China’s Data Regulator to Develop Standards for Embodied AI Data
China’s National Data Administration has announced it will create standards for embodied AI data and guide local authorities on the matter, just ten days after seven companies requested such action. According to Bloomberg, the regulator will support companies investing in data resources.
This move follows a meeting chaired by Liu Liehong, head of the agency, on September 10th, attended by research institutes, technology companies, and humanoid robotics organizations. The industry is shifting towards more data-driven approaches, prompting this response from the regulator.
The Challenge: Data for Training Robots
The bottleneck in humanoid robotics is not model complexity anymore but the availability of real training data. Embodied AI foundation models require approximately 10 million hours of real training data, as estimated by the China Academy of Information and Communications Technology. Currently, only between 100,000 and 1 million high-quality hours of such data exist globally.
China’s Training Ground Initiative
China is rapidly expanding its embodied AI training grounds, with over 70 already operational across half of the country’s provinces, and 46 more planned or under construction. These grounds aim primarily at industrial manufacturing, concentrated in three key regions: the Yangtze River Delta, Beijing-Tianjin-Hebei, and Pearl River Delta.
Europe’s Data Regulations
In contrast, Europe has established rules for this data without significantly collecting it. The Data Act came into force on September 12, 2025, governing who can access the data generated by connected products like industrial machinery. It also prevents using that data to build competing products. While the Data Union Strategy published in November 2025 aims to increase AI data access, most of its actions remain unimplemented, including the promised data labs.
Despite these efforts, China is not yet securing the market. Only 23% of Chinese enterprises surveyed this year expressed satisfaction with available robots, and training grounds collect data rather than charging customers for it.