According to a report from the China Academy of Information and Communications Technology (CAICT), as of the end of June 2026, there are over 70 training venues for embodied artificial intelligence operating in China, with more than 40 other facilities under construction or planning. These sites, sometimes referred to as 'robot schools,' are viewed as national infrastructure for collecting real-world interaction data, rather than as a singular commercial product.
Clusters of these venues are concentrated in several key regions: the Yangtze River Delta, the Beijing-Tianjin-Hebei region, and the Pearl River Delta. According to CAICT's figures, the highest number of locations (14) is registered in Zhejiang province, followed by Jiangsu, Beijing, Guangdong, and Shandong provinces, each having eight such facilities.
Although China is home to over 140 humanoid robot manufacturers and exported approximately 14,400 humanoid units in 2025—accounting for about eight out of ten global shipments—day-to-day reliability still depends on the scarcity of real-world collected data. Training venues address this gap by simulating repetitive tasks such as bottle grasping, tray loading, and parcel scanning, while motion capture operators create labeled trajectories.
In Guangdong province, a local venue functions as a matchmaking center between robot suppliers and users in healthcare or energy sectors. The national pilot base in Hangzhou is positioned as a common connection point for state-owned enterprises, technology companies, and application partners, as founders note significant cost savings compared to developing proprietary datasets independently.
Government policy has begun to align with this boom. In June 2026, the Ministry of Industry and Information Technology and the State Administration for State Assets Supervision and Administration launched a special initiative concerning the training of humanoid robots and embodied AI in real-world conditions. Under this initiative, each provincial region was tasked with identifying at least 20 priority scenarios and creating minimal-changeable, testable training spaces.
Concurrently, CAICT has developed an industry standard for embodied AI dataset quality, which is set to take effect on November 1st. This standard, developed in collaboration with over 40 organizations, aims to shift the approach to datasets—from prioritizing large-volume collection to emphasizing quality and evaluation methodologies.
Operators are also experimenting with business models that go beyond simply selling raw data. Among these is the 'training-as-a-service' model, which integrates data collection, model deployment, and application testing. For robotics teams, the immediate signals are the availability of specific production capacity and the approaching quality standard: dozens of open venues, a planned second wave of development, scenario quotas from MIIT/SASAC, and the November 1st dataset standard—these indicate infrastructure, not the presentation of a new chassis.