Завод площадью 14 000 квадратных метров в Гуанси заявляет, что способен выпускать одного промышленного гуманоида каждые десять минут. Линия действительно существует. А вот сможет ли мир покупать 10 000 таких роботов в год — это уже другой вопрос — и... ...
Сообщение Planet-Today.com. Перевод заголовка и краткого описания выполнен автоматически.
A 14,000-square-meter plant in Guangxi says it can finish one industrial humanoid every ten minutes. The line is real. Whether the world will buy 10,000 of them a year is a different question — and that is the part worth reading slowly.
 |
| Source: Video Screenshot |
Key Takeaways by Planet Today
Capacity is not the same as shipments: UBTECH commissioned the Liuzhou “super smart factory” on 12 September 2026 with a designed annual run-rate above 10,000 industrial humanoids and a 10-minute takt time. That figure describes line capability under design assumptions, not independently audited year-round output.
Siemens is software, not a co-owner: The plant was built with Siemens Digital Industries Software — digital twins, Plant Simulation, and a jointly developed manufacturing operations system. It is a strategic industrial-software partnership signed in March 2026, not a confirmed legal joint venture that splits equity in the factory.
The labor argument cuts both ways: China faces a shrinking working-age population and fewer young people willing to stand on assembly lines. Humanoids can cover dull and hazardous tasks. They can also displace the same wages that still hold up household demand in factory towns.
Hardware lead, software race: Chinese firms now dominate unit shipments of humanoids. U.S. and European firms still argue they lead in general-purpose models, chips, and foundation software. Both claims can be true at once. Neither settles who captures profit.
The next test is the order book: A factory that can build 10,000 machines a year only matters if customers keep them running without an army of technicians. Validation cycles, safety rules, and unit cost will decide more than the opening video.
What just happened
On 12 September 2026, UBTECH Robotics started production at what it calls the world’s first intelligent manufacturing plant designed for 10,000-unit annual capacity of industrial humanoid robots. The site sits in Liuzhou, in the Guangxi Zhuang Autonomous Region. The building covers 14,000 square meters and stands 13.8 meters high. The intended products are the Walker S series of bipedal industrial humanoids and the Cruzr series of wheeled embodied platforms.
The latest public account of the opening was published by the Global Times on 14 September 2026. International trade and technology desks followed within two days. India Today reported the same numbers on 16 September. UBTECH’s own account, posted on 15 September, put the slogan in one line: “One humanoid robot rolls off the line every 10 minutes.” See the company post on X / UBTECH Robotics.
That is the official picture. It is also the picture that travels fastest: robots assembling robots, a digital twin of the hall, no human hands on the small parts. The slower picture is less cinematic. A takt time of ten minutes is a design cadence. A commissioned plant can exist months before it runs at that pace for a full year. Quality, yield, rework, battery packs, wiring looms, software loads, and customer acceptance all sit between a ribbon-cutting and a real 10,000-unit year.
Sci-fi is real! UBTECH just launched the world’s first 10,000-unit humanoid robot factory in Liuzhou. With 1 robot rolling off the line every 10 minutes, this is a massive leap for AI manufacturing! pic.twitter.com/VUUuQpnrQ7
— Lookout (@watchmancz) September 14, 2026
How the line is supposed to work
UBTECH and Siemens Digital Industries Software signed a strategic cooperation framework in Shenzhen on 16 March 2026. The stated aim then was already the same number now printed on the factory wall: 10,000 industrial humanoids a year. Siemens was to supply the digital thread — product design, simulation, process planning, and production management — while UBTECH supplied the machines. Shenzhen municipal coverage of the signing is here. Independent industrial analysis of the same deal is on ARC Advisory Group.
Before the physical hall was locked in, engineers built a 1:1 digital twin in Siemens Plant Simulation. The model was not a marketing render. It mapped aisles, workstations, shelves, AGV routes, empty-tote return cycles, and buffer thresholds. The question the software was asked to answer is the only one that matters on a mixed-model line: can a ten-minute cycle run without parts jamming and without vehicles blocking one another?
On the floor, the company says Cruzr Y1 and Cruzr S2 wheeled manipulators handle depalletizing, palletizing, loading, tote replenishment, and material transfer. Collaborative arms, power-assist manipulators, unmanned logistics vehicles, and 360-degree rotating worktables take the final-assembly stations. Automatic guided vehicles and unmanned forklifts move parts from warehouse to station. Finished machines go into what UBTECH describes as the first fully automated stereoscopic warehouse built for humanoids: 65 square meters, 112 stored robots, more than 50 percent better space use than a conventional layout.
Quality is the part that sounds least like science fiction and most like an auto plant. Each completed robot is said to run more than four hours of whole-machine dynamic testing — joints, gait, payload — then pass through an automotive-grade lighting tunnel for a 360-degree appearance check. Industry reporting adds a fastener count that explains why digital torque stations exist at all: more than 2,000 screws across about 50 specifications on a single chassis. Those details are collected in Humanoids Daily’s 13 September factory brief.
Traceability is closed-loop, at least on paper. Each robot carries a unique serial number. A manufacturing operations system — UBTECH’s Yanshee MOM, tied to Siemens Intelligence Center X — is described as the plant’s “smart brain” for work-order scheduling, material distribution, and quality monitoring.
“Mass production is inseparable from the support of China’s supply chain. Without the Chinese supply chain, mass-producing complex components would be extremely difficult unless cost advantages were completely sacrificed.”
— Pang Jianxin, vice president, UBTECH, quoted by the Global Times
That sentence is more important than the ten-minute slogan. Humanoid robots are not a single invention. They are a pile of motors, reducers, force sensors, batteries, harnesses, cameras, and controllers. China’s argument is that it already makes those piles at a cost the rest of the world has not matched. The counter-argument is that a cheap pile is not the same as a machine that works an eight-hour shift without a babysitter.
What mass media is saying
State-linked Chinese English-language coverage treats the opening as a stage change: from laboratory prototypes and small-batch trials to “large-scale intelligent manufacturing.” The Global Times piece leans on national industrial-robot output as supporting weather — 98,677 industrial robots produced in July, up 30.2 percent year on year; 635,056 in the first seven months, up 28.5 percent, according to the National Bureau of Statistics as cited in that report. Those figures are for industrial robots in the classic sense, not humanoids. They still show the wider machine-tool climate in which the Liuzhou hall sits.
Western and specialist technology desks have mostly repeated the same specifications — 14,000 square meters, ten minutes, 10,000 units, Siemens software — while adding a softer note about the leap from demo to factory. Interesting Engineering and The AI Insider stay close to the company and Global Times numbers. Spanish and Korean general-interest papers did the same translation job on 15 September. The tone is wonder with a thin layer of unease: the factory looks like a scene from a film, and films about self-replicating machines rarely end with a productivity bonus.
Financial wires have been cooler. UBTECH’s Hong Kong-listed shares moved on the news; German-language market notes also recorded that the company was still loss-making in the first half of 2026 even as humanoid revenue jumped. That is not a scandal. Scaling factories burn cash. It is a reminder that a production hall is a cost center until invoices are collected.
What more skeptical and specialist outlets are saying
The sharper reading sits in trade analysis rather than in opposition blogs. Black Scarab’s 13 September note is useful because it separates three things that press releases glue together: designed capacity, demonstrated takt, and accepted deliveries.
UBTECH’s own interim numbers, as summarized in that analysis and in market reports, show real commercial motion. Full-size embodied-humanoid revenue in the first half of 2026 was reported around RMB 590 million, up from about RMB 38 million a year earlier, with 921 units sold against 45. Total company revenue was about RMB 1.27 billion. The same period still showed a loss of about RMB 339 million, and accounts receivable remained heavy. Humanoids Daily adds a growth rate of 1,445 percent in full-size humanoid revenue for that half and says enterprise industrial clients accounted for more than 96 percent of that income. Those are company-reported figures, not a third-party census of robots still working on customer floors six months later.
The skeptical case is not that the factory is fake. The skeptical case is that humanoid manufacturing has not yet earned the word “automotive.” A ten-minute interval at an opening is not a year of yield. A 75 percent laboratory success rate on a vision-language-action loading task, cited in that same critical brief, is the sort of gap that factory managers notice and journalists skip. Customers named in UBTECH materials — BYD, NIO, Geely, Dongfeng Liuzhou Motor, FAW-Volkswagen, Foxconn, SANY — prove that trials exist. They do not, by themselves, prove fleet-scale repeat orders or a positive return on the integration cost.
Social media, as the original brief noted, reaches for the dystopia button first. “Robots building their own kind” is an old science-fiction sentence. It is also a poor description of Liuzhou. Cruzr units moving totes and AGVs feeding stations is industrial automation eating another slice of kitting and logistics. It is not a closed loop in which a finished Walker walks over and designs the next Walker. The distinction matters. Fear attached to the wrong picture makes the real picture harder to judge.
Labor: shortage, rejection, or both?
The policy case for the factory is demographic. China’s working-age population has already peaked. Young workers increasingly prefer service and platform jobs to night shifts on a line. Manufacturers have said for years that they cannot fill every station at the wage they want to pay.
The “30 million manufacturing workers” figure that circulates with this story is best treated as an older industry-planning estimate of a skilled-labor gap, not as a census count published with this factory. More recent bank research is easier to pin down. In May 2026 Barclays estimated that China’s working-age population could shrink by about 37 million over the next decade and that humanoids might offset as much as 60 percent of that shortfall by 2035 — a stock of up to 24 million machines, or roughly 4 percent of the present labor force. Summaries of that note ran at The Next Web and Bloomberg.
The Financial Times made the same demographic point in June: the UN path has China’s 15-to-64 population falling toward 300 million by 2100 from a peak near one billion. Factory managers quoted in that reporting do not sound ideological. They sound tired. One Sany executive called replacement of humans with robots in labor-heavy shops “inevitable.” See the FT line as carried in The Australian Financial Review’s reprint.
There is a second labor story that does not fit the shortage poster. Some eastern-China factory owners told diaspora and opposition-leaning outlets that they adopt robots because people are expensive once pensions, housing funds, and overtime are counted — not because the gates are empty. Unemployment among the young and a cooling gig economy can coexist with a shortage of people willing to do a particular shift. Both can be true in the same province. The policy tension is obvious. Beijing’s 15th Five-Year employment plan still talks about keeping surveyed urban unemployment inside 5.5 percent and about training millions of workers. A plant that needs fewer assemblers does not automatically create the technicians, safety officers, and repair crews those assemblers are supposed to become.
The claim that humanoids will take “up to 20 percent of factory workload” appears in secondary commentary around this opening. Treat it as an ambition or a scenario, not as a measured share of Chinese manufacturing hours in 2026. Workload share depends on the task. Moving a tote is one problem. Routing a wire harness through a moving torso is another.
The geopolitical frame: hardware scale versus software bets
It is fashionable to call this a “robotics cold war.” The phrase is lazy and partly accurate. Export controls on advanced chips already treat compute as a strategic good. Humanoids sit at the junction of motors, batteries, cameras, and models. The state that can make the body cheaply and the state that can make the model general will each claim they are ahead. They will be measuring different races.
On physical volume, China is not hiding. Market trackers cited in mid-2026 put Chinese makers at the vast majority of global humanoid shipments. Planet Today’s own August file on the Beijing humanoid games noted official talk of domestic output above 100,000 units in 2026 and third-party data putting Chinese firms above 97 percent of first-half global shipments. That piece is 2,056 Humanoid Robots Hit Beijing Games: What the Medals Won’t Show. Contests and factories are not the same evidence, but they rhyme: China is putting metal on floors at a pace others have not matched.
On the other side of the ledger sit U.S. and European bets: foundation models, custom silicon, and a smaller number of high-cost prototypes aimed at general tasks rather than a single factory station. Siemens itself is not only a vendor to Shenzhen. It has run its own humanoid logistics pilots in Germany. Bosch, under pressure in auto parts, has said it will push sensors and training data for humanoids — a European industrial answer to the same labor math. That thread is in Bosch Pivots to Robots: $19.2B Sensor Market Opportunity Emerges.
A German software house helping a Chinese humanoid maker hit 10,000 units a year is itself a geopolitical fact. It is also a commercial fact. Siemens sells digital factories. UBTECH needs a digital factory. Washington can tighten chip rules. It cannot easily ban a Plant Simulation license after the hall is already running. Readers who want the wider argument about who writes the rules for advanced AI, rather than who welds the ankle, can set this plant beside King Charles AI Summit in Scotland: Who’s Coming and Why. One story is about dignity charters. The other is about takt time. They collide in the same decade.
None of this requires a morality play. A country with fewer young assemblers will automate. A country that fears losing the next industrial platform will subsidize the machines. A country that sells the software layer will take the contract. Those are incentives, not hidden hands.
Can the market absorb 10,000 humanoids a year?
This is the curiosity gap that the opening videos skip. Building the machine is no longer the rarest skill in this industry. Finding a job the machine can do for less than a person, for long enough to pay back the integrator, still is.
Morgan Stanley and others spent 2025 and 2026 raising China shipment forecasts — from the low tens of thousands toward 50,000 in some 2026 notes — then warning that a large share of early units still go to research, education, and pilots rather than to unsupervised production. Reuters Breakingviews, in June, pointed at another risk: too many firms, too much subsidy, official talk of a bubble among some 150 humanoid makers. Capacity announcements are how a bubble looks from the inside. They are also how a real industry looks just before it finds a standard product. History does not sort those two on opening day.
Price is the quiet variable. Specialist reporting says UBTECH wants unit production cost under $20,000 in the longer run, with more than 90 percent of the supply chain inside China. At that price a factory manager can do arithmetic a banker will recognize. At three or four times that price the same manager keeps the temp agency on speed dial. Safety regulators in Europe and North America will not move at Chinese factory speed. Insurance, liability, and union rules will slow imports even if the hardware is ready. The global market can “absorb” 10,000 units only if someone signs for them after the trial period, not when they leave Liuzhou.
There is a feedback loop worth watching. If the robots that build robots are also the product, every design change that removes a fastener or shortens calibration pays twice: once on the line, once at the customer. That is the optimistic industrial case. The pessimistic case is a warehouse of finished Walkers waiting for software that can handle a bin that was moved overnight.
What is settled, and what is still open
Settled, as of 16 September 2026:
- UBTECH says the Liuzhou industrial humanoid plant began production on 12 September 2026.
- Designed capacity is above 10,000 units a year; designed takt is one unit every ten minutes.
- The hall is about 14,000 square meters and 13.8 meters high.
- Primary models named are Walker S and Cruzr industrial platforms.
- Siemens Digital Industries Software is the digital-manufacturing partner; the March 2026 agreement set the 10,000-unit target.
- Company materials describe Cruzr units, AGVs, unmanned forklifts, four-plus hours of end-of-line testing, and a 65-square-meter automated warehouse for 112 finished robots.
- UBTECH marked its 1,000th Walker S2 at Liuzhou on 26 December 2025 and said more than 500 units had been delivered by then. See the earlier Global Times report of 30 December 2025 and the company December 2025 post.
Still open:
- Sustained monthly output at the advertised takt, after maintenance, model changeovers, and rework.
- How many 2026 units will be accepted by paying customers rather than used inside UBTECH or in subsidized pilots.
- Net job effect in Liuzhou and at customer plants once technicians and supervisors are counted.
- Whether export markets will certify these machines on a timeline that matches the production hall.
- Whether unit economics turn positive before the next capacity announcement.
A note on welfare, without a sermon
If humanoids take the worst stations — heat, night, heavy lift, repetitive strain — that is a gain for the people who no longer stand there. If the same machines arrive faster than retraining, pensions, and local services, factory towns will feel it as a wage shock, not as progress. China is not unique here. Every industrial country that aged before it automated had a version of this argument. The difference is speed. A plant designed for 10,000 bodies a year compresses a transition that used to take a generation.
Readers who want the human side of China’s blue-collar shift, rather than the robot side, already have a long paper trail in English-language business press this year: gig work cooling, migrant hours still long, insurance coverage thin. Those files do not disprove the shortage story. They show why “just move them into services” is a slogan, not a plan.
How to read the next six months
Ignore the next opening ceremony. Watch three numbers.
First, accepted deliveries, not designed capacity. A quarterly unit figure that a customer has signed for is worth more than a takt-time video.
Second, hours of unsupervised work per robot per shift. A machine that needs a person beside it is a very expensive fixture.
Third, service cost. The first year of a humanoid fleet is a maintenance story wearing a technology costume.
If those three move the right way, Liuzhou is a template and other firms will copy the digital-twin-plus-mobile-manipulator layout. If they do not, the world will still have a striking factory and a familiar glut. Both outcomes are compatible with the facts published this week. That is why the honest headline is not that the robots have taken over. It is that a production system built to make them in volume now exists, and the bill for that system will come due in orders, not in adjectives.
Related on Planet Today
- 2,056 Humanoid Robots Hit Beijing Games: What the Medals Won’t Show — shipment share, official output talk, and the gap between contests and paid work.
- Bosch Pivots to Robots: $19.2B Sensor Market Opportunity Emerges — the European industrial answer to the same labor and sensor bottleneck.
- King Charles AI Summit in Scotland: Who’s Coming and Why — who wants to write the rules while others write the takt time.
Primary sources
- UBTECH / Global Times factory report, 14 September 2026: https://www.globaltimes.cn/page/202609/1370528.shtml
- UBTECH X announcement, 15 September 2026: https://x.com/UBTECHRobotics/status/2099830594817646663
- Humanoids Daily factory technical brief, 13 September 2026: https://www.humanoidsdaily.com/news/inside-ubtech-s-10-000-unit-super-factory-how-robots-are-building-humanoids-in-liuzhou
- Black Scarab capacity-versus-output analysis, 13 September 2026: https://www.blackscarab.ai/news/ubtech-liuzhou-humanoid-robot-factory
- India Today wrap, 16 September 2026: https://www.indiatoday.in/technology/news/story/china-opens-new-humanoid-factory-where-robots-build-new-robot-every-10-minutes-2996063-2026-09-16
- Shenzhen Daily / Nanshan on the March 2026 Siemens agreement: https://www.szns.gov.cn/english/news/content/post_12690271.html
- Interesting Engineering on the factory, 14 September 2026: https://interestingengineering.com/ai-robotics/ubtech-starts-10000-unit-robot-factory
- Barclays labor-offset research as reported 19 May 2026: The Next Web
Original source for this file: company and state-media announcement cycle of 12–15 September 2026, anchored on UBTECH’s Liuzhou commissioning and the Global Times report of 14 September 2026 (author line: Liu Yang and Li Xuanmin), with the company’s 15 September 2026 X post as the primary social primary source. Full URL: https://www.globaltimes.cn/page/202609/1370528.shtml
Disclaimer for fact-checkers. This article separates designed capacity from verified annual output. Core physical specifications (site, size, intended takt, Siemens software role, Cruzr logistics roles, four-hour test window, 65 m² warehouse) come from UBTECH statements as carried by Global Times and corroborated by multiple trade desks. Global Times is a Chinese official English-language paper; it is a primary carrier of the company line, not an independent audit of the line’s yield. Western technology sites largely republished those specifications in the first 72 hours and should not be treated as a second measurement. Black Scarab and Humanoids Daily add useful caveats on finances and takt-versus-throughput; they are specialist outlets with their own frames. Siemens is a commercial party to the software stack. Barclays figures on labor offset are model outputs, not government census data. No claim in this piece should be read as a forecast that 10,000 humanoids will ship, sell, or replace a fixed share of factory hours in 2026.
