More and more data points are indicating how China has all but caught up with the US in AI.
The latest comes from NeurIPS — the world’s premier machine learning conference — where China has overtaken the United States as the working location of first authors for the first time. According to data from NeurIPS and Paper Copilot, analysed by The Economist, China now accounts for 2,152 top AI researchers by this measure, compared to 1,810 for the US. The gap, while not vast, represents a symbolic and substantive inflection point: as recently as 2019, China trailed not just the US but also Britain.
The trend lines tell the more important story. In 2016, China ranked fourth — behind the US, Britain, and France. By 2025, it had climbed to first. The US, meanwhile, has slipped to second. Every other country in the top 10 — South Korea (244), Britain (230), Germany (210), Singapore (157) — trails both by a wide margin, making this effectively a two-horse race.
This shift in research geography aligns with a broader pattern. Jensen Huang has noted that 50% of the world’s AI researchers are Chinese and that China accounts for 70% of AI patents filed last year. Meanwhile, Databricks co-founder Andy Konwinski has observed that Western open science has “lost the number one spot to China,” with Stanford and Berkeley PhD students now reading more interesting papers from Chinese startups like DeepSeek and Moonshot AI than from American labs.
The researcher migration dynamic is also worth noting. For years, Chinese-origin researchers moved to the US for better salaries and more ambitious labs. That calculus is shifting. OpenAI has already lost a researcher to Tencent, as China’s domestic AI scene — with companies like DeepSeek, Alibaba, ByteDance, and Moonshot AI releasing competitive models — now offers genuine frontier opportunities at home.
The output from those researchers is increasingly hard to dismiss. Chinese open-source models have overtaken US models in global downloads for the first time, and 80% of startups seen by a16z are now using Chinese AI models. Google DeepMind CEO Demis Hassabis has acknowledged that the US lead may now be measured in months, not years.
The NeurIPS data is a lagging indicator — it reflects where researchers are working, not just where they were trained. That China leads on this metric suggests the talent is now building at home, not just being educated abroad. For US AI policy and the labs that depend on attracting global talent, that is the number worth watching.