Eleven of the fifteen countries surveyed by the UK research firm Public First now say China has passed the United States in artificial intelligence, a perception gap that has opened while US labs keep posting the strongest commercial numbers in the field. The Public First AI Global 2026 report covers more than 18,000 respondents across 15 countries: eight advanced economies (Canada, France, Germany, Japan, Poland, Singapore, the UK, and the US) and seven emerging markets (Brazil, India, Indonesia, Mexico, Nigeria, South Africa, and Vietnam). Eleven of the fifteen pick China as the AI leader. Only Japan, India, Vietnam, and the United States itself still side with the US.
The split is sharper among US allies than the headline number suggests. In Canada, 40% of respondents rate China as the AI leader against 27% for the United States. The UK lands at 44% to 26%. Mexico tilts hardest of the close partners, with 49% naming China and 36% naming the US. France mirrors Canada almost exactly. Across the non-American sample, 19% say they do not trust an American-built AI model, a number the report flags as material: where a model is made is now a buying decision, not a footnote.
What the world is actually measuring when it rates the AI race is the part the leaderboard misses. Public First tracks three things at once: perceived capability, trust in the model, and confidence that AI is being deployed for public benefit. The US still wins on the first axis. Anthropic's revenue grew roughly tenfold year over year, and OpenAI's revenue more than doubled. Data-center construction is one of the few persistent tailwinds in US gross domestic product. On the second and third axes, the US has been retreating. Net sentiment toward AI in the US and UK has fallen roughly 75% over the past two years, and Americans are the only advanced-economy population to view new data centers on net negatively. Every other surveyed advanced economy is net positive on build-out.
Brookings frames the divergence as a strategic split rather than a temporary mood. The US model is resource-heavy, corporate-consolidated, and concentrated in a handful of frontier labs. China's model is, in the think tank's framing, people-centric, intensely competitive among domestic firms, and tightly regulated by the state. Both approaches can produce capable models. Only one of them is producing a population that wants the model deployed in its schools, hospitals, and courts.
That is why the US self-rating is the hinge of the story, not a side note. Fifty-one percent of Americans still say the US leads in AI, 24% say China leads, and 25% say they do not know. A country that is one bad earnings cycle away from a plurality saying it has fallen behind is not a country that has won the perception race, no matter how many frontier-model releases its labs ship. Nearly half of Americans are not confident in US AI primacy, and the rest of the Public First sample is less confident still.
The straight reading of the poll is that "who is winning AI" is now at least three questions, and the US is ahead in one of them. The strategic choice for the next policy cycle is whether Washington treats the trust and deployment gap as a problem to solve, or accepts further isolation as the cost of a model the world increasingly does not want to import.