AI risks · Concentration of power

Do a few tech companies control AI?

The most advanced AI models are built by a small number of companies, because developing them is extremely expensive. That gives these firms considerable power and means that a fault in one widely used system could affect many sectors at once.

EmergingSocietal impact:

Why the market is concentrated

The International AI Safety Report 2025 found that a small number of companies dominate the market for general-purpose AI. Developing a state-of-the-art model can cost hundreds of millions of US dollars, mainly for computing power, skilled staff and large datasets. Market leaders also benefit from self-reinforcing effects: large companies can spread development costs over more customers, and can use data from users of older models to train new ones. The report found that this concentration persisted through 2024.

In April 2024 the UK Competition and Markets Authority (CMA) pointed to an interconnected web of more than 90 partnerships and strategic investments involving the same few firms: Google, Apple, Microsoft, Meta, Amazon and Nvidia.

Why it matters

Concentration raises two kinds of concern. One is about competition and influence: the CMA warned that firms controlling key inputs could restrict access to them, use strong positions in other markets to steer customers, and use partnerships to extend their market power. The other is resilience: if banks, hospitals and authorities all rely on the same few models, a bug or security flaw in one of them could cause simultaneous disruption across many sectors. The International AI Safety Report notes that there is little research yet on how to predict or limit such single points of failure.

Counter-trends

The picture is not static. The International AI Safety Report 2026 notes that new ways of training models across distributed computers may reduce dependence on large centralised data centres, and that openly released models bring research and commercial benefits, especially for actors with fewer resources. Openly released models also have a downside: once published they cannot be withdrawn, and their safeguards are easier to remove.

What is being done

Competition authorities such as the CMA are reviewing partnerships and investments in the AI sector. In the EU, providers of general-purpose AI models have had transparency and copyright obligations since August 2025, and the EU AI Office can request documentation, evaluate models, require corrective measures and impose fines. Amendments that entered into force in 2026 also give the AI Office centralised oversight of AI systems built on general-purpose models.

What you can do

  1. If your organisation depends on AI services, plan for outages and avoid building everything around a single supplier.
  2. Find out where your data is stored and processed, and what the terms say about how it may be used.
  3. Consider whether smaller, open or European alternatives meet your needs.
  4. Follow how competition authorities and the EU AI Office use their powers; their decisions shape the market.

Sources

  1. International AI Safety Report 2025 – International AI Safety Report (chair: Yoshua Bengio), 2025
  2. CMA outlines growing concerns in markets for AI Foundation Models – Competition and Markets Authority (UK), 2024
  3. International AI Safety Report 2026 – International AI Safety Report (chair: Yoshua Bengio), 2026
  4. AI Act – regulatory framework for AI – European Commission, 2026

Data updated: 2026-11-19

More risks

AI risks