China's cyberspace regulator (CAC), citing a CCTV report, says the country has built and put into use more than 70 embodied-AI training grounds, with 40-plus more under construction or planned, clustered in the Yangtze River Delta, Beijing-Tianjin-Hebei and the Pearl River Delta. An embodied-data platform based in Zhongguancun is said to have delivered 1.5 million hours of human-video data; in Shenzhen, a wheeled-arm humanoid trained on that data handles tasks such as switching electrical cabinets and opening industrial valves. The report also cites 2025 figures — core AI industry above 1.2 trillion yuan with more than 6,200 companies — and a push by the industry ministry and others to close the loop from real-world training to scaled deployment.
Our takeThe framing is the point: a regulator counting training grounds the way it counts infrastructure is describing capacity, not demos. That shifts the contest from who has the best robot footage to who owns the data and the sites to train on. Worth watching whether the training-ground model becomes a national standard other players must plug into.