US holds 72% of global AI compute; EU trails at 5%
What's the deal? The US has received 72% of the world's AI compute delivered since 2022, versus 5% for the EU and 10% for China, according to a new estimate from research group Epoch AIDealroom has a profile for this one. Try Dealroom →. Of the 27.4 million H100-equivalents that arrived from 2022 through June 2026, 19.8 million went to the US.
How was it measured? Epoch AI built its AI Chip Imports explorer from customs records across 16 major destinations, inferring AI hardware from server and parts trade. Chips made in China for Chinese data centers never cross a border, so the group added domestic production from its own sales data.
By the numbers: China ranks second at 2.6 million H100e, followed by the EU at 1.4 million and Malaysia at 0.7 million. No other country topped 2%.
The US lead is widening. It took 60–63% of each year's new compute from 2022 to 2024, then 72% in 2025 and 78% in the first half of 2026.
Rest of Europe: Outside the EU, Epoch's country table puts the UK at about 340,000 H100e (roughly 1.2% of the total), Norway at 130,000, Iceland at 96,000 and Switzerland at 37,000. Added to the EU's 1.4 million, that comes to roughly 2 million H100e, or about 7% of the world total.
The China caveat: China holds 18% of the AI chips in the data but only 10% of the compute, because much of its hardware is lower-performance. That includes export-compliant chips like Nvidia's H20 and domestic chips from HuaweiDealroom has a profile for this one. Try Dealroom → and Cambricon, which make up about 40% of China's total.
One caveat: Epoch treats the figures as estimates, not a census. The method infers hardware from trade records, and some counted chips may not yet be running.
Why it matters: A country that hosts large AI data centers can train and serve frontier models within its borders and gains influence over who uses them. That is why governments from the Gulf to Europe and India are competing to host capacity.
What's the EU doing? Europe has launched a tender for up to seven AI "gigafactories," backed by up to €10 billion in public funding and intended to attract at least €20 billion privately. Construction is only expected to start in 2027.
The signal: Access to large compute clusters is concentrating in the US, reinforced by heavy hyperscaler data-center spending. Europe's ambitions to develop and deploy advanced AI now depend on turning those plans into operating capacity.
Read more: epoch.ai
Data: Epoch AI chip imports explorer; chart shared by Samuel Albanie on X
Image credit: Epoch AI
