UBTECH turns on a 10,000-a-year humanoid factory where robots build robots — but capacity is not the same as shipments.

On September 12, Hong Kong-listed UBTECH began operations at a 'super smart factory' in Liuzhou, in South China's Guangxi Zhuang Autonomous Region, that it says is the world's first humanoid plant designed for a 10,000-unit annual capacity — capable, it claims, of rolling out one industrial humanoid every 10 minutes. The 14,000-square-meter (150,700-square-foot), 13.8-meter-tall building was jointly built with Siemens Digital Industries Software, which supplied the digital twin, the Plant Simulation modeling and a co-developed Yanshee manufacturing operations system that UBTECH calls the plant's 'smart brain.' It primarily makes the bipedal Walker S industrial series and the wheeled Cruzr service series, and UBTECH's shares rose more than 3% on September 14 after the launch. The reason this is more than a ribbon-cutting: building a machine of thousands of parts repeatably and cheaply is the wall the whole humanoid industry keeps hitting, and Liuzhou is the largest declared production base yet aimed at it.

The signature detail is that robots are helping build the robots. On the floor, UBTECH's own wheeled Cruzr Y1 and Cruzr S2 units handle depalletizing, palletizing, loading and material transport, while collaborative arms and 360-degree rotating worktables do assembly — each robot takes more than 2,000 screws across 50-plus specifications, with automated torque logging and one-item-one-code serial-number traceability. Automated guided vehicles and unmanned forklifts move materials, and a 65-square-meter automated warehouse buffers up to 112 finished units while improving floor-space use by more than 50%. Before shipping, every robot gets more than four hours of whole-machine testing and a pass through an automotive-grade lighting tunnel for a 360-degree visual inspection. UBTECH vice president Pang Jianxin argued the project is only possible because of China's supply chain: 'Without the Chinese supply chain, mass-producing complex components would be extremely difficult unless cost advantages were completely sacrificed.'

The number deserves a skeptic's asterisk. The 10,000-unit, one-every-10-minutes figure is a planned design capacity, not verified round-the-clock output, and the mixed line covers Walker S bipeds and wheeled Cruzr platforms rather than 10,000 bipedal humanoids. UBTECH's actual biped volume is far smaller: it marked its 1,000th Walker S2 at the site on December 26, 2025, with more than 500 delivered at that point, and in the first half of 2026 it sold 921 full-size humanoids while posting an interim loss of 339 million yuan (about $48 million) on 1.27 billion yuan (about $179 million) of revenue. For scale, Counterpoint Research estimated global humanoid shipments topped 22,000 units in the first half of 2026 with Chinese firms holding all five top spots — AgiBot around 9,700 and Unitree more than 7,000, with UBTECH fourth — though mixed product lines make those counts hard to compare. UBTECH's Walker deployments at BYD, NIO and Foxconn are real, but still task-by-task rather than full-line.

For a procurement team, Liuzhou is the strongest evidence yet that a Chinese maker can industrialize humanoid production, and cheaper units at volume are ultimately what make a purchase pencil out. But capacity is a promise about supply, not proof of demand or usefulness: the number to watch is not the line's 10-minute takt but how many Walker S2s actually ship to paying customers and run reliable multi-shift work, and UBTECH's own first-half figures — 921 sold, still lossmaking — show how early that is. The same constraint that shadows every Chinese humanoid applies here too: UBTECH is China-made, so its robots are shut out of new US sale under the FCC's July import ban, no matter how fast the line runs. Read alongside last week's XPeng automated-line milestone, Liuzhou reinforces that China is winning the manufacturing race; whether that converts into shipped, deployed and profitable robots is the question 2027 will answer.