The Economist: 15% of GPT-5 Contributors Hold Degrees from Chinese Institutions
According to statistics from The Economist, 15% of the 483 contributors to OpenAI's GPT-5 project hold at least one degree from a Chinese institution. This list includes researchers as well as those in marketing, design, and leadership, not just pure paper authors.
The same article mentions that after Meta announces its superintelligence lab personnel around June 2025, a leaked list shows that half are described as coming from China. Subsequent English reports indicate that about 50% of the approximately 44 people on the list are from China, about 40% from OpenAI, about 20% from DeepMind, and about 15% from Scale, with approximately 75% being first-generation immigrants and about 75% holding PhDs.
NVIDIA CEO Jensen Huang has repeatedly stated that about 50% of AI researchers globally come from China, or that about 50% of developers are in the Chinese market. MacroPolo and others have estimated that the proportion of top talent starting in China is about 47%, which he rounded up to 50%.
The Economist tracked the educational trajectories of about 600 randomly selected papers and nearly 4,000 authors from NeurIPS 2025: in 2019, 29% of attending researchers started their careers in China, rising to about half by 2025; during the same period, the U.S. starting point dropped from 20% to 12%. Nine out of the top ten undergraduate institutions are in China, with Tsinghua accounting for about 4% and MIT for about 1%.
Among NeurIPS authors employed by U.S. institutions, about 35% hold undergraduate degrees from China, which is comparable to the proportion holding undergraduate degrees from the U.S. Fewer Chinese undergraduates are graduating and going abroad, with about three-fifths continuing their graduate studies in China by 2025.
In market mechanisms, the ones paying are the U.S. labs that need to stack cutting-edge models, while the ones selling are research employees trained in Chinese undergraduate programs. The event is driven by talent density, with funding flowing from visas and open research to high-priced signing bonuses and domestic talent retention subsidies; benefiting are universities that can supply undergraduates to both Chinese and U.S. labs, while policies that base team security on a single nationality assumption are under pressure.
Source: Public Information
ABAB AI Insight
U.S. leading labs do not cultivate an equivalent scale of undergraduate funnel themselves. Institutions like Tsinghua write training data into the NeurIPS author list, OpenAI incorporates some of them into the GPT-5 contributors, and Meta then poaches talent from OpenAI with salaries ranging from millions to hundreds of millions. The figures 15%, 50%, and 50% refer to the same pipeline: output from Chinese institutions, realized by U.S. labs.
The capital path is poaching, not expanding universities. About 40% of Meta's superintelligence lab comes from OpenAI, indicating that moving people between labs with dollars is faster than waiting ten years in campus settings. On the Chinese side, high salaries, housing, and research funding have increased the proportion of students staying in China for graduate studies from about one-third to about two-thirds, with the return rate rising from 12% to 28%. Money is competing for the same batch of undergraduate resumes.
This is comparable to the return of TSMC engineers in the semiconductor industry and physicists moving to U.S. national labs after the collapse of the Soviet Union. The industry is in a phase where the production location of talent and products is separated: first authors of papers are increasingly affiliated with Chinese institutions, while product releases are still predominantly from U.S. labs. Jensen Huang ties the developer market and researcher nationality together, fearing that export controls might push people and chips out of the U.S. stack.
Structurally, this belongs to the reconstruction of the industrial chain. The mechanism is: the constraints of model competition have shifted from GPUs to "who is still willing to put people in the same building." Tightening visa policies accelerate the retention of people in production locations for publishing papers, while high-priced poaching locks in those already in the U.S. into closed labs. Talent is no longer flowing unidirectionally to Silicon Valley but is split between open-source conferences and closed-source labs.
ABAB News · Cognitive Laws
- The origin of undergraduate education determines papers, while the origin of labs determines products.
- Poaching is faster than establishing schools, and also more expensive.
- The place where people stay is where the stack will be rebuilt.