A rare 10-day tour of China's leading artificial intelligence laboratories reveals an industry compensating for chip access restrictions through extreme work culture, rapid commercialization, and strategic product focus rather than pursuing raw capability parity with US rivals.
The delegation visited Unitree, MiniMax, Moonshot, Zhipu AI, Ant Group, Xiaomi, ModelScope, and Galbot across Beijing, Hangzhou, Shanghai, and Shenzhen, according to a detailed account published by the tour participants.
The Compute Constraint
The most pressing challenge emerged consistently across every lab: insufficient access to advanced chips. "Basically every AI researcher brought up the same complaint: their companies lack sufficient access to the advanced compute resources necessary to train and run AI models," the visitors reported.
While US companies deploy Nvidia's latest Blackwell chips, Chinese labs cannot legally obtain equivalent hardware due to export restrictions. The gap creates a widening technological disadvantage in raw model training capacity.
Working Around Restrictions
Rather than accepting this handicap, Chinese firms employed multiple strategies. The most visible: extreme work culture. Researchers worked through Labor Day holidays and weekends. Sleeping cots appeared throughout offices. "Caffeinated and sugared beverages peppered across every horizontal surface." One researcher described entering constant overtime after DeepSeek's R1 model release in January created competitive pressure.
