Unitree's factory runs largely by hand, author says after tour

The author of the josef.cn blog toured Unitree's factory in Hangzhou on a Wednesday and was not allowed to take pictures. In the building that makes actuators, the motors that drive each joint, there were one or two new automated lines. The rest of the floor was filled with people operating old-school winding machines, which wrap copper wire around a motor's core. The author does not know what share of motors come off the automated lines. The essay's thesis: everyone thinks China's robot factories are automated, they are not, and so the robots are Chinese but the robots that make robots will be Western.

The author backs this with Unitree's own filing figures. Unitree has a viral video (28 million views) of a humanoid doing assembly in its own factory, which it calls "an actual deployment". Its IPO prospectus calls it a pilot and says assembly is "currently mainly manual". At the end of 2025 all of Unitree's machinery was worth about $1.7 million on its books, at a company that made almost 32,000 robots that year: some $50 of machinery per robot. Labour was 8% of the cost of making its robots in 2025, outsourced assemblers included, and most of the rest was parts. With those hands it made about 26,000 robot dogs in 2025, more than the roughly 20,000 that IDC reckons the whole world shipped the year before. Halve a cost that is 8% of the bill and you save 4%. Unitree expects the motor line it is building to cut the labour in each motor by more than half, but the cost of one robot model by under 2%. By the author's arithmetic, labour per robot had roughly halved since 2023, and the filing credits scale, standardised processes and more skilled staff, none of which is a machine.

The second reason is the product. The tour host said Unitree's CEO, Wang Xingxing, asked a G1 running GPT-6 to fetch a bottle of water from another room; it did, slowly. A Unitree slide puts robots' "ChatGPT moment", when they can do 80% of tasks they have never seen 80% of the time, two or three years out, and the slide is from last year. Until then, humanoids are mostly toys for researchers; the host named Nvidia, Amazon, Meta and universities. Three quarters of Unitree's humanoid revenue came from research and education in the first nine months of 2025, and under 5% from factories or other industrial work. The author says half or more of what Unitree files as "industry" is robots giving guided tours at companies. Asked what the GD01 rideable mecha, from $650,000, is for, the answer was that Wang had wanted one since childhood. A custom machine only works on a frozen product, so one built for this year's robot is scrap next year. The exception is the motor, the part robots share most, which is why Unitree is automating its joint modules first. The author saw the same pattern at a drone factory, where circuit boards and motors were machine-made and final assembly was by hand. On the same Wednesday, Wang told a conference in Wuzhen that Unitree had spent ten years on the robot's body and would now put its weight on the brain.

The author concedes that his May claim that China's new factories "bolt robots into every one" was not true of the plant visited, but still believes the West should "erase the factory floor", and left Hangzhou more bullish on the West. The next day at BYD, the author saw a car plant where everything seen was automated.

History is the guide. Ford's first moving line at Highland Park in 1913, which put magnets on Model T flywheels, split the work into 29 operations, each worker doing, in Ford's words, "as nearly as possible only one thing with only one movement". Machines came once the Model T was frozen. When the Model A replaced it in 1927, Ford shut its plants for six months and rebuilt or scrapped most of its machine tools, and Chevrolet outsold it for the first time; a year later GM changed a Chevrolet model in three weeks. The author says Unitree, with a new robot every year, is GM in 1927, not Ford. In 2011 Foxconn's Terry Gou promised a million robots within three years; by late 2016 Foxconn had about 40,000 and said it was slowing down "because humans are cheaper and more flexible to adapt to an ever-shorter life cycle of consumer products". Programming and integrating a robot is typically 50 to 70% of the cost of the application, and GM's robots at Hamtramck in the 1980s reportedly attached Buick bumpers to Cadillacs because they could not tell the cars apart. When hands get too pricey, a factory has three doors: organise them better, find cheaper hands elsewhere, or automate the work away, which has always needed a frozen part.

The author argues America shuts the cheap doors. An hour of factory work costs about six times as much in America as in China (the author's estimate), so at American wages labour would be about 40% of Unitree's costs, not 8%. The motor line expected to save 1.7% of one robot model's cost would save about 9% at American wages. Precedent: after about half a million Mexican farmworkers were shut out in 1964, the share of California tomatoes picked by machine went from 20% in 1965 to 99.9% in 1970, and the tomato was bred tougher and squarer to fit. Policy is closing doors now. In July the FCC stopped approving new humanoids and robot dogs unless made in America with more than 65% American parts by cost (75% from 2029). From August 2027, stators of drones bought by the Pentagon programme must be wound outside China and a handful of other countries.

Western volume is arriving, the author says: Figure claims its BotQ factory went from one humanoid a day to one an hour in less than four months; Tesla says most of a line to build a million Optimus robots a year is at its Fremont plant; Agility, which makes Digit humanoids in Oregon, listed on Nasdaq on 5 November with Foxconn leading about $201 million of the money, though it sold just $1.8 million of robots last year; and 1X says it makes its own motors, from a spool of copper to a finished actuator, on automated machines built in Hayward, California. America cannot yet build the machines itself: on 28 October the White House found that it "relies on foreign imports for most types of robotics, industrial machinery, and their parts and components". Machine builders are moving in, though: Mazak has built 40,000 machines in Kentucky, Yaskawa will make robots in Wisconsin and Fanuc has bought a site in Michigan.

AI is cutting the programming bill, the author argues. HD Hyundai and Palantir say AI agents cut the average time to fix a robot-programming issue from 21.5 days to 7.9. Apple estimated up to nine months to hire 8,700 industrial engineers in America versus 15 days in China. An engineer's time is three to five times more expensive in America, so each hour AI saves is worth that much more. The author calls the robot version "agentic robotics": an off-the-shelf model writes the robot's program, runs it and fixes it, on any task a sensor can check. Magnets are the other gap: MP Materials has finished qualifying its magnets with GM and its Texas plant begins deliveries before the end of 2026, and BMW makes an EV motor in Steyr without permanent magnets. The warning: China already leads in solar-panel equipment and many electric-car motor machines, and Chinese hairpin winding lines have caught up with German and Italian ones, undercut them by as much as half and begun exporting to Europe. At American wages even a winder at twice the price still saves 8 to 9% of a robot's cost; what the West cannot afford is one that is slow to change over.

Key facts

• The author toured Unitree's Hangzhou factory and saw one or two new automated lines in the actuator building; the rest of the floor was people operating old-school winding machines.
• Per Unitree's filing as read by the author, machinery was worth about $1.7 million on its books at the end of 2025 against almost 32,000 robots made (some $50 per robot), and labour was 8% of production cost.
• Unitree expects its motor line to cut labour per motor by more than half but the cost of one robot model by under 2%; the author estimates the same line would save about 9% at American wages.
• The author argues that humanoids are mostly research toys (three quarters of humanoid revenue from research and education, under 5% from industry), so a frozen product to automate around does not yet exist.
• The author's case for the West: labour-scarce America, FCC and Pentagon rules shutting out Chinese hardware, and AI cutting robot-programming costs, though the figures at American wages are the author's own estimates.

Why it matters

The common picture is that China's robot makers run on automated lines. This first-hand account, backed by Unitree's own filing figures, says the opposite for one of the best-known firms: about $50 of machinery per robot, labour at 8% of cost, assembly "currently mainly manual". If that holds, the factory automation of robots is still unclaimed. The author uses Ford, GM and Foxconn to argue that machines arrive when a product freezes and falls in cost enough, and that AI is lowering the programming cost that once made custom machines pay only for frozen products.

Who it affects

Western machine builders and robot makers, such as those named in the essay (Figure, Tesla, Agility, 1X, Mazak, Yaskawa, Fanuc), are the audience of the author's argument. It also bears on Chinese robot makers like Unitree, on policymakers weighing rules such as the FCC approval limits and the Pentagon drone stator rule, and on anyone tracking where humanoid robots will be built and who builds the equipment.

How to use it

Read it as a framework rather than a how-to. The author's test for where automation arrives: is the part stable (like the motor, the part Unitree's robots share most), and how expensive is labour locally? The essay lists three doors for a factory facing dear labour: organise people better, find cheaper hands, or automate the work away. Watch the figures the author cites as signals: the share of humanoid revenue from industrial work, Western factory ramp-ups, and how fast AI agents shorten robot-programming time.

How solid is it

It is an opinion essay based on one visit, and the author says photography was not allowed. The Unitree numbers (machinery book value, robot count, labour share, revenue split) are given as coming from Unitree's filing and prospectus. The labour-per-robot halving is the author's own arithmetic. The American-wage figures (six times hourly cost, about 40% labour share, about 9% saving) are the author's estimates. The Figure, Tesla, 1X and HD Hyundai/Palantir claims are reported as the companies' own statements, with no independent confirmation given.

Risks and caveats

The author does not know what share of Unitree's motors come off the automated lines, and one or two lines were seen on one floor. The author concedes the earlier May claim about China's factories bolting robots into every one was not true of this plant, and reports that BYD's car plant, visited the next day, was fully automated in everything seen. The case that the West will win is a forecast resting on assumptions: that AI keeps cutting programming costs, that policy keeps shutting out Chinese hardware, and that Western firms build the machines. The author also notes that China already leads in solar-panel equipment and has caught up in hairpin winding lines, which is how the West could lose.

“Everyone thinks China's robot factories are automated. They aren't.”

— Author of the josef.cn blog, subtitle of the essay