AI computing power share by country
At the end of 2025, the United States had 81% of the computing power of the AI supercomputers Epoch AI knows of.
Source: Epoch AI2025Updated 7 Nov 2026
Map. Arrow keys move the map, plus and minus zoom, N and P step through the countries with data, Enter opens the one shown, Escape clears the selection.
Loading the map…
The map could not load.
Borders show who controls an area in practice, as drawn by Natural Earth. Disputed borders are dashed. Borders shown do not imply endorsement. How we treat borders
Summary
The highest: the United States 81%, China 6.9% and Germany 1.7%.
36 countries and territories have a figure for 2025. Counts only the AI supercomputers Epoch AI has documented in operation; many private systems are not included. Computing power is counted in H100-equivalents: how many of Nvidia’s H100 chips, a common chip for training AI, would do the same work.
Source: Epoch AI2025Updated 7 Nov 2026
Countries and territories with data
| United States | 81% |
|---|---|
| China | 6.9% |
| Germany | 1.7% |
| Norway | 1.3% |
| Japan | 1.2% |
| France | 1.2% |
| Switzerland | 1.1% |
| Finland | 0.8% |
| Saudi Arabia | 0.5% |
| Italy | 0.5% |
| United Kingdom | 0.5% |
| South Korea | 0.4% |
| Iceland | 0.3% |
| Sweden | 0.3% |
| Spain | 0.3% |
| India | 0.3% |
| Canada | 0.2% |
| Israel | 0.2% |
| Denmark | 0.2% |
| Thailand | 0.1% |
| Russia | 0.1% |
| United Arab Emirates | 0.1% |
| Taiwan | 0.1% |
| Brazil | 0.1% |
| Poland | 0.1% |
| Singapore | 0.1% |
| Netherlands | 0.1% |
| Australia | <0.1% |
| Hong Kong | <0.1% |
| Luxembourg | <0.1% |
| Czechia | <0.1% |
| Slovenia | <0.1% |
| Vietnam | <0.1% |
| Mexico | 0% |
| Malaysia | 0% |
| Philippines | 0% |
Source: Epoch AI2025Updated 7 Nov 2026
Largest known AI supercomputers
| Supercomputer | Country | H100 equivalents | Working since |
|---|---|---|---|
| xAI Colossus Memphis Phase 3xAI | United States | 275,796 | 2025 |
| xAI Colossus Memphis Phase 2xAI | United States | 200,000 | 2025 |
| xAI Colossus Memphis Phase 1xAI | United States | 100,000 | 2024 |
| Meta 100kMeta AI | United States | 100,000 | 2024 |
| OpenAI/Microsoft Goodyear ArizonaMicrosoft,OpenAI | United States | 100,000 | 2024 |
| Oracle OCI Supercluster H200sOracle | United States | 65,536 | 2024 |
| Tesla Cortex Phase 1Tesla | United States | 50,000 | 2024 |
| Lawrence Livermore NL El Capitan Phase 2US Department of Energy | United States | 44,143 | 2024 |
| CoreWeave H200sCoreWeave | United States | 42,000 | 2024 |
| Meta GenAI 2024bMeta AI | United States | 24,576 | 2024 |
| Meta GenAI 2024aMeta AI | United States | 24,576 | 2024 |
| "Jupiter, Jülich"EuroHPC JU,Julich Supercomputing Center | Germany | 23,536 | 2025 |
| Oracle OCI MI300xOracle | United States | 21,649 | 2024 |
| NexGen Cloud Hyperstack AQ Compute SupercomputerNexGen Cloud | Norway | 16,384 | 2024 |
| Oracle OCI Supercluster H100sOracle | United States | 16,384 | 2024 |
| Microsoft Azure EagleMicrosoft | United States | 14,400 | 2023 |
| NVIDIA CoreWeave Eos-DFW Phase 1Nvidia,CoreWeave | United States | 10,752 | 2023 |
| Alps Supercomputer Phase 2ETH Domain | Switzerland | 10,752 | 2024 |
| Oracle OCI Supercluster A100sOracle | United States | 10,332 | 2023 |
| Imbue 10k ClusterImbue | United States | 10,000 | 2023 |
Source: Epoch AI2025Updated 7 Nov 2026