Unitree Robotics (China)'s humanoid robot has attracted global attention when performing top martial arts moves at New Year's programs in this country.
Besides Unitree Robotics' robot product, Chinese phone manufacturer Honor is also preparing to launch its first humanoid robot at the MWC exhibition in Spain from March 2-5, 2026.
Robots are a priority area in the "Made in China 2025" strategy, although initially focusing on factory automation rather than humanoid robots. However, rapid advances in multi-mode AI are promoting "AI embodiment", i.e., self-operating machines in the real world. Chinese officials believe that this trend may help compensate for labor shortages and improve productivity.
At the starting point, Chinese companies are surpassing US competitors in speed and output. Selina Xu, a Chinese AI policy expert, said that China possesses a strong hardware supply chain, largely built from the electric vehicle industry, from sensors to batteries, along with a world-leading manufacturing platform. This allows Chinese businesses to improve products faster than the West.
As a result, Chinese robots are not only cheaper but also launched on the market at a faster pace. Ms. Xu said that Unitree alone shipped about 36 times more robots than its two US rivals Figure and Tesla last year.
According to Forbes' report, the total number of humanoid robots shipped globally last year was only 13,317 units, a very small number compared to market forecasts that could reach 2.6 million units by 2035. However, the report also noted that it is not clear how many of them are actually commercially sold or just demonstration models, according to Techcrunch.
By 2025, the leading Chinese companies in output include Agibot and Unitree, followed by UBTech, Leju Robotics, Engine AI and Fourier Intelligence, showing an initial advantage leaning towards Beijing.
Although hardware is progressing rapidly, software and AI are still bottlenecks. Companies are betting on vision-language-action models and "world models", but technology is still in its early stages. "World models" are systems that help robots understand and predict how things move in the real world, to act correctly and safely.
The big challenge is the lack of practical training data. Unlike large language models that can collect data from the Internet, human-shaped robots need physical data from the real environment. Therefore, many companies have to rely on simulations, while still needing to collect data in real life to complete the "brain" for robots.
According to Ms. Yuli Zhao, Strategy Director of Galbot Human-shaped Robot Company, China's biggest advantage is the speed of scale expansion from research, production to practical deployment taking place in a very short loop, helping businesses quickly test, adjust and commercialize products.
