Inside China’s robotics revolution
A journalist travels across China visiting robotics companies to assess how close the country is to deploying humanoid robots at industrial scale. China's robotics boom is driven by deep learning advances, massive state investment, and a dense manufacturing supply chain, with companies racing to automate factory work currently performed by hundreds of millions of workers. The piece explores the technology's limits, the human cost of automation, and the paradoxical interdependence between Chinese and American industry.
Summary
The piece opens with Chen Liang, founder of Gucci Robotics in Shanghai, who builds automation machines for Chinese car brands like BYD and NIO. His company focuses on the hardest part of car manufacturing — final assembly — and has already eliminated the need for human workers in wheel, dashboard, and window installation. Chen estimates 80% of final assembly remains to be automated, and that is his long-term goal. A General Motors team was observed at Gucci's warehouse testing machines destined for a Canadian plant, where the purchase would eliminate 12 assembly operators. The GM engineer on site, tasked with hitting annual job-reduction targets, chose Gucci over a German competitor because only Gucci could accommodate a moving assembly line.
The journalist then travels to Beijing to observe a meeting between Chen and Galbot, one of China's most prominent humanoid robotics startups. Galbot develops Vision Language Action Model (VLA) robots — machines designed to operate in fluid, unfamiliar environments rather than following pre-programmed scripts. Their robots have been deployed in pharmacies dispensing medication and are being tested in retail and factory settings. A key bottleneck for this technology is the scarcity of training data, collected either through teleoperation (humans manually guiding robots through tasks) or virtual simulation environments. Chen proposed a collaboration to train Galbot's humanoids to perform screw-driving in under eight seconds — a task trivial for humans but requiring numerous micro-decisions for an unscripted robot.
The piece then visits Unitree, the world's leading humanoid robot shipper by volume, whose robots went viral performing at the Lunar New Year Gala and a pop concert. Unitree's robots start at $1,600, compared to tens of thousands for American equivalents. This price advantage is attributed to China's dense hardware supplier ecosystems in the Yangtze and Pearl River Deltas, where prototype tweaks can happen in under a day. However, Unitree faces scrutiny over potential military connections — Chinese state TV showed its robot dogs equipped with machine guns during military drills, and US lawmakers have suggested cutting off the company's access to American semiconductors. Unitree denies selling to the military.
The journalist then visits Lejiu Robotics' training center in Beijing, where roughly 100 teleoperators in their late teens and early twenties guide robots through hundreds of repetitive tasks per shift to build training datasets. These workers are recruited through labor dispatch companies from vocational colleges and earn around 6,000–10,000 yuan per month. The journalist personally tries teleoperation at Saibot, a dexterous robotic hands startup, discovering it requires humans to move at machine-registerable speeds and suppress natural human behaviors to avoid corrupting data — in essence, humans must act more like robots to train robots to act more like humans.
A visit to a Huawei electric vehicle factory in Hefei shows Chen's automation machines in action, with robotic arms installing dashboards in seconds. Chen observes that even pick-and-place tasks remain too complex for current humanoids due to the variety of component types and constantly changing parts. He believes final assembly will be close to fully automated by the mid-2030s. When asked about the social consequences for the 120 million Chinese factory workers, Chen is detached — he suggests higher-skilled workers could train robots and that lower-skilled workers should simply change careers.
The piece frames the broader geopolitical and economic context: China accounts for over half of global new factory robot installations annually, and state investment is deeply integrated with private enterprise. Xi Jinping's government has moved away from market-driven innovation toward party-directed technology priorities. Cities compete to attract robotics startups, acting as patrons backing local champions. Meanwhile, the US-China decoupling narrative is complicated by the reality that American manufacturers like GM depend on Chinese automation suppliers, and Chinese firms like Gucci value American customers partly because they pay on time. The journalist concludes with a speculative framework: the US may lead development of general-purpose humanoids while China floods the world market with cheap, reliable single-task robots.
Key Insights
- Chen Liang argues that young Chinese workers are increasingly unwilling to do factory work, which he uses to counter the argument that automation destroys desirable jobs — his implication being that even in China, where factory culture is deeply ingrained, the labor supply for such work is drying up voluntarily.
- A General Motors engineer revealed that GM sets explicit annual job-reduction targets requiring the elimination of a set number of factory workers across North American plants, framing automation as a corporate mandate rather than an incidental byproduct of efficiency.
- Galbot's pick-and-place robots are described as more commercially significant than acrobatic humanoids because acrobatic robots follow pre-programmed instructions and cannot operate off-script, whereas VLA-based robots aim to handle unfamiliar environments — a distinction the piece argues is poorly understood by the public.
- Unitree's robots start at approximately $1,600, compared to tens of thousands of dollars for comparable American machines, an advantage attributed not to wage differences but to the density of hardware suppliers in China's manufacturing clusters, where prototype iteration can take less than a day.
- The journalist argues that a plausible future division of labor between the US and China in robotics is one where the US leads development of general-purpose humanoids while China supplies the world with cheap, reliable single-task robots — with Chinese companies pulled toward commercialization and American ones insulated by deeper venture capital.
- Teleoperation workers at Lejiu Robotics are sourced through labor dispatch networks from vocational colleges, performing 15 tasks per day, 10 times each on 8-hour shifts, earning wages comparable to delivery drivers — representing a new invisible labor class whose work enables AI training but whose contributions are largely unacknowledged.
- Chen Liang acknowledged that the 120 million Chinese factory workers and their successors currently in vocational training for advanced manufacturing 'definitely need to change careers,' but offered no concrete plan for lower-skilled displaced workers, reflecting a broader detachment in the Chinese robotics industry toward the social consequences of automation.
- The US-China decoupling narrative is complicated by on-the-ground reality: American manufacturers actively source Chinese automation equipment, and Chinese firms like Gucci value American clients not only for market access but because, as Chen noted, 'Americans pay on time' — contrasting with Chinese clients who frequently delay payments and demand impossible timelines.
Topics
Full transcript available for MurmurCast members
Sign Up to Access