China has unveiled its largest domestic data training center for humanoid robots in Beijing [1, 2].
The facility marks a strategic shift in robotics development, prioritizing the software and data needed for robots to perform useful tasks over mere mechanical design.
Opened in 2025 [2], the center spans approximately 10,000 square meters [1]. The facility is designed to address the gap between robot hardware and real-world application. Zhu Kai, the center's general manager, said the current inability of robots to perform helpful work stems from a lack of learning data.
"The future development of robots will be determined by data, not hardware," Zhu said [1].
Industry projections for the sector show rapid growth, though estimates vary. One U.S. research firm reported that global humanoid shipments reached approximately 19,100 units in the first half of 2026 [1]. Meanwhile, analysts at Morgan Stanley have projected that shipments from Chinese manufacturers alone could reach 50,000 units in 2026 [3].
The Beijing center focuses on the accumulation of diverse datasets to improve robot dexterity and decision-making. This includes training robots to handle everyday objects, such as condiment bottles, and brushes, to refine their interaction with human environments [1].
Zhu said that without sufficient data, hardware improvements alone cannot make robots functional for the general public [1]. The center aims to provide the necessary scale of information to bridge this gap, positioning China as a leader in the humanoid supply chain.
“"The future development of robots will be determined by data, not hardware,"”
The establishment of a massive, dedicated training center suggests that the robotics industry has hit a 'data wall.' While the physical shells of humanoid robots have become sophisticated, the intelligence required to navigate human spaces depends on massive amounts of proprietary training data. By centralizing this process, China is attempting to create a data moat that could accelerate the commercialization of robots faster than competitors relying on fragmented data collection.
