The real bottleneck holding back humanoid robotics is no longer computational power—it is the sheer volume of recorded real-world footage needed to train these systems. China has recognised this constraint and is responding with scale: more than 70 dedicated training grounds are already operational across the nation's regions, with another 46 in development or construction. Europe's closest equivalent is a single company's facilities.
China's National Data Administration announced plans to establish standards governing embodied AI training data and coordinate guidance to regional authorities, following a 10 September meeting led by Liu Liehong. The move came just ten days after seven industry players had approached the same regulator requesting precisely this: standardised data frameworks and public infrastructure for embodied AI development.
The scale of the challenge is substantial. According to the China Academy of Information and Communications Technology, embodied AI foundation models require approximately 10 million hours of authentic training footage. Current global supplies fall dramatically short, with estimates suggesting only between 100,000 and 1 million hours of high-quality data exist worldwide.
China's training ground network spans more than half the country's provincial regions, with roughly 86% focused on industrial manufacturing applications. Three geographic clusters—the Yangtze River Delta, the Beijing-Tianjin-Hebei region, and the Pearl River Delta—form the backbone of this infrastructure.
Europe's regulatory approach without the data infrastructure
The European Union has taken a different path: establishing comprehensive rules without yet building the systems to generate training data at scale. The Data Act entered force on 12 September 2025, determining which parties may access information generated by connected devices and industrial equipment. The legislation also prevents that data from being repurposed to develop competing products, with design requirements applying to connected products entering the EU market from 12 September 2026 onwards.
Europe's nearest equivalent to a Chinese training ground is NEURA Robotics, a private company constructing robotic training facilities. The firm plans ten locations total, with five expected operational by year's end, distributed across Europe, the United States, and China.
The European Commission released a Data Union Strategy on 19 November 2025 that prioritises expanding data access for artificial intelligence development. However, ten months later, most planned initiatives remain incomplete, including the establishment of dedicated data laboratories.
Despite China's infrastructure advantage, the market remains unsettled. Only 23% of Chinese enterprises surveyed this year expressed satisfaction with available robotic solutions, and training grounds accumulate footage rather than generate immediate commercial returns.
Source: The Next Web
