In a 2025 sample of NeurIPS paper authors, the number of Chinese-origin AI researchers working in the United States was 30 times the number of US-trained researchers working in China. The ratio was 46 to one in 2022.
The finding comes from a Carnegie Endowment for International Peace study, “Who’s Ahead in the Global AI Talent Race?” It examined 10,280 authors whose papers were accepted at NeurIPS 2025, one of the most selective machine-learning conferences. Researchers compared where authors earned their undergraduate degrees, pursued further study and currently work.
China accounted for 57% of the sample’s undergraduate degrees, an 11-percentage-point increase from 2022. The US share was 13%. China was also the current workplace for 41% of the researchers, compared with 34% in the US. The United States recorded a net gain of 2,145 elite researchers, while China had a net loss of 1,729.
The study records education and current employment, but not when researchers moved. Its figures therefore do not fully capture the impact of recent student-visa restrictions, a $100,000 fee on new H-1B petitions or cuts to federal research funding. Most moves predated those policies, the researchers said.
China has improved its retention rate. The share of China-trained researchers now working in China rose to 69% from 57% in 2022. The US rate was 89%. South Korea retained 77% of its homegrown AI researchers, while Singapore recorded a net gain of 218 researchers.
The results describe a selective sample of NeurIPS authors, not the entire AI workforce or all researcher migration. Yingyi Ma, a Syracuse University sociologist, warned that a gradual decline in the US appeal to talent could build over decades and take time to appear in the data.