In 2025, global humanoid robot shipments reached 13,318 units. Chinese companies produced 87% of them. AgiBot led with 5,168 units. Unitree Robotics ranked second. UBTECH followed. These are not prototypes. These are deployed machines working in real industrial environments.
For context: 13,000 humanoid robots in one year was considered impossible in 2022. The machines were too slow, too fragile, too expensive. Three engineering decisions changed that fast.
Stella’s Analysis
First: the actuator supply chain. Brushless DC motors with harmonic drives are now manufactured at scale inside China. The supply chain for precision actuators — bearings, encoders, torque sensors — was previously dominated by European and Japanese suppliers. Chinese manufacturers vertically integrated. They built it themselves. Unit costs dropped 40 to 60 percent in two years.
Second: AI training strategy. Instead of solving general-purpose manipulation, Chinese teams focused on narrow, repeatable tasks: box handling, connector insertion, part inspection. AgiBot ran 100 robots simultaneously to collect data and published a dataset of over 1 million action trajectories. This is the principle Elon Musk applied to Tesla manufacturing: constrain the problem tightly, then scale aggressively.
Third: ship-and-iterate discipline. CATL’s Xiaomo robot achieves 99% success on connector insertion. But 1% failure on a line running 300 cycles per day is still 3 failures. The engineering answer is redundancy plus real-time monitoring, not waiting for perfection. Chinese manufacturers shipped and fixed in production. Western teams are still in the lab.
Industry Impact
Chinese EV factories have started replacing human workers on specific stations: battery module feeding, harness routing, quality scanning. These are repetitive, ergonomically demanding tasks. Robots don’t get RSI. They don’t call in sick. On a three-shift line, the economics work at current robot prices.
The second-order effect is subtler. Factory layout is designed around human reach envelopes of 600 to 700 mm. Humanoid robots operating in the same space work with existing fixtures without line redesign. That is the real cost advantage. We explored this in our analysis of robot body business and industrial cost transformation.
Beyond automotive, logistics companies are deploying humanoids in sorting centers. Retail and hospitality are trialing them for service operations. The application map is expanding fast.
Trends
Global output will at least double in 2026. Unit costs are falling. Goldman Sachs estimates humanoid robot costs could drop below $30,000 by 2030. Chinese producers are already below $50,000 for some configurations in 2025.
The next battleground is software. Hardware parity is nearly achieved among the top Chinese makers. The differentiator will be task libraries: how many industrial operations a single robot can perform without custom programming. Whoever builds the biggest task dataset wins the deployment contracts.
For hardware limits in specialized applications, see our review of specialists in high-speed and high-payload robotics. For quality integration, read innovations in China’s industrial inspection landscape.
