Update: DeepSeek AI and the Great Talent Competition

In April 2025, we analyzed the 223 researchers behind DeepSeek's first five papers and found that the company's talent ran deeper than the prevailing story allowed. This update extends the analysis to all seven of DeepSeek's foundational papers, adding V3.2 (December 2025) and V4 (April 2026). The author pool has grown to 356, but the findings from last year hold. More importantly, most have sharpened. DeepSeek's researchers are more cited and more credentialed than they were a year ago, and still over half of them have never held an affiliation outside China. Among those who did train in the United States, most returned home, and the length of their stay did not meaningfully predict whether they returned. The data points to two distinct problems for U.S. policy. The first is retention. American institutions train researchers who leave. The second is independence. China now produces, at an unprecedented scale, frontier AI researchers who never pass through the United States at all. Neither of these of these problems are likely to revolve without concerted policy changes. The question is no longer whether the United States can maintain a permanent talent advantage. The question is whether it can compete on the terms the data now describes, which means treating the comparative advantage of U.S. research institutions as something to be earned rather than assumed, and treating the domestic Chinese pipeline as a development to be acknowledged.