In mid-October, four seemingly scattered pieces of news pieced together to form the same curve.
In Changzhou, the Nanpeng precision reducer project for humanoid robots has been completed, and the new factory has entered trial operation. With 125 new processing and auxiliary machines and 2 automated production lines added, the annual production capacity will reach 300,000 units/sets of precision planetary reducers and high-precision harmonic reducers upon commissioning. Jin Jianpeng, Assistant to the General Manager, stated verbatim: "Currently, the company's orders on hand are fully booked until the end of the year, and there is an urgent need to release our full production capacity."
In Foshan, Jihua Qujie's robot joint module production line with an annual capacity of 100,000 units has been put into operation. The designed takt time is 90 seconds per set, with a target first-pass yield rate of over 96%. It took this company seven years to go from laboratory project initiation to scaled mass production.
Leadshine reported a series of impressive figures during its investor relations activities: the annual orders on hand for frameless torque motors exceed 1 million units, with a monthly production capacity of 150,000 units, and it is advancing towards an annual production of 3 million units; orders on hand for three types of joint modules—harmonic, planetary, and integrated wheel—exceed 60,000 units, with deliveries accelerating and scaling up.
At the China International Industry Fair in Shanghai, Welling Robotics, a subsidiary of Midea, showcased its full-stack self-developed core components, while also presenting a set of capacity data: the production line automation rate exceeds 80%, with an annual delivery of 200,000 reducers (330,000 in progress), 60,000 joint modules (150,000 in progress), and 210,000 servo motors.
Reducers, joint modules, and motors—the three core components of the humanoid robot "body"—were pushed to the critical point of capacity realization within just one week.
However, while the supply curve for components is rising steeply, the curve for complete machine production has not followed the same slope.
01. Complete Machine Production Forecasts Stall at the "100,000 Units" Scale
Regarding the output of complete humanoid robots in 2026, figures from various institutions vary, but the scale highly converges.
According to statistics in the "2026 Humanoid Robot Industry Development Report," in the first half of 2026, China's humanoid robot shipments exceeded 40,000 units, with its global share further increasing to 97%. Earlier, the Ministry of Industry and Information Technology officially predicted that China's humanoid robot production would exceed 100,000 units for the entire year of 2026.
Now, let's put the component-side figures on the same table.
Summing up the disclosed capacities of just Nanpeng (300,000 reducers), Welling (330,000 reducers in progress + 150,000 joint modules), Jihua Qujie (100,000 joint modules), and Leadshine (estimated annual delivery of 800,000 to 900,000 frameless torque motors, with over 60,000 joint modules on hand), the annual capacity reserve for reducers has easily surpassed the 600,000-unit scale, joint modules have exceeded 300,000 units, and motors are heading straight for the million-unit level. A standard humanoid robot typically requires 40 to 50 degrees of freedom for movement, with each degree of freedom corresponding to a set of reducer + motor + joint module. Based on this conversion relationship, the current component-side capacity reserve corresponds to a complete machine "body" supply capacity that is already several times or even more than ten times the expected shipment volume of complete machines.
Even if overseas procurement is included, their nascent complete machine production will not significantly consume China's component capacity. Moreover, the above four companies are merely a microcosm; there are far more than just these four component manufacturers accelerating capacity expansion domestically.
Leaderdrive provides a more intuitive footnote. Its Deputy General Manager, Chu Jianhua, revealed in an interview that the shipment volume of harmonic reducers in 2025 is close to 500,000 units, and "this year's capacity planning is being prepared for nearly 1 million units."
02. The Bottleneck is Not in the "Body", but in the "Brain"
There is already a repeatedly used phrase in the industry: "strong body, weak brain."
Xinhua Finance directly pointed out in a research report that over the past two years, the technology at the "body" level of humanoid robots has become increasingly mature; dancing, boxing, and running are no problem, and they have even started "working" in automobile factories, but their "intelligence remains limited." Sun Bo, Director of the Smart Brain Innovation Center at the Shanghai AI Research Institute, breaks down "strong body, weak brain" into three dimensions: autonomous decision-making, generalization ability, and real-time performance—low accuracy in recognition and grasping, slow execution of movements, and the need for massive retraining when switching to a new scenario.
The interview with Securities Times is even more straightforward. A person in charge of the robotics sector at a listed company said that the current mainstream model still involves collecting data and training models for specific scenarios, leaving robots trapped in the dilemma of "failing when the scenario changes." The robots on the Spring Festival Gala stage demonstrate physical control within the scope of the "cerebellum," rather than the interaction and operational capabilities that the "brain" should possess. "For different tasks such as delivering components, driving screws, and assembly on the production line, as well as the vastly different environments and needs in home scenarios, it is simply impossible to achieve full-coverage data collection; even slight changes in the scenario require retraining the model."
Reducers and joint modules solve the problem of "whether it can move and how accurately it moves," while "whether it can judge on its own what to move and how to move" is a problem on another level. The former is a precision manufacturing issue, and the Chinese supply chain has repeatedly verified its capabilities in industrial robots and automotive components over the past twenty years; the latter is an algorithm and data issue for embodied intelligence, with no ready-made scalable path to transfer.
Thus, a rare industrial sequence has emerged: the capacity construction of upstream components is running ahead of the commercialization pace of complete machines. Nanpeng's "orders on hand booked until the end of the year," Leadshine's "million-unit motor orders," and Jihua's "90-second takt time" reflect the supply chain's bet on the "first year of mass production," rather than the matched shipment scale already realized on the complete machine side.
03. The Window Period for Component Suppliers is Being Squeezed by Three Forces
Advanced capacity reserve is not a problem in itself. The issue is that during the window period of advanced reserve, the competitive landscape is tightening at an accelerating pace.
The first pressure comes from the self-research of complete machine manufacturers. The prospectus of Unitree Robotics discloses that over 90% of the company's components are self-developed, with total external procurement in 2025 amounting to only CNY 794 million, of which more than half are low-value "iron parts" such as joint housings and structural components; high-value components like motors, reducers, and controllers are basically kept in-house. Zhiyuan has spun off its dexterous hand business to operate independently as "Critical Point," directly entering the field to compete with suppliers; the depth of self-research for Tesla's Optimus is also continuously increasing. Potential major customers are simultaneously potential biggest competitors—this relationship has already taken shape in the component segment.
The second pressure comes from the influx of cross-industry capacity. Haoneng Shares is investing CNY 1 billion to build a capacity of 5 million reducers, Midea is planning a 300,000-unit production line, and Zhongding Shares has raised CNY 1.9 billion to enter the market. "This model of 'betting huge amounts' is exactly the same as the capacity race in the power battery industry back in the day." When the capital density on the supply side reaches this extent, it is only a matter of time before what is scarce today becomes oversupplied tomorrow.
The third pressure comes from the transmission of complete machine price cuts to the supply chain. The average unit price of Unitree's humanoid robots has dropped by 72% over two years, falling to CNY 166,000 per unit, but the gross profit margin has instead risen to 63.18%—the room for price cuts is "not squeezed out of suppliers, but extracted from self-research." When complete machine manufacturers place the main battlefield for cost reduction on self-research substitution rather than supplier bargaining, if component suppliers only provide "substitutable standard parts," their profit margins will be systematically compressed.
04. The Real Match Point: From "Selling Shovels" to "Irreplaceable"
Returning to the capacity figures at the beginning of the article. 300,000 reducers, 100,000 joint modules, and 1 million motor orders, set against the backdrop of an unlocked "brain" and complete machine shipments still climbing at the scale of tens of thousands of units, read more like a capacity race rather than natural expansion driven by demand.
This does not mean that component investment is wrong. The "body" of humanoid robots is a definite hardware foundation, and the number of degrees of freedom, precision requirements, and reliability standards will only go up. The issue lies in the pace and positioning.
For suppliers who have already secured their positions, the task during the window period is not to continue piling up capacity, but to transform "substitutable capacity" into "irreplaceable capabilities."
For complete machine manufacturers and investors, this set of mismatched data between capacity and shipments provides a sober reference: when the annual capacity of components has been rolled out at the million-unit level, while complete machine shipments are still hovering at the scale of tens of thousands of units, the focus of industrial contradictions in the next stage will most likely not fall on the supply of reducers or motors, but on the question of "who can make robots truly work on their own."
The takt time for reducers can be achieved at 90 seconds per set. But the speed at which a robot "learns" a new task is still far from reaching the stage of being measured in seconds.
This article is compiled based on publicly available online information, for reference only, and does not constitute investment advice.