Ask people around the world who is winning the AI race and a striking number now answer: China. In a new survey of more than 18,000 people across 15 countries by the London consultancy Public First, clear majorities in 11 of them said they believe China has pulled ahead of the United States on AI capability and innovation. The list of countries that ranked China first is the uncomfortable part for Washington. It includes Canada, Britain and France, three of America's closest allies.
This is a verdict on perception, not a measurement of the technology, and the two are not the same thing. But perception has its own gravity, and in this case it tracks something real. Stanford's AI Index reported this year that the gap between the best US and Chinese models has narrowed to a few percentage points on the hard benchmarks. The public appears to have absorbed two years of headlines about cheap, capable Chinese systems and drawn the obvious conclusion. The view held across rich economies such as Germany and Japan and emerging ones such as Brazil, India and Nigeria alike.
Then comes the twist that complicates the headline. Being seen as the most capable is not the same as being the most trusted. When Public First asked whose models people actually trusted with their data, the ranking rearranged itself. People trusted their own country's models first. After that, Japan came out as the most trusted source overall, with the US just behind. China, for all its perceived lead on raw capability, did not top the trust list at all. Instead it split the room, drawing both strong trust and strong distrust in unusual measure.
The Americans should not feel smug about second place. Nineteen per cent of non-Americans said they did not trust a US model either, a reminder that suspicion of foreign AI is not aimed only at Beijing. The real signal sits in the gap between admiration and trust. People can believe China builds the most impressive systems in the world and still have no intention of handing those systems their personal data. More than three-quarters of respondents said it mattered to them that their data stay inside their own country, a loud vote for digital sovereignty.
What makes the findings genuinely interesting is how little this stated distrust governs actual behaviour. Of the people who said they did not trust US models, most were using ChatGPT, Claude or Gemini anyway. The same respondents who believe China is ahead, and who want their data kept at home, still reach for whichever assistant works best when there is a task in front of them. Belief, preference and habit are pulling in three different directions at once.
That disconnect is the part policymakers should sit with. Export controls and chip bans are built on the premise of defending an American lead. If publics in allied capitals already believe that lead is gone, the political ground under those measures is softer than it looks. Capability, trust and convenience are turning out to be separate races, and so far no single country is winning all three.