Documented AI compute sits almost entirely in one country: 977,475 H100-equivalents verified in the United States, none elsewhere in the excerpts

Asked:

“For each country in the world show available for future use energy sources (gw) and available chips for AI to estimate chances to win AI race. Show on map. Make it beatiful”

Best-effort map from the verified rows available in the supplied source excerpts: 9 operational GPU clusters from Epoch AI’s GPU Clusters database (all in the United States, latest first-operational date July 2025), plant-level nameplate capacity from the WRI Global Power Plant Database excerpt (Afghanistan complete at 0.30 GW, Albania partial), and 2024–2025 electricity generation context from Ember via Wikipedia. Every country stays on the map; those without verified figures are gray, never zero.

⚠ Coverage badge — partial extracts: the Epoch CSV and WRI excerpts supplied here contain only their opening rows, not the full files. This page is an infrastructure proxy, not a complete inventory and not a forecast of who “wins”.

AI infrastructure readiness — verified evidence per country

documented AI chips (cyan→violet, readiness signal) verified installed power capacity (GW) only no verified data in supplied excerpts

All 977,475 documented H100-equivalents (932,080 physical AI chips across 9 operational clusters, ~1.33 GW of cluster power) are in the United States; the largest single site is xAI Colossus Memphis Phase 3 at 275,796 H100e. epoch.ai

Epoch AI records over 500 GPU clusters worldwide and locates 95% to at least country level — the 9 US rows here are only the top of the file, so other countries’ compute is under-represented, not absent. epoch.ai

The only complete country capacity picture in the WRI excerpt is Afghanistan: 0.30 GW across 9 plants — 0.24 GW hydro, 0.04 GW gas, 0.02 GW solar (2017 data). github.com/wri

For scale context, China generated 10,580 TWh of electricity in 2025 versus 4,520 TWh in the US — but national grid capacity is not the same as power contractually available to AI data centers. wikipedia.org

Leaderboard — countries with usable verified figures

AI infrastructure readiness score (0–100): 75% normalized log(1 + H100-equivalents) + 25% normalized log(1 + installed GW). A transparent infrastructure proxy, not a forecast. Compute is weighted more heavily because chips, not grid size, are the binding constraint. Only one country has verified chips in these excerpts, so cross-country normalization is degenerate; scores are shown as partial.

Context — electricity generation, top 15 producers

Annual generation in TWh (Ember, 2024–2025). Shown separately: generation (TWh) is not installed capacity (GW), and the two are never mixed in the score.

All countries represented in the evidence

Country	Installed GW (verified)	Energy breakdown	AI chips	H100 equiv.	Clusters	Latest date	Source

Documented operational clusters behind the US figure

Cluster	Owner	H100 equiv.	AI chips	Power (MW)	Operational

Method: built from four supplied result sets — the opening excerpt of Epoch AI’s gpu_clusters.csv (9 complete operational rows, all US, first-operational dates Sep 2024–Jul 2025, H100-equivalents preferred over raw chip counts), the opening excerpt of the WRI Global Power Plant Database (plant nameplate MW ÷ 1,000 = GW; Afghanistan complete, Albania truncated mid-country and marked partial), Ember 2024–2025 generation tables via Wikipedia (TWh, context only), and Epoch documentation/metadata pages. A UNdata extraction covered 2024 but held only an Afghanistan shell row with no capacity value and is treated as missing, not zero. Countries without verified rows are gray “No verified data”. The readiness score is an infrastructure proxy — 75% compute, 25% power — and national capacity is not power contractually available to AI data centers. Cut for space: sources beyond the top 15 generation rows and unverified third-party country rankings.

This report was generated automatically by Keenable SELECT at a user's request, from publicly available web sources linked herein. Keenable does not review, verify, or endorse its contents and makes no representation as to accuracy, completeness, or timeliness; AI-based extraction may contain errors. Nothing in this report is investment, legal, financial, or other professional advice. All trademarks and referenced content remain the property of their respective owners; no affiliation or endorsement is implied. To report an error, rights concern, or request removal: legal@keenable.ai.