The 2026 Wuhan Artificial Intelligence and Robotics Expo opened on October 10 with 50,000 visitors on day one, 200 exhibitors, and a declaration from analysts at Zhejiang Securities that would have seemed fantastical three years ago: China's humanoid robot industry has reached a mass production verification phase. The phrase sounds technical, but what it means is concrete. Chinese companies are no longer building robots to demonstrate capabilities at trade shows. They are building factories to make them.
What Actually Happened
The Wuhan Expo, held in the city's Optics Valley district, featured 15 national debuts of new robotic systems across October 10 to 12. Analysts present at the event characterized the state of the domestic industry as having crossed a threshold from prototype demonstration to mass production verification, meaning manufacturers are now validating production-line processes at scale rather than hand-building individual units for showcase deployments. Component suppliers present at the expo described accelerating order books: Siling reported its robot components entering small-batch production, and Joyson announced a dedicated robotics subsidiary to supply controllers and energy modules to multiple humanoid manufacturers. The component supply chain is the leading indicator, because it has to scale before the final assembly lines can.
The factory-level data behind that characterization is striking. UBTECH's new Liuzhou facility in Guangxi province opened in mid-September 2026, with planned annual capacity exceeding 10,000 humanoid robots. One report described the line as capable of producing a robot approximately every 10 minutes. That pace, if sustained, puts UBTECH in a class that no Western robotics company has approached. For comparison, Figure AI, which has the most visible US deployment at BMW's Spartanburg, South Carolina plant, has not disclosed comparable production figures. What Figure has disclosed is that its robots have supported the production of more than 30,000 BMW X3 vehicles at that facility, which provides a useful benchmark for how deployment scale translates to industrial output. But BMW is buying robots; UBTECH is manufacturing them, at a rate that makes the BMW deployment look like a pilot program.
The Western side of the production equation is also moving, if more slowly. In October 2026, Hexagon Robotics disclosed to Forbes that its AEON humanoid is now in active pilot deployment at BMW's Leipzig facility, alongside a commitment from industrial components manufacturer Schaeffler to deploy at least 1,000 Hexagon AEON robots across its global factory network by 2032. Hexagon's robot has an unusual design: its legs end in wheels rather than feet, allowing rapid movement on flat floors while retaining the leg joints needed to navigate stairs and dock with equipment. The Schaeffler deal is the first major multi-year fleet commitment from a Tier 1 automotive supplier, which is a different category of industrial validation than a single pilot deployment. And the data infrastructure side of the industry is maturing in parallel: Mecka AI raised a $60 million Series B from Sequoia, NVIDIA, and Microsoft's M12 in early October, specifically to build human-motion training datasets for humanoid robot manufacturers.
Why This Matters More Than People Think
The mass production threshold matters more than production numbers alone suggest because of what it does to the economics of the industry. When humanoid robots are hand-built at hundreds or low thousands of units per year, their cost structure is dominated by skilled assembly labor and low-volume component pricing. A UBTECH factory producing 10,000 units per year can negotiate fundamentally different supply agreements, amortize tooling costs across a 10,000-unit annual production run, and build the process engineering muscle that reduces defect rates over time. The first 10,000-unit factory is not just a production facility. It is a learning system that generates the data needed to design the 50,000-unit factory that comes next. The companies that build the first mass-production lines are not just ahead on units. They are ahead on the learning curve that determines who builds the dominant factories of 2028 and 2030.
The geographic dimension of this shift is real and underappreciated. China's manufacturing advantage in humanoid robots is not primarily about cheap labor, which is declining in relevance as factory automation advances. It is about supply chain depth. China produces the motors, reducers, actuators, force sensors, and lithium-ion battery packs that go into humanoid robots. When a Chinese company like UBTECH or DEEP Robotics scales production, it can draw on domestic suppliers who are already running at scale. A US or European manufacturer trying to scale faces a supply chain that is either import-dependent or requires parallel investment in domestic component manufacturing. That gap is not unbridgeable, but it is real, and it is visible in the current production numbers. The Wuhan Expo's component supplier announcements are a reminder that the manufacturing ecosystem is deepening even as the headline robot companies get most of the attention.
The critics argue, however, that production volume is not the same as commercial deployment. The bear case for China's humanoid robot manufacturing surge is that the industry is producing robots for internal demonstration purposes, export commitments that haven't converted to actual deployments, and government procurement with subsidized economics that won't survive market pricing. A crowded expo hall in Wuhan doesn't prove paid production-line deployments. Some analysts estimate that 65 percent of Chinese humanoid units currently in circulation are still being used for research, exhibitions, or evaluation rather than live commercial work. If that figure is accurate, the industry's headline production numbers overstate the genuine commercial traction by a wide margin. The transition from mass production verification to actual commercial revenue at scale still involves solving reliability, durability, and ROI demonstration problems that no manufacturer has fully cracked.
The Competitive Landscape
The global competitive picture in humanoid robots is more complicated than a simple China-versus-West framing suggests. Figure AI, the US company that has generated the most commercial visibility through its BMW partnership, is navigating a transition from its second-generation Figure 02 to the Figure 03, which has been moved into logistics work at the Spartanburg plant. Figure has not disclosed production numbers, but BMW's stated intention to build up to 30,000 Boston Dynamics Atlas robots annually by 2028 provides a Western production ambition that would, if achieved, exceed UBTECH's current capacity. Boston Dynamics, owned by Hyundai, is taking a different approach: its Atlas electric humanoid entered pilot programs in multiple industrial settings in 2026, with Hyundai emphasizing precision assembly applications where Atlas's manipulation capabilities are differentiated from Chinese alternatives focused primarily on mobility and carrying tasks.
The robot training data market is emerging as a secondary competitive front. Mecka AI's October 2026 fundraise from Sequoia reveals a constraint: even companies with robots in mass production still need massive volumes of human-motion data to train the AI behaviors that make those robots useful. Mecka pays people to record daily tasks using body sensors and iPhones, then sells the labeled data to humanoid robot makers. Its reported annual run-rate revenue of more than $100 million suggests that multiple major robot manufacturers are paying for this data at scale. The company with the best training data pipeline will have robots that are faster to train, more reliable in deployment, and capable of handling a wider range of tasks. This is a software problem embedded inside a hardware business, and it is where much of the long-run competitive differentiation will be determined.
The Hexagon-Schaeffler partnership represents a structurally different model from both the Chinese production approach and the Figure-BMW partnership. Schaeffler is simultaneously a customer deploying AEON robots and a supplier providing the precision actuators that go into them. This vertical integration creates a feedback loop that no purely software-focused or purely manufacturing-focused competitor can replicate: Schaeffler's actuator engineers can tune their components to the demands the AEON robots create on the factory floor, while the robot deployments give Schaeffler visibility into which actuator specifications matter most for real-world reliability. That kind of customer-supplier integration is how Toyota built its manufacturing advantage in conventional automotive. It takes time to develop but becomes very difficult to displace once established.
Hidden Insight: The Real Race Is for the Training Data and the Software Stack
China's lead in humanoid robot production creates a misleading impression about where the long-run competitive advantage will reside. Building a humanoid robot at scale is an engineering and manufacturing problem. Building a humanoid robot that can do genuinely useful work reliably across diverse unstructured environments is an AI problem. The hardware is increasingly commoditized: the motors, actuators, and mechanical frames that go into humanoid robots are becoming as standardized as the components in a smartphone. The software, specifically the AI stack that decides how a robot responds to a novel object, an unexpected instruction, or a floor obstacle it hasn't encountered before, is where the moats will be built. Chinese manufacturers have an early production advantage, but the AI capability question is much more open.
The data dependency revealed by Mecka's fundraise is the specific lever worth watching. Current humanoid robot AI systems are trained on a combination of simulation data, human demonstration data, and real-world trial-and-error. Simulation data is cheap to generate but doesn't fully capture the physics of the real world. Human demonstration data is expensive to collect but transfers reliably to robot behavior. The companies that build the best human demonstration datasets will have AI-capable robots faster and at lower cost than competitors who try to use simulation alone. Mecka is building that infrastructure for anyone who will pay. But the AI labs embedded inside major robot manufacturers, including Tesla for Optimus, Figure, and UBTECH, are also building proprietary training data pipelines. The outcome of this parallel development race is unclear, but whoever solves the data problem first will be able to train reliable robot behaviors in weeks rather than months.
There is a second hidden dimension: the humanoid robot industry's energy economics are about to become a binding constraint. A humanoid robot running an 8-hour shift in a factory consumes approximately 3 to 5 kilowatt-hours of electricity, not counting the server-side compute required to run its AI inference. At scale, a fleet of 1,000 humanoid robots working full shifts requires the same power as a small data center. As humanoid deployments move from hundreds to thousands to tens of thousands of units, power consumption becomes a factory planning variable that today's facilities haven't designed for. This is not a problem unique to robotics: the AI industry broadly is grappling with power demand that outpaces grid capacity. But it is an underappreciated constraint on how quickly humanoid robot deployments can actually scale inside existing factory infrastructure.
Finally, the geopolitical dimension of the production race deserves consideration that most industry coverage skips. China's aggressive industrial policy support for humanoid robots, which includes subsidies, state procurement, and preferential access to manufacturing infrastructure, creates a floor for Chinese production volumes that private market economics wouldn't generate alone. US and European companies are competing partly against market forces and partly against a state investment strategy. The distinction matters for predicting outcomes: Chinese manufacturers can sustain production at volumes that wouldn't be commercially rational if they had to cover their full costs, which allows them to build learning-curve advantages faster than a purely market-driven competitor could. Western policymakers are beginning to recognize this dynamic, but the policy responses, including potential import restrictions and domestic manufacturing subsidies for robotics, are at early stages and won't change the near-term production gap.
What to Watch Next
The Wuhan Expo runs through October 12, and any major production commitments or deployment announcements made on the final day will indicate the current state of Chinese manufacturers' commercial confidence. The more important 30-day indicator is UBTECH's Q3 production report, which should provide the first official quarterly output figure from a humanoid manufacturer at the 10,000-units-per-year scale. If the Liuzhou factory is running at its rated capacity, that would confirm the production claims made at the expo. If production is running below rate, it would suggest the mass production verification phase is still working through process reliability issues that haven't been publicly disclosed.
The 90-day signal to watch is Figure AI's announcement of its next commercial customer. Figure has one confirmed major deployment at BMW Spartanburg. A second major deployment, particularly with a logistics customer like UPS or a non-automotive manufacturer, would confirm that Figure's technology works across domains rather than just in the specific environment it has optimized for. The Journal reported in 2025 that UPS was in talks with Figure about deployment, but no formal agreement has been announced. If that deal or something equivalent closes in Q4 2026, it would be a key data point about Western humanoid robots' ability to move beyond automotive pilot programs.
The 180-day horizon to watch is the component supply chain. The Wuhan Expo announcements from Siling, Joyson, and other component makers suggest the Chinese supply chain is deepening. Watch for any announcements from US-based motor or actuator manufacturers about humanoid-specific production investments. Companies like Harmonic Drive and Allied Motion Technologies supply precision motion components that go into humanoid robots but have not publicly described humanoid-specific capacity investments. If those announcements emerge in the first half of 2027, it would indicate the US supply chain is beginning to respond to the production ambitions of Western robot manufacturers. If they don't, the Western manufacturers will remain dependent on Asian suppliers for key components, which is both a cost disadvantage and a supply chain risk.
China is winning the humanoid robot production race, but the factory count doesn't tell you who wins the AI race: the real competition is for the data and the software that decide whether these robots can actually think.
Key Takeaways
- Wuhan Expo opens October 10 with mass production milestone: Zhejiang Securities characterized China's humanoid robot industry as having reached the mass production verification phase, with 200 exhibitors and 15 national debuts at the event.
- UBTECH factory ships 10,000+ units per year: UBTECH's Liuzhou, Guangxi facility, which opened in September 2026, is the first humanoid robot factory in China at this production scale, with capacity for one robot approximately every 10 minutes.
- Hexagon-Schaeffler commits to 1,000-unit fleet by 2032: Schaeffler, a Tier 1 automotive supplier, committed to deploying 1,000 AEON humanoid robots across its global factories, becoming the first major industrial components maker to announce a humanoid fleet at this scale.
- Training data is the hidden bottleneck: Mecka AI's $60M Series B from Sequoia, NVIDIA, and Microsoft reveals that even production-scale robot makers still need massive human-motion datasets to train reliable AI behaviors, creating a parallel competitive race in data infrastructure.
- 65 percent of Chinese humanoids remain in research use: Despite production-scale ambitions, analysts estimate that most Chinese humanoid units currently in circulation are used for research, exhibitions, or evaluation rather than commercial production deployments.
Questions Worth Asking
- If 65 percent of Chinese humanoid robots are still being used for research and demonstrations rather than commercial deployment, does the "mass production verification" label describe the manufacturing capability or the commercial readiness, and are those the same thing?
- The AI stack that makes humanoid robots genuinely useful across diverse environments is still being developed. If the hardware commoditizes quickly while the AI capability gap between manufacturers widens, who actually wins: the company with the most robots or the company with the best training data?
- China's humanoid robot industry benefits from state subsidies that allow production volumes that private economics alone wouldn't justify. At what scale does that state support create a structural disadvantage for Western companies that have to cover their full costs, and what is the appropriate policy response?
