China Has More Than 150 Humanoid Robot Companies. How Many Can Survive?

China’s official count of more than 150 humanoid robot companies captures how quickly capital, engineering talent and supply chains have moved into the sector. It includes established robotics manufacturers, young startups and companies crossing over from automotive, electronics and artificial intelligence.

That number does not tell us how many firms have a finished product, a production line, a paying customer or a robot that can complete a shift with few interventions. A crowded field can accelerate component development and price competition while also producing many machines that look different in marketing material but solve the same narrow demonstration task.

A prototype can be assembled for an exhibition, trained for one controlled routine and repaired by the engineers who built it. A commercial product must arrive on schedule, pass site acceptance, work around existing equipment, survive repeated cycles and be serviceable by someone other than the founding team.

The strongest survival signal is not a new gait video. It is a customer ordering more units after operating the first batch. Repeat purchases reveal whether the robot creates enough value to cover integration, supervision, maintenance, charging, spare parts and software support.

Chinese manufacturers can draw on dense supplier networks for motors, reducers, batteries, cameras, processors, machined parts and electronics. Shorter supplier distances can reduce iteration time and make redesigns cheaper than in a fragmented supply chain.

Humanoid performance still depends on the complete system. A capable actuator does not guarantee stable locomotion. A dexterous hand does not guarantee reliable manipulation. Perception, calibration, thermal management, wiring, safety logic and task software must work together under the lighting, dust, floor conditions and human traffic of the deployment site.

Lower prices make pilots easier to approve, especially for research laboratories, universities and factories that want to test physical AI. The same price pressure can expose companies whose bill of materials, warranty costs and support model were never designed for sustainable margins.

A manufacturer can ship more robots and still lose money on every installation if engineers must remain on site, components fail early or the customer’s task changes. Survivors will need control over costs without removing the sensors, redundancy and service capacity required for safe operation.

Useful comparisons should track delivered units, paid deployment sites, productive hours, human interventions and repeat orders. These figures are harder to publish than a factory capacity target because they expose whether robots continue working after the launch event.

Failure data matters too. Mean time between failures, recovery time, damaged components and rejected task attempts determine whether a robot saves labor or creates a new supervision job. Public shipment claims should therefore be separated from installed units and from units performing productive work.

Not every company needs to become a full humanoid manufacturer. Some teams may survive by supplying hands, actuators, teleoperation systems, simulation tools, batteries or foundation models to several robot brands. Others may narrow their focus to inspection, logistics, research or hazardous environments.

The companies that remain will probably combine a defensible task, evidence from real sites, a maintainable machine and enough capital to support customers through several hardware generations. China’s company count shows the scale of the contest. It does not reveal the final winners.