The world’s most capable AI models come from a startlingly small number of places.
Two countries build almost everything at the frontier, and the gap between them is measured in months, not years.
TL;DR
- Over half of the world’s large-scale AI models were developed in the United States, about a quarter in China, followed by the EU and the UK.
- In 2024, U.S. institutions produced 40 notable AI models, versus 15 in China and three in Europe.
- Every model at the capability frontier since 2023 has come from the United States, with Chinese models trailing by about seven months on average.
The model counts here come from Epoch AI’s AI Models database and the Stanford AI Index, the two standard trackers of frontier AI development.
The concentration of AI power
| wdt_ID | wdt_created_by | wdt_created_at | wdt_last_edited_by | wdt_last_edited_at | Country or bloc | Standing in frontier AI |
|---|---|---|---|---|---|---|
| 1 | emmanuel-ashemiriogwa | 04/08/2026 01:44 PM | emmanuel-ashemiriogwa | 04/08/2026 01:44 PM | United States | Most large-scale models; every frontier model since 2023 |
| 2 | emmanuel-ashemiriogwa | 04/08/2026 01:44 PM | emmanuel-ashemiriogwa | 04/08/2026 01:44 PM | China | About a quarter of large-scale models, share rising |
| 3 | emmanuel-ashemiriogwa | 04/08/2026 01:44 PM | emmanuel-ashemiriogwa | 04/08/2026 01:44 PM | European Union | Trails, tens of large-scale models |
| 4 | emmanuel-ashemiriogwa | 04/08/2026 01:44 PM | emmanuel-ashemiriogwa | 04/08/2026 01:44 PM | United Kingdom | Home base of DeepMind's UK work |
The gap between U.S. and Chinese models ranges from four to fourteen months, and no Chinese model had matched OpenAI’s o3 as of early 2026.
But the race is close and moving. U.S. and Chinese models have traded the lead multiple times since early 2025;
DeepSeek-R1 briefly matched the top U.S. model in February 2025, and by March 2026 the leading U.S. model was ahead by just 2.7%.
Who Stands Out?
The United States holds roughly three-quarters of global GPU cluster performance, which is the raw ingredient of frontier training.
That hardware advantage, reinforced by export controls on advanced chips, is the main reason American labs stay at the front.
The United Kingdom’s contribution runs largely through DeepMind, and Canada punches above its weight in foundational research through institutions like the Vector Institute, but neither builds frontier models at the scale of the U.S. or China.
The frontier is, for now, a two-country contest with everyone else contributing pieces.
Why concentration matters
When almost every frontier model comes from two countries, the direction of the entire technology, its safety norms, its languages, its built-in assumptions, is set by a handful of labs. That concentration is a strategic fact, not a footnote.
ELI5
The most powerful AI systems in the world are built by just a few countries, mostly the United States and China. The U.S. is a little bit ahead right now, but China keeps catching up within months
.
Sources
