1. Introduction

A superintelligent economy may be understood as an economy in which artificial intelligence systems substantially exceed human capabilities across scientific research, software development, financial analysis, logistics, design, strategic decision-making, and potentially autonomous business operations.

In such an economy, competition law would face a structural problem: traditional antitrust law generally assumes that human-controlled firms compete using relatively observable assets, prices, contracts, and business strategies. Superintelligent systems could instead compete through autonomous agents capable of continuously changing prices, negotiating contracts, acquiring competitors, allocating computational resources, designing products, and coordinating large economic networks.

The central competition-law question would therefore become:

How can competition remain open when intelligence itself becomes a scalable economic infrastructure controlled by a small number of firms?

Existing competition law already provides important principles for addressing this problem. The FTC, DOJ, CMA and other authorities have specifically identified AI-related risks involving concentration of computing resources, data, foundation models, distribution channels and strategic partnerships.

Importantly, “superintelligence” is not presently a separate legal category under conventional competition statutes. The analysis would therefore generally proceed through existing concepts such as monopoly, abuse of dominance, exclusionary conduct, vertical foreclosure, tying, refusal to deal, merger control, collusion and essential facilities.

2. Why Superintelligence Creates a Distinct Antitrust Problem

The economic characteristics of advanced AI could make market power unusually scalable.

A conventional firm might need thousands of employees to expand production. A highly autonomous AI firm could potentially deploy thousands or millions of software agents with comparatively low marginal labour costs.

This could create several forms of concentration:

1. Compute concentration
2. Foundation-model concentration
3. Data concentration
4. Cloud infrastructure concentration
5. AI-chip concentration
6. Distribution-platform concentration
7. Autonomous-agent concentration
8. AI-generated intellectual-property concentration
9. AI ecosystem concentration
10. Capital and acquisition concentration

The FTC has already identified control over essential inputs—including computational resources and specialised chips—as potential sources of competitive bottlenecks in generative-AI markets.

3. Market Definition in a Superintelligent Economy

Traditional market definition asks whether products are sufficiently substitutable.

In an AI economy, however, the relevant market could involve several layers.

Layer 1 — Compute

Examples include:

• AI accelerators;
• GPUs;
• specialised AI processors;
• high-performance servers;
• data centres;
• electricity and cooling infrastructure.

Layer 2 — Cloud AI infrastructure

This includes:

• model-training infrastructure;
• inference infrastructure;
• cloud AI APIs;
• storage;
• networking.

Layer 3 — Foundation models

Examples include:

• general-purpose language models;
• multimodal models;
• reasoning models;
• autonomous-agent models;
• scientific models.

Layer 4 — Agentic services

These may perform:

• purchasing;
• coding;
• financial analysis;
• legal research;
• logistics;
• customer service;
• scientific experimentation.

Layer 5 — Consumer and business applications

AI could become integrated into:

• search;
• operating systems;
• office software;
• finance;
• healthcare;
• education;
• transportation;
• commerce.

Consequently, an apparently competitive downstream market might actually depend upon a highly concentrated upstream AI infrastructure market.

4. Intelligence as an Economic Input

One of the most important developments would be the transformation of intelligence into an economic input.

Historically, firms purchased:

• labour;
• capital;
• machinery;
• information;
• raw materials.

A superintelligent economy could add another critical input:

machine intelligence.

If one or a few companies controlled the most capable systems, competitors might have to purchase intelligence from them.

This raises a classic leverage concern.

A dominant AI provider could potentially operate simultaneously as:

AI infrastructure provider → foundation-model provider → operating-system provider → application provider → marketplace operator.

The company could therefore compete against businesses that depend upon its own infrastructure.

That creates significant vertical-integration concerns.

5. Compute Concentration

Compute could become an important antitrust bottleneck.

Suppose a small number of firms control access to the most advanced AI chips or data centres. They might theoretically:

• reserve capacity for themselves;
• offer inferior capacity to rivals;
• impose discriminatory pricing;
• restrict access during periods of scarcity;
• enter exclusive agreements with AI developers;
• bundle computing with proprietary models.

Such conduct could resemble traditional foreclosure theories.

The legal analysis would ask:

1. Is the firm dominant?
2. Is the input genuinely difficult to replicate?
3. Are competitors dependent upon it?
4. Is access technically or economically feasible elsewhere?
5. Is discrimination occurring?
6. Does the conduct exclude equally efficient competitors?
7. Are there legitimate capacity, security or technical justifications?

6. Foundation-Model Dominance

A superintelligent foundation model could become a gateway to numerous downstream markets.

For example:

Foundation model → AI assistant → search → advertising → commerce → payments

If the model provider uses dominance in one layer to foreclose competitors in another, conventional abuse-of-dominance principles may become relevant.

The CMA's AI Foundation Models work specifically examined risks to competition arising from concentration and relationships surrounding foundation models, and developed principles aimed at maintaining fair, open and effective competition.

7. Data as a Competitive Advantage

Superintelligent systems could depend heavily upon access to:

• user interaction data;
• search data;
• behavioural data;
• scientific data;
• transaction data;
• proprietary enterprise data;
• synthetic data.

A dominant platform might use its position to accumulate more data, which improves its AI system, which attracts more users, which generates still more data.

This produces a potential:

data → intelligence → users → data feedback loop.

Antitrust authorities could therefore examine whether exclusive data arrangements or discriminatory data access artificially preserve dominance.

However, data possession alone would not automatically establish an antitrust violation. The relevant questions would include its substitutability, replicability, competitive significance and the conduct surrounding its use.

8. Autonomous Pricing and Algorithmic Collusion

Superintelligent agents could independently monitor competitors and modify prices continuously.

This creates a difficult distinction between:

Independent adaptation

Each AI system independently observes market conditions and chooses prices.

and:

Collusion

AI systems communicate or are deliberately designed to coordinate their behaviour.

The traditional distinction between conscious parallelism and agreement could become harder to apply when autonomous systems interact at machine speed.

Possible risks include:

• price coordination;
• output coordination;
• market allocation;
• customer allocation;
• wage coordination;
• procurement coordination.

Current antitrust principles would still distinguish unilateral algorithmic behaviour from an agreement or concerted practice. The technological sophistication of the system would not by itself transform independent conduct into an unlawful agreement.

9. Autonomous Agents and Tacit Coordination

Superintelligent agents might observe market conditions far more effectively than human managers.

Suppose competing AI agents repeatedly learn:

• competitors' pricing;
• production capacity;
• inventory;
• demand;
• bidding strategies.

They could potentially converge on stable market outcomes without explicit human communication.

This creates a major governance question:

When does autonomous learning become unlawful coordination?

Antitrust law would likely need to examine:

• design instructions;
• communication architecture;
• common ownership;
• data-sharing arrangements;
• intentionality;
• predictability;
• monitoring systems;
• whether firms deliberately engineered coordination.

A particularly important distinction would be between coordination as an intended commercial objective and merely parallel outcomes generated by independent optimisation.

10. Autonomous Mergers and Acquisitions

Superintelligent firms could identify potential acquisition targets far faster than human management.

They might automatically:

1. scan thousands of companies;
2. identify emerging competitors;
3. predict their technological trajectory;
4. acquire them at an early stage;
5. integrate their data;
6. absorb their researchers;
7. discontinue competing technologies.

This creates a modern form of killer-acquisition risk.

Traditional merger review may therefore need greater attention to:

• potential competition;
• innovation pipelines;
• nascent technologies;
• data assets;
• research teams;
• computational resources;
• future competitive constraints.

The relevant question would not simply be:

“Does the target currently have substantial market share?”

It could also be:

“Could the target become an important source of future competitive constraint?”

11. Self-Preferencing by Superintelligent Platforms

Suppose an AI platform operates a marketplace and simultaneously sells its own AI products.

The platform's algorithm could rank its own services more favourably.

Potential conduct could include:

• preferential ranking;
• preferential access to APIs;
• better computational resources;
• lower platform fees;
• exclusive access to user data;
• preferential recommendations.

This resembles traditional self-preferencing concerns, but the scale and opacity could be considerably greater because AI systems may make millions of ranking decisions automatically.