi-Layer Market Problem

Autonomous enterprise networks create difficulty in defining the relevant market.

A single company might simultaneously operate:

1. cloud infrastructure;
2. foundation AI models;
3. enterprise software;
4. agent marketplaces;
5. payment systems;
6. advertising services;
7. data services.

Thus, the relevant question may not be:

"What is the company's market share?"

but:

"At which layer does the company possess a bottleneck through which competing autonomous enterprises must pass?"

This is particularly important because conventional market shares may understate ecosystem power.

5. Competition Concerns

A. Excessive concentration

High concentration can facilitate:

• unilateral market power;
• exclusionary conduct;
• higher prices;
• reduced innovation;
• reduced choice;
• inferior interoperability.

B. Self-preferencing by autonomous agents

Suppose an enterprise AI agent receives:

"Find the cheapest suitable supplier."

If the agent systematically recommends suppliers affiliated with its own platform, competition concerns can arise.

The issue becomes particularly serious where the agent controls the decision-making interface through which customers interact with the market.

The EU's recent DMA enforcement against Google illustrates the broader concern: the European Commission stated in July 2026 that Google had given preferential treatment to its own services in search rankings, while also imposing measures concerning interoperability of competing AI services with Android.

6. Exclusive Access and Foreclosure

A dominant autonomous-network operator may impose:

• exclusive API arrangements;
• exclusive cloud arrangements;
• exclusive agent distribution;
• exclusive data access;
• exclusivity payments;
• restrictions on multi-homing.

The concern is that competitors may technically exist but be unable to obtain sufficient scale.

This resembles the traditional competition-law concept of foreclosure.

7. Interoperability as a Competition Issue

Interoperability becomes especially important where autonomous agents must communicate.

Consider:

Enterprise A → AI Agent → API → Enterprise B → Cloud → Payment System

If one company controls the API layer, it can potentially determine:

• which agents communicate;
• what data they receive;
• what functionality is available;
• how quickly competitors obtain access.

The European Commission's 2026 Android measures are an important contemporary illustration. The Commission stated that competing AI assistants needed effective access to Android functionality so that they could compete with Google's own AI services.

8. Data Access and Data Portability

Concentration may also arise from control over commercially essential data.

Competition concerns may arise where a dominant network:

• refuses reasonable access to data;
• makes portability technically difficult;
• restricts API access;
• imposes discriminatory data-access conditions;
• combines data from multiple markets;
• uses exclusive data to train autonomous systems.

The competitive advantage can become cumulative.

Data concentration cycle

More users

↓

More transactions

↓

More data

↓

Better AI

↓

Better autonomous decisions

↓

More users

This is one of the most significant structural risks associated with autonomous enterprise networks.

9. Algorithmic Coordination and Tacit Collusion

Autonomous enterprise networks create a particularly difficult issue where independent AI agents interact.

Suppose competing suppliers deploy autonomous pricing agents.

Each agent independently observes:

• competitors' prices;
• inventory;
• demand;
• capacity;
• market conditions.

The agents may repeatedly adjust prices.

Even without a traditional human agreement, the resulting market could become less competitive.

The CMA has specifically identified algorithmic pricing and agentic AI as areas requiring competition-law attention and has warned of possible new forms of collusion.

The legal distinction remains important:

Autonomous parallel behaviour is not automatically equivalent to an unlawful agreement.

Competition authorities would need to establish the applicable legal elements, including evidence of coordination, communication, facilitating practices, or other conduct recognised by the relevant jurisdiction.

10. Tying and Bundling

A dominant AI infrastructure provider could potentially require enterprises to purchase:

Cloud + AI model + agent platform + identity + payment infrastructure

as a package.

This can raise tying or bundling concerns where the provider possesses dominance in one layer and uses that position to strengthen another.

11. Case Laws

The following cases are particularly useful for analysing market concentration in autonomous enterprise networks. Most are precedents by analogy, because fully autonomous enterprise-agent markets are still developing.

Case 1: Google LLC and Alphabet Inc. v European Commission — Google Android

Case: C-738/22 P, judgment of 2 July 2026.

The European Court of Justice considered Google's conduct concerning Android, search, app stores, contractual restrictions, tying, exclusive pre-installation payments and Android forks. The Court's 2026 judgment dealt specifically with exclusionary effects and the competitive significance of contractual restrictions.

Relevance to autonomous enterprise networks

The case demonstrates how competition law can examine ecosystem-level power rather than treating every technological product as completely independent.

An autonomous enterprise network could similarly combine:

operating system + AI assistant + application marketplace + data + cloud + enterprise services.

The case is therefore relevant to questions concerning:

• ecosystem dominance;
• tying;
• interoperability;
• exclusive distribution;
• competing operating systems;
• network effects.

Case 2: Google LLC and Alphabet Inc. v European Commission — Google Shopping

Case: T-612/17.

The European courts examined Google's treatment of competing comparison-shopping services and the effects of preferential placement in general search.

The litigation is particularly relevant to the concept of self-preferencing.

Relevance

An autonomous enterprise agent may itself become the principal interface through which businesses access suppliers.

If an agent:

ranks → recommends → selects → purchases

and systematically favours the platform's affiliated services, the competitive problem may extend beyond ordinary search ranking.

The Google Shopping jurisprudence therefore provides a useful analytical framework for evaluating algorithmic intermediation and preferential treatment. The EU's later legal materials continue to cite Google Shopping when discussing network effects and exclusionary conduct.

Case 3: United States v. Microsoft Corp.

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

Microsoft's conduct involving Windows and competing browsers remains one of the foundational cases concerning:

• platform power;
• network effects;
• interoperability;
• exclusionary conduct;
• leveraging dominance from one layer into another.

Relevance

Autonomous enterprise networks may similarly have a dominant foundational layer.

For example:

dominant cloud/operating environment

↓

enterprise AI agent

↓

enterprise applications

↓

business transactions

The Microsoft framework is useful for analysing whether control of one technological layer can be used to restrict competition at another.

Case 4: FTC v. Amazon.com, Inc.

The U.S. Federal Trade Commission and state plaintiffs sued Amazon alleging that it used interlocking strategies to maintain monopoly power in online retail and marketplace services. The allegations include anti-discounting mechanisms, restrictions associated with Prime eligibility, search-result practices and other strategies affecting sellers and rivals.

Relevance

The case is particularly important for autonomous enterprise networks because marketplaces increasingly become automated decision environments.

An enterprise agent may decide:

• which marketplace to use;
• which supplier to select;
• which fulfilment provider to choose;
• what price to accept.

If a platform controls those automated decisions, conduct affecting ranking, fulfilment, seller access or marketplace participation can have amplified competitive consequences.

Case 5: United States v. Google LLC — Search and Advertising

The U.S. Google litigation provides another important precedent for analysing concentration around digital intermediaries and distribution channels.

Relevance

Autonomous enterprise networks can transform a conventional search or platform service into a decision gateway.

Instead of a human searching:

"Which supplier should I choose?"

an autonomous agent may independently determine:

supplier → price → contract → payment → fulfilment.

The entity controlling the agent interface could therefore occupy a strategic position similar to a digital gatekeeper.

The competition analysis should examine whether such control results from:

• genuine innovation;
• legitimate efficiency;
• exclusive arrangements;
• discriminatory access;
• foreclosure;
• leveraging.

Case 6: Microsoft Corp. v Commission

Case: T-201/04, General Court of the European Union.

The European Microsoft case concerned interoperability and Microsoft's conduct involving its operating-system dominance and related software markets.

The case is especially useful for the proposition that technical interoperability can have major competitive significance.

The EU courts continue to cite the Microsoft judgment in discussions involving network effects and competitive restriction.

Relevance

For autonomous enterprises, interoperability may involve:

• agent-to-agent communication;
• API access;
• identity;
• data portability;
• cloud switching;
• enterprise software compatibility.

A refusal or restriction can therefore become a competition issue where the controlled interface constitutes an important competitive bottleneck.

Case 7: Intel Corp. v Commission

Case: C-240/22 P, judgment of 24 October 2024.

The Intel litigation concerns exclusionary conduct and the assessment of rebates by a dominant undertaking.

Relevance

Autonomous enterprise networks may use:

• volume-based AI discounts;
• cloud credits;
• preferential API pricing;
• enterprise-agent subsidies;
• loyalty incentives.

Such arrangements should be examined not simply by asking whether discounts exist, but by considering whether they contribute to exclusionary effects.

The EU's recent legal materials continue to cite Intel alongside other major Article 102 cases in analysing exclusionary conduct.

Case 8: Tomra Systems ASA v Commission

Case: C-549/10 P.

Tomra concerned exclusionary arrangements and the ability of a dominant firm to restrict rivals' access to customers through contractual mechanisms.

Relevance

The analogy becomes significant where an autonomous enterprise platform obtains:

• exclusive enterprise customers;
• exclusive supplier relationships;
• long-term AI contracts;
• exclusivity through agent marketplaces.

Because autonomous systems can operate continuously, long-duration contractual restrictions could potentially have stronger cumulative effects than traditional one-off contracts.

12. Concentration Through Vertical Integration

A major concern is vertical concentration.

Consider a hypothetical autonomous enterprise network:

The individual markets might each appear contestable.

But the combined ecosystem may give Firm A a significant strategic bottleneck.

This is why competition authorities increasingly examine digital ecosystems rather than relying exclusively upon conventional market-share calculations.

13. The "Agent Bottleneck" Theory

A particularly important emerging concept is the agent bottleneck.

An AI agent could become the gateway through which an enterprise accesses hundreds of suppliers.

For example:

Enterprise

↓

AI procurement agent

↓

Supplier selection

↓

Negotiation

↓

Contract

↓

Payment

↓

Logistics

If the agent provider controls the ranking and selection process, it could potentially determine which suppliers receive business.

This creates a competition concern even if the agent provider itself does not manufacture the products being purchased.

14. Concentration and Market Definition

Traditional competition analysis might define:

"enterprise procurement software"

as the relevant market.

For autonomous networks, however, authorities may need to examine several overlapping markets:

1. AI model market;
2. AI agent market;
3. cloud infrastructure market;
4. enterprise software market;
5. procurement marketplace;
6. data-access market;
7. API/interoperability market;
8. transaction market.

This produces a layered-market approach.

15. HHI and Traditional Concentration Measures

The Herfindahl-Hirschman Index (HHI) can still be useful.

If four firms have market shares:

• A = 50%
• B = 25%
• C = 15%
• D = 10%

then:

HHI = 50² + 25² + 15² + 10²

= 2500 + 625 + 225 + 100

= 3,450

However, HHI may be inadequate for autonomous networks if the critical competitive resource is not sales but:

• data;
• compute;
• APIs;
• agents;
• ecosystem participation;
• switching costs;
• interoperability.

Thus, competition authorities may need a multi-dimensional concentration analysis.

16. Proposed Autonomous-Network Concentration Framework

A useful framework is:

Step 1 — Identify the network

Determine:

• participating firms;
• agents;
• platforms;
• infrastructure providers;
• suppliers;
• customers.

Step 2 — Map technological layers

Identify:

Infrastructure → Intelligence → Data → Agent → Application → Transaction

Step 3 — Identify bottlenecks

Ask:

• Who controls access?
• Who controls APIs?
• Who controls data?
• Who controls identity?
• Who controls the agent marketplace?

Step 4 — Measure concentration

Examine:

• market share;
• HHI;
• number of competitors;
• switching costs;
• multi-homing;
• customer dependency.

Step 5 — Examine network effects

Determine whether:

more users → more data → better AI → more users

creates durable entry barriers.

Step 6 — Examine conduct

Investigate:

• tying;
• bundling;
• exclusivity;
• self-preferencing;
• discriminatory access;
• refusal to interoperate;
• data restrictions;
• predatory conduct;
• algorithmic coordination.

Step 7 — Assess innovation effects

Consider whether concentration affects:

• AI innovation;
• new business models;
• interoperability;
• independent agent development;
• supplier diversity.

17. Special Problem: Autonomous Agents as Economic Actors

A difficult legal question is whether competition law should treat an autonomous agent merely as:

a tool controlled by its enterprise

or as:

an autonomous decision-making mechanism capable of generating market effects.

Current competition law generally attaches responsibility to the human or corporate undertaking rather than treating an AI system itself as a legal undertaking.

Therefore, the fact that:

"the algorithm decided"

does not necessarily eliminate the enterprise's competition-law responsibility.

The CMA's recent discussion of AI and collusion expressly emphasises that businesses need to understand the technologies they use when those technologies shape commercial decisions.

18. Autonomous Networks and Merger Control

Concentration can arise not only from conduct but also from mergers and acquisitions.

Authorities should potentially examine acquisitions involving:

• AI agent companies;
• enterprise software providers;
• data platforms;
• cloud providers;
• autonomous logistics companies;
• robotics platforms;
• procurement networks.

A transaction may raise concerns even where the target has modest current revenue if it possesses strategically important:

• data;
• technology;
• talent;
• customers;
• interoperability infrastructure.

This is particularly important in rapidly developing technology markets.

19. Remedies

Potential competition-law remedies include:

Structural remedies

• divestiture;
• separation of business units;
• restrictions on acquisitions.

Behavioural remedies

• non-discriminatory API access;
• interoperability obligations;
• data portability;
• non-preferencing;
• transparency requirements;
• restrictions on exclusivity.

Ecosystem remedies

• multi-homing rights;
• switching mechanisms;
• open technical standards;
• independent ranking systems;
• access to essential interfaces.

The EU's 2026 Android measures illustrate the growing importance of interoperability as a regulatory response where platform control can disadvantage competing AI services.

20. Autonomous Enterprise Networks: Competition-Risk Matrix

21. Key Legal Principles Emerging From the Case Law

The cases collectively support several important principles for autonomous enterprise networks:

Principle 1 — Size alone is not unlawful

Competition law generally focuses on market power and conduct, rather than simply penalising firms for becoming large.

Principle 2 — Ecosystem power matters

Google Android and Microsoft demonstrate why competition analysis may need to examine interconnected technological layers rather than isolated products.

Principle 3 — Interoperability can be competitively significant

Where competitors depend upon a dominant technological interface, restrictions on interoperability may have exclusionary consequences.

Principle 4 — Ranking and recommendation systems can affect competition

Google Shopping demonstrates the importance of algorithmically mediated access to customers.

Principle 5 — Contractual restrictions can reinforce network concentration

Exclusivity, loyalty arrangements and similar mechanisms can prevent competitors from obtaining sufficient scale.

Principle 6 — Autonomous decision-making does not remove competition-law risk

The use of AI does not automatically immunise conduct from competition scrutiny. Current CMA work expressly identifies AI-driven commercial decision-making as a developing competition-law issue.

22. Conclusion

Market concentration in autonomous enterprise networks represents a shift from conventional firm-level market power toward ecosystem-level and infrastructure-level control.

The principal competition-law concern is not simply that one company may have a high market share. It is that a company may control several interconnected layers:

Compute → Cloud → Data → AI Model → Agent → API → Marketplace → Transaction

When those layers reinforce one another through network effects, data advantages, switching costs and interoperability control, a relatively small number of firms may acquire substantial structural and strategic power.

The most relevant case-law principles come from Microsoft, Google Android, Google Shopping, Amazon, Intel and Tomra, even though none was decided on a fully developed market of autonomous enterprise agents. They provide established competition-law concepts—ecosystem power, exclusion, interoperability, tying, self-preferencing, exclusivity, foreclosure and network effects—that can be applied to this emerging technological environment.