The AI race: US leads at the frontier, but China is gaining ground in adoption | Merics

There is a particular genre of think-tank document that exists to let everyone go home feeling represented, and MERICS has delivered a fine specimen: the U.S. leads at the AI frontier, but China is gaining ground in adoption. Two categories, two winners, no losers. It is the geopolitical equivalent of splitting the bill. One suspects the frontier is the bit with the enormous datacentres, and adoption is the bit where the model gets installed on a factory floor to count crates.

Which is not nothing — the commentary notes that Chinese open-weight models offer ready access without relying on a U.S. provider. Ready access. That is the register you use for a breakfast buffet, not for a technology stack, and it happens to be the precise property that every safety team with a research budget has spent two years writing fretful memos about. Weights that leave the building do not come back, and they do not fill in a landing card when they turn up somewhere new. Meanwhile the frontier lead that everyone has been so keen to quantify to the decimal turns out to be a lead in a race where the other side has quietly decided to be scored on installs.

And then the detail that deserves a paragraph of its own: all Chinese AI services require approval from the Cyberspace Administration of China. Every model, every chatbot, every set of 'readily accessible' open weights passes through a regulator's inbox first. The Cyberspace Administration has managed to build the thing Western legislators keep convening hearings about — a licensing regime with actual teeth — and the commentary files it as a footnote under adoption. One regulator, one queue, one answer, and no ambiguity about who is responsible when the thing misbehaves. Splendid. Our lot publish white papers about the distance between capability and law; theirs simply stamps the form and ships.