DeepSeek AI: China Retains 70% of Mobile Researchers
Key insights
- 70.3% of internationally mobile DeepSeek researchers ended up in China, including most who spent five or more years in the U.S.
- 53.5% of researchers with known affiliations spent their entire careers at Chinese institutions, requiring no U.S. training pipeline.
- Median researcher citations more than doubled from 249 to 681 in one year, reflecting DeepSeek's rapid adoption across the AI field.
Why this matters
China's ability to develop frontier AI researchers entirely within its own institutions ends the strategic assumption that US training pipelines gave Washington leverage over Chinese AI development. The finding that US tenure duration did not predict whether internationally mobile researchers stayed in the US undercuts policy proposals built around extended visa programs or academic exchanges as retention tools. For AI founders and practitioners, the 4,108 median citations among US-affiliated DeepSeek researchers who mostly returned to China confirms that some of the most impactful people in the field are now doing frontier work entirely outside the US ecosystem.
Summary
A Hoover Institution update tracking DeepSeek's research team finds 70.3% of internationally mobile researchers ended up in China, most after five or more years in the U.S.
Amy Zegart and Emerson Johnston at Hoover and Stanford HAI expanded the study from 223 to 356 researchers across seven foundational DeepSeek papers, including V3.2 and V4. Of those with known affiliations, 53.5% spent their entire careers at Chinese institutions, never training in the U.S. at all.
Essentially: (DeepSeek, Chinese research institutions) are sustaining frontier AI development without US training pipelines.
- Median researcher citations more than doubled from 249 to 681 in one year, tracking rapid field adoption of DeepSeek's 2025-2026 work.
- The 80 researchers with any U.S. affiliation averaged 4,108 citations, the highest-cited group, but most are now based in China.
- A stable 31-member Key Team appeared on all seven papers, with their median citations rising from roughly 700 to 1,200.
The US faces two compounding losses: training high-impact researchers who return to China while China independently generates frontier AI talent without depending on US institutions to seed it.
Potential risks and opportunities
Risks
- US-affiliated DeepSeek researchers, averaging 4,108 citations and now mostly China-based, represent a concentration of frontier AI capability that US institutions and companies can no longer access or recruit from.
- The finding that US tenure duration did not predict researcher retention could invalidate extended-visa strategies as a policy lever, leaving US policymakers without a tested retention tool for international AI researchers.
- China's demonstrated capacity to produce frontier AI talent entirely within its own institutions removes a core assumption behind US chip-export controls, which presupposed that hardware restrictions would bind Beijing if talent pipelines ran through US universities.
Opportunities
- US AI labs and universities can use Zegart and Johnston's citation-mobility data to build targeted retention packages for high-impact researchers before relocation, using the 4,108-citation US-affiliated cohort as a concrete benchmark.
- Policy organizations focused on STEM visa reform gain a longitudinal, peer-reviewed dataset from Hoover Institution and Stanford HAI that could support evidence-based congressional proposals on researcher retention.
- AI companies expanding internationally, particularly in markets outside the US-China talent competition, can use these talent-flow findings to identify high-quality researcher communities not yet consolidated around Beijing or Silicon Valley.
What we don't know yet
- Whether specific US immigration policies or research-funding conditions affected return decisions among the 13 researchers who spent five or more years in the U.S.
- Whether the seven published DeepSeek papers capture the full scope of the company's research output, or whether a parallel internal pipeline exists outside this tracked dataset.
- What institutional affiliations the stable 31-member Key Team currently holds, and whether any retain active US academic appointments as of mid-2026.
Originally reported by hoover.org
Read the original article →Original headline: Hoover Institution Update: DeepSeek's Research Team Grew to 356 Authors Across Seven Papers, 70% of Internationally Mobile Researchers Settled Back in China — US Talent-Pipeline Gap Called 'Critical'