A large share of core contributors at Chinese labs are current students, integrated as peers on the model team. OpenAI, Anthropic, and Cursor offer no internships at all, so that talent pool simply does not enter US frontier work.
Building a frontier model is a multi-objective optimization where someone's genuinely good idea gets shelved so the whole model wins. Cultures that reward researchers for fighting publicly for their contribution pay a quality tax for it.
Ask Chinese researchers about AI economics or how models should behave and the question lands as a category error, not evasion. Their system rewards building, and offers no path to scientist stardom the way US podcasts do.
Inside China the labs read as one ecosystem rather than warring tribes: broad respect for DeepSeek's research taste, shared wariness of ByteDance's Doubao as the country's only frontier closed lab, and routine movement between employers.
Fresh entrants are an advantage, not a gap. Having never internalized an earlier AI hype cycle, students adopt each new paradigm (mixture-of-expertsA model built from many specialist sub-networks where only a few activate per token, giving big-model capability at small-model running cost.Full definition → scaling, then RL scaling, then agents) without unlearning the last one.
Terms in this piece · Glossary
mixture-of-experts — A model built from many specialist sub-networks where only a few activate per token, giving big-model capability at small-model running cost.
Why it matters
A rare insider account of how China's top AI labs actually operate — their student-heavy teams, fast-follower dynamics, and 'build-not-buy' data mentality — giving practitioners tracking frontier models real ground truth beyond secondhand analysis.
Key quotes
“All of the Chinese labs fear Bytedance with their popular Doubao model, which is the only frontier closed lab in China.”
“Ego and desires for career advancement do get in the way of making the best models.”
