China’s robot makers excel in hardware but lack the AI brains for general use

China produces roughly 90% of the world's humanoid robots, with about 12,800 to 13,000 units expected by 2025, but these machines lack the AI sophistication needed for general-purpose work, operating at only 30% to 50% of human efficiency in industrial settings. The sector has only 500,000 hours of training data against a need for 100 million hours, a 200x gap, according to the report. Chinese government investment in humanoid robotics reached at least $230 million in the first half of 2026, nearly quadrupling from $62 million a year earlier, while about 500 companies compete in the space.

Photo: Tima Miroshnichenko / Pexels China’s robot makers excel in hardware but lack the AI brains for general use Chinese factories are churning out nearly all of the world's humanoid robots, but getting them to actually think remains a very different problem. China now accounts for roughly 90% of the world’s humanoid robot production, with approximately 12,800 to 13,000 units rolling off assembly lines by 2025. The machines look impressive. They walk, they gesture, they sometimes dance at trade shows. What they don’t do particularly well is think. The country’s humanoid robot sector has become a manufacturing juggernaut, leveraging the same supply chain wizardry that turned it into the dominant force in electric vehicles. Some models retail for as little as $6,000 to $12,000, a fraction of what Western competitors charge. But beneath the sleek exteriors lies a persistent weakness: the artificial intelligence powering these machines isn’t sophisticated enough for the kind of adaptable, general-purpose work that would justify the hype. The hardware is easy, the brain is hard In industrial settings, Chinese humanoid robots currently operate at just 30% to 50% of human efficiency. That’s not a rounding error. It’s a fundamental gap that separates a demo-ready prototype from a commercially viable worker. The core issue is that most of these robots are limited to scripted tasks. They can repeat a motion they’ve been programmed to perform, but throw them into an unstructured environment, say a factory floor where layouts shift or parts arrive in slightly different orientations, and they struggle. Adaptability requires a type of AI called physical intelligence, and building it demands staggering amounts of training data. The sector currently has access to around 500,000 hours of high-quality training data. Developing robust physical intelligence requires roughly 100 million hours. That’s a gap of about 200x. Why China got the hardware right China’s dominance in humanoid robot hardware was built on top of the country’s electric vehicle supply chain, which already produces motors, sensors, batteries, and precision components at scale. The EV industry essentially subsidized the robotics sector’s bill of materials, with actuators, lithium batteries, and control systems already being mass-produced for cars. This cost advantage has attracted significant government attention. Investment in humanoid robotics from Chinese government sources soared to at least $230 million in the first half of 2026, nearly quadrupling from the $62 million deployed during the same period in 2025. The data bottleneck and the overcapacity risk Training data for physical intelligence requires robots operating in real-world environments, making mistakes, learning from those mistakes, and gradually building embodied knowledge. Simulation helps, but it has well-documented limitations when it comes to bridging the gap between virtual and physical reality. Meanwhile, the sheer volume of Chinese production raises a familiar specter: overcapacity. The country has a well-documented pattern of flooding sectors with subsidized production capacity before the underlying economics fully support it. Solar panels, EVs, and steel all followed this trajectory. With roughly 500 companies currently operating in China’s humanoid robot space, a shakeout feels almost inevitable. What to watch from here The near-term commercial path for Chinese humanoid robots likely runs through highly structured environments where tasks are repetitive and predictable. Warehouses with standardized layouts, simple assembly tasks, and controlled conditions represent the low-hanging fruit. These applications don’t require general intelligence, just reliable execution of defined movements. Disclosure: This article was edited by Editorial Team. For more information on how we create and review content, see our Editorial Policy https://cryptobriefing.com/editorial-policy/ .