China's AI research output surpasses the US in published papers, but closed US labs skew the picture
Apr 6, 2026
Key Points
- China surpassed the US in AI research papers presented at top conferences in 2025, a milestone driven by US frontier labs moving research behind closed doors.
- US-based researchers at OpenAI, DeepMink, Anthropic, and similar labs have stopped publishing as work consolidated into private companies around 2021, making publication counts a lagging indicator of actual capability.
- China's open-source research ecosystem continues generating cross-pollination benefits, but unclear departures of top talent at Kuaishou suggest shifting dynamics in how China manages its researchers.
Summary
China's lead in AI research paper authorship is real but obscured by a structural shift in how the US produces knowledge. Jensen Wong at Nvidia acknowledged the gap directly: in 2025, for the first time, more papers presented at the world's top AI conference had lead authors based in China rather than the US.
The catch is that this metric captures only published research. US-based AI researchers at frontier labs—OpenAI, DeepMind, Anthropic, and others—have largely stopped publishing as their work moved behind closed doors. Between 2015 and 2020, the number of active AI researchers in the US grew sharply; it plateaued around 2021 as the best talent consolidated into private labs. China's research ecosystem remains largely open-source, so Chinese researchers continue publishing at scale.
Ilya Sutskever at Safe Superintelligence exemplifies the US trend: he does not publish and operates with deliberate secrecy.
The tension is whether published volume still signals genuine competitive advantage. One argument holds that open research networks create valuable cross-pollination and generate the kind of bizarre, exploratory ideas that can seed breakthroughs. China benefits from that pipeline. But the counterpoint is blunt: if the frontier work happens in closed US labs, publication counts become a lagging indicator of actual capability, not a leading one.
Another signal: something appears to be shifting in how China treats its own top researchers. The reorganization at Kuaishou, where extremely talented researchers departed under unclear circumstances, hints that China's approach to talent management may differ from the US model—though the full picture remains unclear.