China’s humanoid robots recently grabbed global attention with kung fu flips during the nation’s televised Spring Festival Gala. Meanwhile, Chinese phone maker Honor is set to unveil its first humanoid robot at the Mobile World Congress in Spain. Robotics was flagged as a priority under the country’s “Made in China 2025” plan, though it was originally focused on factory automation rather than humanoids. Now, rapid advances in multimodal AI are accelerating so-called embodied AI, which are autonomous machines operating in the real world. Officials say this push could help offset labor shortages and drive productivity gains.
At this early stage of humanoid robot development, Chinese companies are outpacing their U.S. rivals in both speed and volume, according to Selina Xu, a China and AI policy lead at the office of Eric Schmidt. Xu stated that China has a more robust hardware supply chain, much of it built up through the electric vehicle sector, and the world’s strongest manufacturing base. This allows companies to iterate far faster than Western competitors. As a result, Chinese robots are not only cheaper, but companies can also release new models more quickly. Xu noted that leading Chinese player Unitree shipped roughly 36 times more units last year than U.S. rivals Figure and Tesla.
Global humanoid robot shipments totaled just 13,317 units last year. That is a tiny base for an industry expected to nearly double annually and reach 2.6 million units by 2035. However, these figures should be viewed with caution. It remains unclear how many units represent commercial sales versus demo models or pilot deployments, underscoring the early-stage nature of the industry. The top humanoid robot makers by 2025 shipments were led by China’s Agibot and Unitree, followed by UBTech, Leju Robotics, Engine AI, and Fourier Intelligence, highlighting Beijing’s early dominance in the sector.
The biggest shift recently has been from demo-driven excitement to operations-driven adoption, according to Yuli Zhao, chief strategy officer at Galbot. Galbot’s humanoid robot, the G1, appeared at this year’s Spring Festival Gala alongside robots from Unitree Robotics, Noetix, and MagicLab. Zhao said more customers are now asking if the robot can run stably in real environments and actually take work off people’s plates. That practical pull is strengthened in China because policy and industrial strategy encourage automation upgrades, and the manufacturing ecosystem makes iteration extremely fast. Zhao added that while increased funding has accelerated progress, the most durable adoption comes when you can show reliable and repeatable value in production or service operations, not just a one-off showcase.
Chinese robotics makers are securing significant investment. Last year, Unitree was valued at around $3 billion after closing its Series C, with ambitions to reach as much as $7 billion in a future IPO. Meanwhile, Galbot has raised more than $300 million in fresh funding, reportedly pushing its valuation to $3 billion, one of the largest financings in China’s humanoid robotics sector to date.
U.S. companies are also moving beyond flashy demos to focus on real-world deployments. U.S. startup Foundation, for instance, plans to build 50,000 humanoid robots by the end of 2027. But China is already targeting a mix of affordable mass-market models and high-end applications, rapidly expanding humanoids across industrial, consumer, and rehabilitation sectors.
When it comes to AI systems and integrated software, it is still unclear where Chinese humanoid firms truly stand. The industry is largely betting on vision-language-action models and world models, but both technologies remain in early stages. Nvidia currently leads the space with its end-to-end humanoid software stack, so most humanoid startups in China are powered by Nvidia’s Orin chips. However, domestic chipmakers are developing homegrown alternatives.
Humanoid robotics makers are still working on fundamental problems. The challenge is enabling robot foundation models to predict the next physical state the robot will face in unpredictable environments. Unlike large language models, humanoid robotics companies cannot simply scrape the internet for training data. So most are relying on simulation environments to generate synthetic data, though real-world data collection remains essential. Because of this data scarcity problem, humanoids are still far from autonomy. The hardware is currently ahead of the software. Safety is also a major hurdle, as one high-profile accident could trigger public backlash. As the industry matures, more regulations are expected.
Given the lack of data, Zhao believes demand for humanoids will grow first in fairly contained workplaces. Early momentum is likely to be in industrial manufacturing, warehouse logistics, and retail, where tasks are repetitive, hours are long, and processes are clear.
Humanoid robot development is not a two-country race. Japan’s robotics ecosystem is targeting humanoid mass production by 2027. Long a pioneer through projects like Honda’s Asimo, Japan leans on precision and advanced control. One area unique to Japan is the increasing use of humanoid robots in eldercare. Coral Capital CEO James Riney believes Tokyo will continue to thrive in the humanoid robotics industry due to labor shortages, a cultural view of robots as friends, and dominance in many parts of the robotics supply chain.
Hyundai Motor’s Boston Dynamics unit introduced a new Atlas humanoid for factory use by 2028, with plans to produce up to 30,000 units annually in the U.S. as part of its AI-driven robotics push.
For China, government policy, industrial strategy, labor shortages, and private capital are all converging to turbocharge the country’s humanoid robotics push. China’s leadership is best understood as a speed-to-scale advantage. The ecosystem compresses the entire cycle from research and development to customer deployment into a very tight loop. That means humanoid companies can move from prototype to real-world deployment faster, learn from real operations, and iterate at a pace that is difficult to match elsewhere.