Invest1 publisher3 min readPublished
UBTECH's 10,000-robot year in Liuzhou fits inside a single shift
UBTECH put a capacity number on the record for its new Liuzhou humanoid plant and left out the price, the shipments and the order book, so the cost case rests on its claim about Chinese component supply.
The Investor · Invest desk
What happened
- UBTECH said on the 13th that it had begun operating a humanoid robot plant in Liuzhou, in the Guangxi Zhuang Autonomous Region, with annual capacity of 10,000 units.
- The facility covers 14,000 square meters, stands 13.8 meters tall, and mainly builds the company's Walker S and Cruzr robot series.
- UBTECH says an in-house automated three-dimensional warehouse holds 112 robots in 65 square meters, lifting space utilization by more than 50% for the same floor area.
- RoboMIND, the open-source robot dataset released last December by the Beijing Innovation Center for Humanoid Robotics, had passed 20 million cumulative downloads as of the 2nd.
Compiled by The InvestorSomething wrong?How this is made
Why it matters
- constraint The warehouse holds about four days of output at full rate, so finished robots have to be pulled by orders within days of leaving the test tunnel; the building cannot store a quarter of unsold units.
- capability Because the floor's unloading, stacking and transport run on the Cruzr models the plant also builds, UBTECH can place some of its own output internally before an external buyer signs anything.
- decision Anyone modelling humanoid unit costs has to decide whether a capacity claim counts as evidence, since UBTECH published no price, no shipment total and no backlog alongside it.
Ten thousand units a year at one unit every ten minutes works out to 100,000 minutes of line time, or about 1,667 hours [1], against the 8,760 hours in a calendar year. Spread across 250 working days that is under seven hours a day [2]. Run the same ten-minute cycle around the clock and the building would turn out 52,560 robots a year [3]. Both figures, the capacity and the cadence, come from UBTECH and were relayed by the state-run Global Times [2].
Four hours of full-unit testing per robot, plus a pass through an automotive-grade inspection tunnel [7], is where volume gets expensive. Ten thousand units at four hours each is 40,000 test hours a year [4]. Fit that inside 1,667 hours of line time and you need roughly 24 test positions running in parallel [5]. Test continuously, nights and weekends included, and you still need five [6].
UBTECH said the plant takes it past small-batch and pilot production, and credited China's component supply chain for making a line of that size possible [9]. Fang Jianxin, a UBTECH vice president, told Global Times that "Chinese robots have already entered the global first tier" [10], and on cost he said that "without China's supply chain, mass-producing complex components would be extremely difficult unless you completely gave up your cost advantage" [11]. Siemens Digital Industries Software took part in building the plant [5].
The capacity is also committed capacity. The 14,000 square meters are tooled for the Walker S and Cruzr lines [4], so a third design competes with the line UBTECH has just finished paying for.
Against conventional robots the volume is small. China's July industrial robot output of 98,677 units, up 30.2% year on year [12], implies about 22,900 more units than the same month a year earlier [8], which is more than twice the Liuzhou plant's entire annual capacity [9].
Training data is running on separate infrastructure. The Beijing Innovation Center for Humanoid Robotics says it has supplied about 30,000 hours of high-quality data, collected with more than 150 robots and 40 robot configurations across more than 30 reproduced environments [15], which is about 200 hours per robot [10].
Three readings sit in the same facts: the nameplate stays a nameplate and 10,000 turns out to be a 2030 figure; the line fills and the binding constraint becomes test bays and software instead of assembly; or the line fills, component prices fall with volume, and the decline reaches every assembler buying from those suppliers. I'd weight the middle one, because a four-hour test scales with bays and headcount and not with cycle time. Fang's component claim would be wrong if an assembler outside China shipped comparable units at comparable prices.
What to watch
- Any monthly shipment or utilization figure for Liuzhou, which would show whether the 10,000-unit line runs a shift or a pilot batch.
- A list or contract price for Walker S and Cruzr units bought at volume by a named customer.
- Whether National Bureau of Statistics output data starts breaking humanoids out from the 635,056 industrial robots reported for January to July.