Supply of first-person video is increasing and prices are falling, but teleoperation data has a higher usable yield, while its collection is constrained by equipment and capacity. Nomura believes near-term value is concentrated in data validation, integration and contact-data collection.
Zheshang Securities is optimistic about demand for robot training data collection and supply-chain companies with existing orders; Huatai Securities sees improvements in Yili’s liquid milk and milk powder businesses, but says a higher tax burden depressed its net margin in the first three quarters.
Demand for embodied AI data is driving collection services, while edge AI is constrained by model size and memory bandwidth. Institutional views on dairy, pharmacies and traditional Chinese medicine center on cyclical improvement and defensive positioning, respectively.
Frontier models are improving robots’ ability to adapt to new tasks, but high-precision perception, real-time control and data collection continue to constrain commercialization. Goldman Sachs believes robots’ competitive advantages will increasingly shift toward hardware-software integration, with upstream sensing and data collection potentially benefiting earlier.
Agibot has launched a robot pre-training data engine, with tests showing that it can increase effective training data by more than 40 times. Huayuan Securities believes data providers’ competitive barriers will extend from collection to action conversion and training infrastructure.
Guosheng Securities is optimistic about demand for IMU data collection driven by robot training, while Changjiang Securities focuses on steelmakers’ cost advantages. Competition for the AI device interface is intensifying, while lower screen production costs are shifting value toward IP and distribution.
Industrial Securities believes there is still a gap between the mass production of humanoid robots and the maturity of embodied intelligence, and that real-world data could become a scarce resource for breakthroughs in physical AI. An industry-chain estimate of a data-collection market exceeding 60 billion yuan depends on two assumptions: the scale of the robot hardware market in 2030 and the share of costs attributable to data collection.
