UBTECH Humanoid Robot Factory Scales to 10,000 Units as Demand Trails Supply

UBTECH has put a humanoid robot factory into operation in Liuzhou, in southern China's Guangxi Zhuang Autonomous Region, built for an annual output of 10,000 industrial humanoids at a line speed of one completed robot every 10 minutes. The company describes the 14,000-square-metre site as the first manufacturing plant designed around four-digit annual humanoid volume. It runs alongside UBTECH's existing humanoid factory in Shenzhen. The nameplate figure is roughly nine times the number of full-size humanoids UBTECH actually shipped in 2025.

Production started this week. The Liuzhou lines will build the Walker S industrial humanoid and the Cruzr service model, with humanoid robots carrying out assembly work on other humanoids. UBTECH's capacity before the new plant topped about 6,000 units by the end of 2025, and the company delivered around 1,079 full-size humanoids last year. The distance between those two numbers is where the story sits.

Inside the UBTECH Humanoid Robot Factory

The Liuzhou site is not simply a faster version of an existing line. It is designed to be largely self-operating, which makes UBTECH its own most reliable customer for the output. A plant that consumes its own product demonstrates that the product can be built. It does not demonstrate that external buyers exist in the quantities the capacity implies.

UBTECH has already crossed 1,000 Walker S2 units and delivered more than 500 of them, which establishes that the hardware can be manufactured and shipped at modest scale. Reaching 10,000 units a year needs a different class of buyer, and the company has not disclosed an order book for the Liuzhou plant.

The Walker S stands 1.76 metres tall, which places it in the footprint of a human workstation rather than a purpose-built cell. That is the design bet behind the whole category: sell into existing factory layouts instead of rebuilding the plant around the machine. It also explains why automotive manufacturing, smart factories and logistics dominate the first wave of deployments, since each runs repetitive tasks in human-scaled spaces.

UBTECH now operates two dedicated humanoid plants, one in Shenzhen and one in Liuzhou. Splitting production between a coastal technology hub and an inland industrial city spreads supply-chain risk and places the higher-volume line closer to the Guangxi procurement programmes that have already bought from the company.

Capacity Versus Orders

The company's published targets run well ahead of its delivered volumes, and the distance has grown with each capacity announcement.

Reaching 5,000 units in 2026 would mean multiplying 2025 deliveries by roughly four and a half. Doubling again the following year would push the Liuzhou line towards a ceiling it has not yet approached. Because UBTECH has not published utilisation or order figures, the schedule reads as a plan rather than a forecast.

Fixed costs sit underneath that plan. A plant sized for 10,000 units carries depreciation, tooling and staffing that must be spread across whatever volume it achieves. Shipping 2,000 units instead of 10,000 multiplies per-unit overhead several times over. UBTECH has pointed to a target of about $20,000 per humanoid by 2030, and the route to that figure runs through utilisation rather than engineering.

Where the Constraint Actually Sits

The hard problem in industrial humanoids has moved from the assembly line to the deployment site. UBTECH's most revealing contract is not a factory order. Last October the company won an $18 million tender from the Guangxi regional government to supply humanoids and related hardware to a training facility, with the aim of producing data for embodied AI, the branch of machine learning that lets robots act in physical environments.

That structure explains the current business model. Robots placed in training and demonstration settings generate operating data, and that data trains the policies that make later units worth buying. UBTECH has more than 100 humanoids working in what it frames as intern roles at its smart factory. Its partnership with Siemens Digital Industries Software, announced in March 2025, targets 10,000 units of annual manufacturing capacity and standardises the software layer that turns hardware into a repeatable product.

State demand is doing real work in that model. The Guangxi tender gives UBTECH a paying customer while private orders build, and publicly backed training programmes create a data pipeline that few commercial clients would fund on their own. Data captured inside a state-funded training centre also differs from data captured on a paying customer's production line, and the second kind is what industrial buyers will ask to see before they commit to a fleet.

Competitors, Pricing and Trade-offs

Liuzhou is not an isolated bet. AgiBot moved its 5,000th humanoid off a Shanghai line in December 2025 and reached its 10,000th unit by March 2026, a faster ramp than UBTECH has managed. XPeng and BYD are building humanoid capacity for use inside their own plants, which hands them a captive demand base UBTECH does not have.

Pricing is still wide enough to complicate any payback case. UBTECH's UWORLD U1 humanoid spans 119,800 RMB to 990,000 RMB depending on the model, or roughly $17,655 to $145,717. Industrial configurations are quoted separately. At those levels, a buyer weighing a humanoid against a fixed industrial arm or an extra shift needs a return that the technology has not yet consistently delivered.

For a manufacturer deciding where to spend, the trade-off is between a humanoid that can be redeployed across tasks and a fixed arm that does one job at a lower unit cost. Humanoids win where the task mix changes; fixed automation wins where volume is stable and the motion repeats. UBTECH's capacity bet only pays if enough sites have the first profile.

The bullish reading of the plant borrows from Chinese solar and electric vehicles: build ahead of demand, absorb the overhead, and let scale pull unit costs down until the product becomes hard to avoid. The bearish reading is that humanoids are not panels. A solar module is standardised and its buyer does not care where it was made. A humanoid's value depends on the site, the task, the software integration and the service terms attached to it.

Two questions remain open in UBTECH's disclosures. The first is how a fleet gets supported once it is installed: a humanoid that fails on a customer's floor needs spare parts, technicians and software updates, and those costs do not fall as the assembly line speeds up. The second is how much of the Liuzhou capacity is already contracted rather than available.

UBTECH is betting on the bullish reading, and its balance sheet carries the bet until orders arrive. The test is not whether the Liuzhou line can produce a robot every 10 minutes. It is whether UBTECH can sign enough automotive, logistics and smart-factory customers to keep that line running at a fraction of its ceiling that still covers fixed costs. AgiBot has shown the output can be reached. No Chinese humanoid maker has yet shown the output can be sold at scale.

Why this matters

The Liuzhou plant makes hardware availability less of a constraint for industrial buyers, which lowers the cost of starting a humanoid pilot. It does not reduce the integration risk that decides whether a pilot becomes a fleet. For investors and competitors, UBTECH's nameplate capacity sets a benchmark that AgiBot, XPeng and BYD will be measured against, and the number that reveals the real winner will be utilisation, not output.

Related Articles

• China pushes 10,000 humanoid robots by 2026 as factories struggle to absorb them
• Humanoid Robot Shipments Jump 272% as AgiBot Overtakes Unitree in China-Dominated Market
• Humanoid Robot Shipments Top 22,000, Yet Factory Use Lags [Update]

✔Human Verified

Researched and cross-referenced against primary sources by the Bytevyte editorial team. This article was generated with the assistance of artificial intelligence and reviewed by the Bytevyte editorial team.