InvestNot yet confirmed elsewhere1 publisher3 min readPublished
China overtakes US in top AI researchers, studies cite visa pushback and DeepSeek's homegrown talent
China hosted 41% of the world's top AI researchers in 2025 to 34% for the US, reversing their 2022 order, Carnegie China found. The study partly blames US visa rules, yet 53.5% of DeepSeek's authors never worked or studied abroad.
The Investor · Invest desk

What happened
- Chinese undergraduate degree holders staying in China rose to 69% in 2025 from 57% in 2022.
- Among DeepSeek researchers who gained US experience, more than 70% eventually returned home, a Hoover Institution and HAI analysis found.
- DeepSeek plans to double department sizes by mid-2026 on competitive salaries and will consider foreign researchers only if they speak Chinese.
- China's human resources ministry estimates its AI talent gap at more than 5 million professionals.
Why it matters
- contradiction The study names US visa rules as a cause, but most of DeepSeek's authors were never within reach of a US visa decision, while its core-contributor figure points back toward foreign training.
- cost American universities and labs pay to train the DeepSeek researchers who return, and the output of that training goes to a Chinese lab.
- decision Looser US visas would reach only researchers who passed through the US, a minority at DeepSeek, so any change in Washington addresses part of China's gain at most.
- constraint Because labs require Chinese, foreign hiring can fill little of a gap above 5 million, so domestic training has to supply most of what Chinese labs need.
China gained 14 percentage points of the elite between 2022 and 2025 [13] and the US lost 12 [14]. Together the two held 73% of NeurIPS-defined top researchers in 2022 and 75% in 2025 [15]. Everyone else slipped from 27% to 25% [15]. A 19-point American lead turned into a 7-point Chinese one [16].
Shares that mirror each other this closely invite a migration story. Carnegie China, though, counted participation in NeurIPS [4], so the share shows where elite work comes from. It cannot separate a researcher who came home from an American lab from one who never left China.
Three readings fit the same shares. One is the visa channel the study names, with researchers pushed back to China after time overseas [6]. Another is the competitive pay at leading Chinese firms that the study also names [6]. The third is a domestic pipeline that involves no visa decision at all, and the 12-point rise in Chinese undergraduates staying home [17] belongs to it.
DeepSeek is the one lab where the sources count heads. Stanford's Hoover Institution and HAI found that 53.5% of authors affiliated with it had never worked or studied outside China [1]. That leaves 46.5% with some foreign experience [12], only part of it American, and the analysis does not say how large that part is. More than 70% of the US-experienced group came home [8]. On that author list, visa policy can account for some fraction of the 46.5% and none of the majority.
The counter-case sits in the same research. One-third of the core contributors to a model Crypto Briefing describes as rivaling OpenAI's came out of China's domestic education system [9]. Beside 53.5% for all authors [1], the inner team looks less homegrown than the full author list, though the two figures use different tests. If the people who built that model mostly trained abroad, the visa reading is strongest among the people who matter most to the lab's best work. I think the evidence supports visa rules as a real but partial cause of China's 14-point gain [13]. A Carnegie breakdown showing that most of the gain came from researchers with US experience would prove that view wrong.
China, for its part, is spending on one pool. Foreign hiring runs into a Chinese working language, the Washington-Beijing rivalry and anti-foreign sentiment fed partly by youth unemployment, according to Crypto Briefing [11]. DeepSeek's openness to foreigners comes with a Chinese-language condition [2]. Its plan to double department sizes therefore draws on Chinese speakers, whether they stayed home or are returning from abroad [2]. Against a shortfall the labor ministry puts above 5 million [10], the only recruiting tool the sources show Chinese labs using is pay [6][2].
What to watch
- A Carnegie China breakdown of the 14-point gain between researchers with US experience and those who never left China.
- Whether DeepSeek meets its plan to double department sizes, and whether it relaxes the Chinese-language condition for foreign hires.
- Any change to US visa rules for AI researchers, which would test how much of the returnee flow policy actually controls.
Clarity's read
What the record supports and how the coverage leans. The claims behind it follow.
Reality
- Evidence40
- Adoption
- Insufficient
- Hype gap+20
- Incentives
- Insufficient
- Confidence40
Claim ledger
Ranked by verification strength, evidence, and original report placement.
- [1]
An analysis by Stanford's Hoover Institution and HAI found that 53.5% of authors affiliated with DeepSeek had never worked or studied outside China.
ReportedSupportedSource: Hoover Institution and HAI analysis, as reported by Crypto Briefing2 sources— create a free account to open themView cited source - [2]
DeepSeek plans to double department sizes across various roles by mid-2026, is offering competitive salaries to do it, and has signaled openness to foreign researchers on the condition that they speak Chinese.
ReportedSupportedSource: Crypto Briefing2 sources— create a free account to open themView cited source - [3]
A Carnegie China study published in September 2026 found that China hosted 41% of the world's top AI researchers in 2025.
- [4]
The Carnegie China study measured elite status by participation in NeurIPS.
- [5]
China's share of top AI researchers was 27% in 2022; the US share fell from 46% to 34% over the same period.
- [6]
The study points to two forces: leading Chinese firms offering competitive pay, and restrictive US visa policies that pushed researchers back to China after time spent overseas.
- [7]
The share of Chinese undergraduate degree holders who stayed in China rose to 69% in 2025 from 57% in 2022.
- [8]
Among DeepSeek researchers who gained US experience, more than 70% eventually came home.
ReportedSupportedSource: Hoover Institution and HAI analysis, as reported by Crypto BriefingView cited source - [9]
The same research found that one-third of the core contributors to an AI model rivaling OpenAI's capabilities came out of China's domestic education system.
ReportedSupportedSource: Hoover Institution and HAI analysis, as reported by Crypto BriefingView cited source - [10]
China's Ministry of Human Resources and Social Security estimates the country's AI talent gap exceeds 5 million professionals.
ReportedSupportedSource: Ministry of Human Resources and Social Security, as reported by Crypto BriefingView cited source - [11]
Barriers to recruiting foreign researchers include working in Chinese, US-China geopolitical tension, and rising anti-foreign sentiment fueled in part by local youth unemployment.
- [12]
46.5% of DeepSeek-affiliated authors had worked or studied outside China.
- [13]
China's share of top AI researchers rose 14 percentage points between 2022 and 2025.
- [14]
The US share of top AI researchers fell 12 percentage points between 2022 and 2025.
- [15]
The US and China together held 73% of top AI researchers in 2022 and 75% in 2025, leaving the rest of the world with 27% and then 25%.
- [16]
The US led China by 19 points in 2022; China led the US by 7 points in 2025.
- [17]
The share of Chinese undergraduate degree holders staying in China rose 12 percentage points from 2022 to 2025.
Sources
1 independent publisher whose own reporting we read for this story.
- cryptobriefing.comChina struggles to recruit foreign AI researchers even as it wins the domestic talent race
1 article · October 6, 2026
Topics and entities
Follow any of these and your For You feed starts watching them — no settings page required.
Topics
- AI talent migrationFollow
- US immigration and visa policyFollow
- US-China Open Model CompetitionFollow
Entities
- Carnegie ChinaFollow
- DeepSeekFollow
- Moonshot AIFollow
- Conference on Neural Information Processing SystemsFollow
- Hoover InstitutionFollow
- Stanford HAIFollow
- Ministry of Human Resources and Social SecurityFollow
- OpenAIFollow