The AI Race
Country and company scorecards on the metrics that define the AI race: model quality, model velocity, training compute, investment, chips, patents, talent, and data-center power.
Country scorecard - share of globalStanford HAI · WIPO · Synergy
Each country's share of the global total on six measurable AI-race metrics. Darker bar = bigger share.
| Country | Private $ | Models | Talent | Patents | DC capacity | DC power |
Data-center power capacitySynergy '24
Top countries by total data-center capacity (GW). Compute = power; these are the physical plants where AI gets built.
Data-center electricity useIEA '24
Annual TWh consumed by data centers. 2026 projections shown as lighter bars.
Private AI investment (cumulative 2013–2024)Stanford HAI '25
Notable AI models released (2025)Stanford HAI '26
Top AI labs by training computeEpoch AI '26
Flagship-model training compute, log scale. Bigger = more compute thrown at the largest public model.
| Lab | Country | Flagship model | Training FLOPs | Log10(FLOPs) |
|---|
Best AI models by benchmark intelligenceArtificial Analysis '26
Frontier models ranked by composite benchmark score. Bar color = country of origin lab.
| # | Model | Lab | Country | Intelligence index |
|---|
Notable AI models released per year, globalEpoch AI '25
Top-tier AI talent concentrationMacroPolo '23
AI patent share (WIPO 2024)WIPO '24
Data sources + expand
| Metric | Series | Provider |
| Private $ | Cumulative private AI investment (2013–2024) | Stanford HAI - AI Index Report 2025 |
| Models | Notable Models database (model counts + training compute) | Epoch AI |
| Model quality | Frontier-model intelligence index (composite reasoning / coding / math evals, max reasoning settings; August 2026) | Artificial Analysis |
| DC power | National electricity consumption by data centers (2024 estimate) | IEA - Electricity 2024 (Data Centres chapter) |
| DC capacity | Hyperscale data-center capacity (quarterly) | Synergy Research Group |
| Patents | Global AI patent filings by country (2024) | WIPO |
| Talent | Top-tier researcher distribution (Global AI Talent Tracker, 2023) | MacroPolo |
| Training FLOPs | Frontier-model training compute (Epoch credible estimates) | Epoch AI |
Caveats: compute and chip figures are estimates - labs and chipmakers rarely disclose exact numbers. All values are static snapshots, not live. Last refreshed July 2026 (Stanford HAI AI Index Report 2026, Epoch AI mid-2026, IEA); the site's quarterly data routine re-checks it, with the next full refresh due with the AI Index Report 2027 (~April 2027).
About this page
A scorecard of the global artificial intelligence race: country-by-country standings on training compute, data-center buildout, model releases and research output, updated as the landscape moves.
For investors the AI race is above all a capital-expenditure story. The buildout of chips, data centers and power is one of the largest concentrated investment waves in market history, and it is reshaping index composition, electricity demand and the earnings of everything in the supply chain.
Frequently asked questions
Why does training compute matter?
Frontier model capability has so far scaled with the compute used to train it, which makes access to chips and power a hard constraint. Compute has become the oil of the AI era.
How does the AI race affect ordinary investors?
Through concentration. A handful of AI-linked mega-caps now drive a large share of index returns, so even a plain S&P 500 holder has a meaningful bet on the race's outcome.