AI computing power
Installed AI-supercomputer capacity by region, the hardware floor under every AI ambition.
Installed AI compute · H100-equivalent GPUs
US 1251k · China 109k · EU 80k · Epoch AI
Only verifiable clusters are counted; coverage of China is partial.
Verified large AI clusters only; coverage of China is partial, so its true capacity is higher than shown.
The story
Artificial intelligence runs on computing power. Training and deploying advanced models takes enormous amounts of specialised hardware, which makes AI compute one of the most important inputs to frontier innovation. More compute means larger models, bigger research programmes and faster routes from idea to market.
This indicator measures AI computing capacity relative to population: a read on a country's ability to take part in the next generation of AI development. Europe has expanded its AI infrastructure in recent years. The United States has built a far larger computing base, and that gives its researchers and startups faster access to the resources that decide who competes at the frontier.
Why it matters
Compute has become a strategic resource for the digital economy. Access to it increasingly determines where frontier research happens, where startups locate and where new AI companies can scale.
As AI spreads through healthcare, manufacturing, finance, defence and science, the economies with computing capacity are the ones positioned to turn it into productivity growth, investment and competitive technology sectors. The ones without it will import the results.
Policy context
AI infrastructure is now a central theme of Europe's competitiveness agenda. The recent initiatives recognise that AI leadership takes more than talent and research: it takes large-scale computing capacity, digital infrastructure and energy systems that can carry it.
Building the hardware is only half the job. The outcome depends on whether startups, researchers and businesses can reach those resources quickly and affordably, and on whether capital markets, energy costs, regulation and adoption let Europe convert AI capability into growth and industrial leadership.
How this number is calculated
Epoch AI's openly licensed dataset of verified large AI clusters, aggregated as H100-equivalent GPUs per region. Only verifiable installations are counted, which means China's true capacity is higher than shown; for that reason this metric is displayed but not yet included in the Momentum Index.
Source & cadence
Related policy insights
No insight references this indicator yet; new analyses appear here automatically.
Browse insights