There is a sentence that foreign robotics engineers repeat with a mixture of admiration and frustration: the hard part of building a humanoid is not the design, it is getting the parts.
China's advantage is that it already has the parts.
Key takeaways
- Vertical integration: leading Chinese humanoid makers design and manufacture their own joint modules, motors and actuators rather than buying them, removing the markup of third-party components and compressing iteration cycles.
- Price: Unitree's G1 education variant is listed between roughly 145,000 and 414,000 yuan — a fraction of comparable Western platforms — which is what turned it into the volume product of the category.
- Scale-up: humanoid production capacity of about 12,000 units a year today, planned to reach 50,000, with total all-category robot capacity of 190,000 at the Ningbo complex.
- Cluster depth: China has more than 6,000 AI companies and a full chain covering intelligent chips, compute clusters, model development and application scenarios.
- Deployment breadth: Chinese warehouse robots are already operating in British retail and logistics, and Chinese humanoids have appeared on Polish television and at international industry events.
Why manufacturing beats prototypes
Robotics is unusual among deep-tech fields in that the bottleneck is often mechanical, not algorithmic. A joint actuator that is 10% more torque-dense, or 20% cheaper, changes what a whole machine can do. Making that actuator in-house changes how fast you can find out whether it works.
This is the same pattern that made China the factory for consumer electronics, applied to a harder product. Where Western humanoid programmes frequently remain at prototype or pilot deployment stage, Chinese vendors have been able to move hardware to customers at volume — a distinction that outside analysis has repeatedly graded as the decisive difference.
It also changes the economics of failure. A company that can build 7,000 units in two months can afford to be wrong about a design.
The dexterity frontier
The remaining hardware problem is the hand. Hands are where humanoids lose to humans: too many degrees of freedom, too little space, too much contact complexity.
The Guardian reported that a new cohort of Chinese AI startups is using the country's manufacturing strengths and its appetite for embodied AI to build fully dexterous robotic hands — framing it as an attack on the field's hardest unsolved problem. Parallel work has produced clever hybrid form factors, such as a gimbal that merges robotic motion control with a phone body, turning a handset into a device that actively moves to track a subject.
The counter-argument
Two things could erode the advantage.
First, software maturity. Independent assessments consistently grade Chinese humanoid platforms as shipping hardware whose autonomy stack is still developmental, with advanced behaviour requiring third-party integration or customer engineering. Hardware leadership does not automatically convert into software leadership.
Second, the intelligence layer may not be defensible. If the models that control robots become commoditised — and Chinese open models are aggressively priced — then the differentiating asset is the machine's body and its cost, which is precisely where China is strongest. But it also means a competitor with equal access to those models can compete on integration and trust rather than silicon.
What the outside analysts are saying
The most interesting external framing is not about benchmarks at all. Forbes argued that the winner of the AI race may not be the country with the highest benchmark scores but the one that makes intelligence cheap to deploy. Stanford's 2026 AI Index was quoted saying that China's strongest AI path is the combination of research goals, industrial capacity, sensible pricing and coordinated deployment.
That is a supply-chain argument, not a model-quality argument. And it is the argument China is currently winning.
Based on company disclosures, third-party market estimates and international media reporting in 2026. Capacity figures are planned or reported targets.