What is AI compute?

What the term covers

Canada's Sovereign AI Compute Strategy describes AI compute as the computing resources AI systems need. It names work such as processing data, running algorithms and training machine learning models. The UK Compute Roadmap, published in July 2025, describes compute more generally as processing power that comes from chips, sits in data centers and is reached through software.

Training compute is often counted in floating-point operations (FLOP), a count of calculations performed. A European Commission questions-and-answers page on the EU AI Act was last updated in September 2025. It says the law sets a threshold of 10^25 FLOP used in training to capture the most advanced general-purpose AI models. It cites a 2024 Epoch AI estimate that training at that scale costs tens of millions of euros.

Why it matters

A 2024 paper by Girish Sastry and 18 co-authors argues compute is an especially effective lever for governing AI. Its reasons: compute is detectable, excludable and quantifiable, and comes from a concentrated supply chain. Stanford's 2026 AI Index reports TSMC fabricates almost every leading AI chip. It says the United States hosts 5,427 data centers, more than 10 times as many as any other country.

Epoch AI, a research organization, estimates on a dashboard updated in February 2026 that the compute used to train frontier language models has grown about fivefold a year since 2020. The International Energy Agency reported in 2025 that data centers consumed about 415 terawatt-hours of electricity in 2024, around 1.5% of the world total. It projected about 945 terawatt-hours by 2030.

Where things stand in 2026

An OECD report said in February 2023 that no country yet had data on its national AI compute capacity or a targeted plan for it. Several governments have since published plans. Canada's strategy draws on 2 billion Canadian dollars over five years announced in its 2024 budget. The UK roadmap committed up to £2 billion through 2030, including more than £1 billion to expand its AI Research Resource twentyfold. GOV.UK labels the roadmap as published under the 2024 to 2026 Starmer government. In the European Union, the EuroHPC Joint Undertaking opened a procurement call on July 30, 2026, for up to seven AI gigafactories. Submissions are due November 12, 2026.

The AI Index describes state-backed investment in AI supercomputing as rising. It says model production remains concentrated in the United States and China.