The study was conducted by the think tank Carnegie Endowment for International Peace. The results show that among every 30 Chinese researchers engaged in artificial intelligence research in the United States in 2025, only one will return to China.
Ma Yang, who led the study, said: "Despite the tense relations between the United States and China, the number of researchers originally from China in the United States has not actually declined, but has increased by four percentage points."
The study analyzed papers accepted by NeurIPS last year to determine the flow of global scientists. NeurIPS is an academic conference focusing on machine learning and computational neuroscience. Ma Yang and his team analyzed where the authors attended university, received their graduate degrees, and ultimately worked.
The findings confirm a common belief: The United States and China produce most cutting-edge AI research. The study found that while China has failed to recruit foreign researchers, it has made great gains in retaining local talent, who have grown up through the domestic education system and are increasingly inclined not to work overseas. The United States, on the other hand, has stayed ahead by hiring foreign researchers, many of whom had previously immigrated from China, India, South Korea, and elsewhere.
The Carnegie study counted researchers who moved from one country to another, but did not specify when they moved. One explanation for why the number of Chinese scientists working in the United States remains high is that they arrived before the Trump administration last year imposed visa restrictions and cut research funding.
“There is a delayed effect — most of the researchers in this study had made their decision to come to the United States to study or work years ago,” said Yingyi Ma, a sociology professor at Syracuse University who studies Asian students in the United States. “The danger in the United States is that its attractiveness to talent is slowly eroding, and this erosion will accumulate over decades. By the time the effects show up in the data, it will be difficult to reverse.”
For years, Chinese scientists and their counterparts around the world have been attracted to the United States for its career opportunities in start-ups and ample academic funding. But in recent years, China has made significant progress in becoming an alternative for AI researchers.
In 2022, only six of the top 30 institutions publishing papers on NeurIPS will be in China; in 2025, this number will rise to 12. Two of China's top universities, Peking University and Tsinghua University, topped the list, overtaking heavyweight American institutions such as MIT, Stanford University and Carnegie Mellon University. Last year, Chinese researchers accounted for 41% of authors published by NeurIPS, exceeding 34% of American researchers.
Kevin Xu, founder of Internet Capital, a hedge fund that invests in artificial intelligence technology, said China now has a large number of workplaces for artificial intelligence researchers, including well-funded universities, large technology companies and start-ups. As Chinese companies such as DeepSeek and Dark Side of the Moon have made a name for themselves on the global stage, they have become increasingly attractive to top talent.
"Being able to work at DeepSeek brings the same career prestige and future success as working at OpenAI, which is very different from what we observed five years ago, when the choice path was between going to Google or going to Tencent," he said.
The rise of Chinese companies may explain why the vast majority of Chinese AI researchers choose to stay at home. But Kevin Xu added that there is another reason why some researchers stay in China: the government is unwilling to let them go.
This year, the Chinese government restricted the two founders of artificial intelligence startup Manus, which was due to be acquired by Meta for $2 billion, from leaving the country. Since then, the travel of Chinese AI entrepreneurs and researchers — as well as their spouses and children — has been further restricted, Bloomberg recently reported.
