Insider Brief
- A new report from the Mercator Institute for China Studies found China dominates global robot manufacturing but remains heavily dependent on U.S. technology for the AI systems powering many of its humanoid robots.
- According to the report, China produced roughly 90% of the world’s humanoid robots last year and now controls much of the robotics supply chain, aided by spillover benefits from its electric vehicle industry.
- Researchers said Chinese robotics firms still rely heavily on AI infrastructure from NVIDIA and continue to trail Western competitors in general-purpose robotics models, creating what the report describes as a strategic software vulnerability.
China made roughly 90 percent of the world’s humanoid robots last year. It operates the largest fleet of industrial robots on the planet. Its factories installed more robots over the past five years than every other country combined. And yet, at the heart of nearly every intelligent robot rolling off a Chinese production line sits technology built by a company in Santa Clara, Calif.
That paradox of overwhelming hardware muscle with stubborn software dependency is one of the critical insights in a new report from MERICS, the Mercator Institute for China Studies, a Berlin-based research institution funded by the German federal government. The report offers the most comprehensive recent assessment of China’s push into what researchers call “embodied AI”: artificial intelligence that doesn’t just think, but acts in the physical world through robots.
The findings matter well beyond tech policy circles. China’s robotics ambitions are reshaping global manufacturing, straining its own social fabric and threatening to repeat, in metal and motors, the same competitive rout it delivered to Western automakers with electric vehicles.
The Hardware Champion
By any measure of physical production, China’s robotics industry is formidable. According to the MERICS report, Chinese companies supplied more than 57 percent of China’s domestic industrial robot market in 2024, overtaking foreign competitors for the first time. In 2025, Chinese factories produced 773,074 industrial robots and 12,870 humanoids, with the latter representing about 90 percent of global humanoid output.
The cost advantage is real and striking. Unitree Robotics, one of China’s leading humanoid makers, sells its G1 model for roughly €11,650. Boston Dynamics‘ Atlas, a comparable American machine, costs more than €120,000. Tesla‘s Optimus sits at around €25,800.
Much of that cost advantage flows from China’s dominance in electric vehicles. The report identifies the EV supply chain as a kind of hidden subsidy for the robotics industry. Batteries developed for cars now power humanoid robots and lidar sensors and cameras built for autonomous driving navigate factory floors. XPeng, the EV maker, uses autonomous-driving algorithms in its Iron humanoid robot. GAC’s GoMate robot runs on EV-derived batteries, as does Spirit AI‘s Xiaomo.
This cross-pollination is no accident. China controls 63 percent of the key companies in the global humanoid robot component supply chain, the report notes, citing Morgan Stanley research. That includes actuators, which are the motors and mechanical joints that give robots their movement, and the rare earth materials needed to make them. When an industry already owns the parts, building a new product with those parts is faster and cheaper.
However, foreign industrial robot makers that include Switzerland’s ABB and Japan’s FANUC still dominate the high-end market. Germany’s Schaeffler and Japan’s THK and NSK supply 90 percent of the specialized ball screws used for precise positioning in robotics applications. Chinese humanoids, the report says, still struggle with multi-step coordination, fine hand movements, and tasks requiring precise hand-eye coordination, such as inserting small parts or assembling cables.
The report also notes that for wide scale implentation of humanoid robots, Chinese firms will still have to trim the cost by about half. However, that is just one barrier.
The Software Gap
For all that manufacturing firepower, Chinese robotics firms face a ceiling they cannot yet break through on their own when it comes to the intelligence layer.
Running a robot in the physical world requires more than motors and sensors. It requires software that can process what those sensors detect, interpret the environment in real time, and decide what to do next. This is the domain of what researchers call “vision-language-action” models, or VLAs, that take in visual and language inputs and translate them into immediate physical actions. Think of it as the robot’s brain.
Google DeepMind pioneered this approach with its RT-2 model and remains a global leader. Nvidia has built an entire ecosystem of tools around it. Chinese firms, the report says, are making progress in targeted areas but still lack foundational generalist models that can handle genuinely novel situations without being pre-programmed for them.
Most Chinese robotics companies rely heavily on Nvidia’s hardware and software. Nvidia’s Jetson modules — compact computers that combine processing chips with an AI development platform — are embedded in robots made by UBTech, Galbot, Unitree, EngineAI, and AgiBot, among others. These firms were among the first to receive Nvidia’s latest Jetson Thor module, the report notes.
Nvidia also provides Isaac Sim, a robotics simulation environment where companies train their robots in virtual environments before deploying them in the real world. Both Unitree and Galbot use it and AgiBot uses it for elements of its development work. The report’s exhibit mapping Chinese humanoid firms against Nvidia’s software and hardware tools is striking as it includes nearly every major Chinese player, each dependent on one American supplier for the core of their robots’ cognitive capability.
Beyond being a technical embarrassment, the analysts also consider is a strategic vulnerability. The report notes that Nvidia’s edge-computing chips used in robotics have not yet been targeted by U.S. export controls, but that could change as they grow more powerful. China’s robotics industry is, in effect, building on a foundation it does not control and cannot easily replace.
Chinese firms and research institutions are working to close the gap. AgiBot released its first general-purpose AI model, called GO-1, in 2025. Galbot is developing specialized VLA models for specific tasks, including one for grasping, one for navigation and one for retail environments. Companies including SenseTime, Huawei, and XPeng are developing what researchers call “world models,” AI systems that build internal representations of physical environments. But the MERICS report concludes that China remains in the early phases of this work, still dependent on U.S.-based research for the foundational breakthroughs.
The Social Bargain Under Strain
Beijing’s robotics push is not taking place in a vacuum. It is unfolding inside a country with nearly 300 million migrant workers, a weak social safety net, youth unemployment running around 17 percent and a Communist Party whose legitimacy rests in significant part on its ability to deliver economic security.
The MERICS report documents a growing tension between the government’s technological ambitions and the human consequences of achieving them. Several studies cited in the report conclude that AI will displace Chinese jobs far faster than it creates them. One estimate suggests robots could ultimately replace at least 70 percent of China’s manufacturing workforce. An August 2025 policy document from China’s National Development and Reform Commission — the country’s top economic planning body — invited citizens to think about AI in terms of “empowerment” rather than “substitution.” The linguistic gymnastics suggest awareness of the underlying anxiety.
That anxiety is already visible in Chinese public discourse. The report notes growing public concern about job security and links the robotics price war to a broader Chinese phenomenon the report calls “involution” — a cycle of self-defeating competition, overcapacity and suppressed profits familiar from other Chinese industries. An NDRC representative acknowledged publicly that an overcrowded humanoid market risked generating overcapacity and cutting into the R&D budgets needed for genuine innovation. The ministry has set up a standardization committee for humanoid robots, a move the report suggests may precipitate a shakeout among China’s roughly 150 humanoid manufacturers.
China’s 15th Five-Year Plan, released in 2026, lists embodied AI as a national priority and acknowledges the need to monitor AI’s impact on employment. But as the MERICS report observes, Xi Jinping’s government has no intention of building a Western-style welfare state. Whatever social adjustments are needed will have to come from other directions such as retraining programs, new industries or managed displacement, without the cushion of robust unemployment insurance or social support systems.
The report’s authors suggest this is the CCP’s core bet, that the productivity gains from AI-powered robotics will be large enough, and distributed widely enough through economic growth, to hold the social contract together. It is a bet that has not yet been tested at scale.
The Warning for the West
The MERICS report’s sharpest warning is aimed at European industry. China’s humanoid makers are cheaper than Western competitors, backed by state policy and capital, embedded in the world’s most capable manufacturing ecosystem and moving fast to replace the high-end components they still import from Europe and Japan.
The parallel the report draws is explicit: this is the EV race again. One dominant Western player — in that case, Tesla — competed against a fleet of Chinese manufacturers with state support, cheap components and the willingness to sustain losses in pursuit of market share. Those manufacturers eventually became technologically serious and are now reshaping global automotive markets. The report suggests the same dynamic is now initializing in robotics.
What separates the two races is the software problem. The robotic brain — the AI that makes an autonomous machine genuinely useful across varied, unpredictable environments — remains an unsolved challenge globally, not just for China. That gap is the one place where Western and American firms retain a meaningful lead. How long it holds may determine whether China’s hardware dominance eventually translates into the kind of market control for which Beijing is planning.
For now, the machines can dance and kickbox. Whether they can truly think is still an open question. For factories, for workers, and for the balance of industrial power in the decades ahead, the answer will matter enourmously.
