I. The Year of Moving from Exhibition Booth to Production Line
Humanoid robots completed an identity shift in 2025: from performers on tech exhibition stages to an industry with shipment volumes, financial statements, and IPO prospectuses.
Shipment figures vary by methodology, but the magnitudes agree. IDC's January 2026 report puts 2025 global shipments at roughly 18,000 units, up about 508% year on year, with Agibot first at roughly 5,200 units and Unitree Robotics second at roughly 4,700—both of the top two are Chinese companies. Unitree's IPO prospectus, for its part, discloses that its 2025 humanoid robot shipments exceeded 5,500 units, with the average price of its flagship G1 model down to 167,600 yuan. Which of the two ranks first remains a matter of statistical methodology, but that Chinese vendors lead the world in shipments is not in dispute.
Capital markets priced this in at the same time. Unitree Robotics obtained its STAR Market registration approval in early July 2026—just 73 days from acceptance to approval—and will become the "first humanoid robot stock"; its FY2025 revenue was 1.699 billion yuan with a net profit of 278 million yuan, a rare profitable specimen in the industry. Overseas, Figure AI completed a financing round of more than 1 billion dollars at a 39 billion dollar valuation in September 2025, and its mass-production plant doubled monthly shipments back to back, from 60 to 240 units, in the first half of 2026.
For 2026, TrendForce expects global shipments to grow more than 700% year on year and surpass 50,000 units, calling it the inflection year for commercialization; the firm also expects China's humanoid robot output to grow 94% in 2026, with Unitree and Agibot together holding nearly 80% of the share.
II. Where the Forecasts Diverge—and Agree
On how large this market can grow, the divergence among investment banks deserves to be presented as it is.
Goldman Sachs forecasts a global humanoid robot market of 38 billion dollars and shipments of 1.4 million units by 2035; Morgan Stanley looks further out and is more aggressive, forecasting a global market of more than 5 trillion dollars and an installed base of more than 1 billion units by 2050, with 90% of them in industrial and commercial settings; Citi's model estimates roughly 648 million humanoid robots at work by 2050. Among domestic institutions, CICC's June 2025 research report estimates China's long-term potential market at more than one trillion yuan.
The numbers diverge wildly, but three points of consensus are clear. First, factories and warehouses will come before homes—Goldman Sachs expects nearly all of the 250,000 units shipped in 2030 to be for industrial use. Second, the cost curve is the lifeline of industrialization—Morgan Stanley expects unit prices to fall from about 200,000 dollars in 2024 to 50,000 dollars in 2050, while data from GGII (Gaogong Robotics) show that by the first quarter of 2026 the unit cost of domestically made complete machines had already fallen to roughly 100,000 yuan, down 33% from 2025. Third, whichever firm's model one takes, the Chinese supply chain is a key variable in the cost-decline assumptions.
Policy is ramping up in parallel. "Embodied intelligence" was written into the Government Work Report for the first time in 2025 and was then listed as a future industry in the recommendations for the 15th Five-Year Plan; the national venture capital guidance fund, backed by 100 billion yuan of central government funding, launched in December 2025, and Beijing, Shanghai, and Shenzhen have successively set up industry funds at the 100-billion-yuan and 10-billion-yuan scales, with Hangzhou and Suzhou following suit.
III. Components: The Capacity Race Outside the Spotlight
Complete machines absorb most of the attention, but the heavy lifting of industrialization happens in components. A humanoid robot needs dozens of rotary and linear joints, and the corresponding screws, reducers, motors, and sensors are setting off a round of capacity competition in China's component factories.
Planetary roller screws are the segment with the greatest upside elasticity. Previously highly dependent on overseas supply—two Swiss companies once held more than half of the global market combined—a single screw was priced in the tens of thousands of yuan. Since 2025, announced investments around screws by A-share listed companies have reached the tens of billions of yuan: Wuzhou Xinchun is planning capacity for 980,000 sets of planetary roller screws in Xinchang, Zhejiang, and Beite Technology plans to invest 1.85 billion yuan in an R&D and production base in Kunshan. Through reverse engineering and process improvement, domestic manufacturers have already pushed the cost of some inverted roller screws down to the thousand-yuan range.
Harmonic reducers show a "one dominant, one strong" structure: Harmonic Drive Systems holds roughly 80% of the global market, and Leaderdrive about 10%. The ramp-up of humanoid robots is rewriting the growth curve—Leaderdrive's 2025 revenue was 570 million yuan, up 47% year on year, and its new fundraising project will add capacity for 1 million harmonic reducers. By industry estimates, a single humanoid robot requires 20 to 40 harmonic reducers, a volume traditional industrial robots cannot match.
Frameless torque motors and dexterous hands are among the segments where localization is advancing fastest. Leadshine disclosed deliveries of more than 120,000 frameless torque motors in 2025, up more than 20-fold year on year; dexterous hands currently run 50,000 to 100,000 yuan apiece at mainstream prices, about 20% of complete-machine cost, and manufacturers such as ZWEI (Zhaowei) and Inspire Robots are pushing the price curve downward through the scale-up of micro screws and micro servo cylinders.
Six-axis force sensors are a sample where domestic substitution was completed ahead of the rest: domestic products are priced at roughly one-third of imports, and the localization rate of six-axis force sensors for collaborative robots approached 80% several years ago.
IV. Industrial Cluster Geography: Where the Orders Are Landing
Put these segments on a map, and a fairly clear division of labor emerges.
Xinchang, Zhejiang, drawing on the precision-machining foundation of its bearing industry, is becoming the concentrated landing point for screw capacity; Suzhou, Jiangsu, backed by a 200-billion-yuan robotics industry target, keeps doubling down on reducers, servos, and controllers, with more than 1,000 robotics and AI companies already in Wuzhong District; Shenzhen, Guangdong, has gathered more than 70 supply-chain companies spanning tactile sensors and dexterous hands through to complete machines, with plans for the cluster to exceed 1,200 companies by 2027; Dongguan, meanwhile, is producing a batch of startups in niche segments such as tactile sensors and direct-drive joint modules. Beijing's Yizhuang has assembled about 300 ecosystem companies around the national-local jointly built innovation center, while Shanghai's Zhangjiang supports the "Qinglong" open-source reference robot and about one-third of the country's robotics output value.
The implication of this map for upstream suppliers is direct: humanoid robots are not a "future customer segment"—they are already the present, real business of thousands of precision manufacturing factories in the Yangtze River Delta and the Pearl River Delta that are expanding capacity and placing purchase orders. Screw factories are buying grinding machines, reducer factories are buying inspection equipment, and dexterous hand factories are looking for suppliers of flexible materials and micro motors—every link's expansion along the chain means new orders for its own upstream.
V. Conclusion: Translating the "Inaugural Year" into Orders
The industry calls 2025 the inaugural year of mass production and 2026 the inflection year for commercialization. For industry research, the real meaning of these labels is this: for the first time, demand has turned from demonstration projects into repeatable procurement, and competition in components has shifted for the first time from "can it be made at all" to "whose capacity and cost are better."
In the phase of most drastic change, the information gap is the opportunity. Which factories are expanding screw capacity, which regions are accumulating dexterous hand suppliers, which OEMs' supplier lists are being enlarged—the answers to these questions are updated every quarter. You can search conversationally, by category and region, for the real factories along this supply chain in Tianxia Gongchang AI's database of 4.8 million real factories in active production.
Data Sources
The data in this article comes from public sources: shipment and cost data from IDC, TrendForce, and GGII (Gaogong Robotics); market forecast reports from Goldman Sachs, Morgan Stanley, Citi, and CICC; Unitree Robotics' IPO disclosures and Caixin reporting; official releases from Figure AI and UBTech; and coverage of component investments by 21st Century Business Herald and Securities Times. Vendor self-reported figures are noted as such; 2025 shipment volumes exist under multiple statistical methodologies, including IDC's and companies' prospectuses, and each is labeled separately in the text.