*This article is excerpted from the 'Humanoid Robot Contract Manufacturer Analysis Report' published by MIR DATABANK
In 2026, the shipment milestones of the humanoid robot industry are being rapidly refreshed: AgiBot's humanoid robots have cumulatively rolled off the production line 20,000 units (as of September), taking only about 6 months to go from 10,000 to 20,000 units; Yushu Technology's humanoid robots have cumulatively rolled off the production line 18,000 units (as of July), with a single G1 model shipping approximately 11,000 units; Youbixuan's Walker S2 has achieved thousand-unit-level deliveries (921 units). Compared with the corporate data on roll-offs and deliveries, the Ministry of Industry and Information Technology's (MIIT) forecast for the annual whole-machine output based on national production lines is even more ambitious: it is expected to exceed 100,000 units in 2026 (without limiting the models and application scenarios). However, an easily overlooked fact is that not all these robots are manufactured by the original equipment manufacturers (OEMs) themselves.
A humanoid robot mass production factory covering the entire process requires an investment of tens of billions to hundreds of billions of CNY. For the vast majority of robot companies, spending money on algorithms and product definition while entrusting manufacturing to professional contract manufacturers is a faster and more asset-light choice. AgiBot is exactly using this strategy to achieve the fastest volume growth rate in the industry within three years.
So, who is building robots for robot companies? Where are the barriers to entry in this business? How many seats are left? We have verified the announcements, financial reports, and investor relations records of multiple related companies in the industry chain, and the following sets of conclusions are worth discussing in detail.
Conclusions First
① The overall volume growth in the industry will simultaneously amplify contract manufacturing opportunities. Most new OEMs lack manufacturing assets. As they scale up from the thousand-unit level to the ten-thousand-unit level, borrowing external capacity is a necessary path.
② Division of labor follows a pattern. The closer to assembly, the higher the proportion of contract manufacturing: the depth of whole-machine contract manufacturing is extremely high, joint modules are relatively high, and components are low (module capacity expansion is faster than that of whole machines, so it is not surprising that structural overcapacity will emerge in 2027).
③ Some seats are locked through equity binding. The cooperation between Zhongding Gufen and Zhuji Dongli, as well as Lansi Keji and AgiBot, both include joint venture or investment clauses; pure commercial orders face the risk of being taken back in the later stage of volume growth.
④ Binding is only a ticket to enter, and long-term retention depends on capabilities. If the yield rate ramp-up, process finalization coordination, and cost reduction speed cannot keep up with the OEM iteration pace, the binding relationship will also be replaced.
01. The Manufacturing Strategies of Leading OEMs Have Diverged into Three Paths
Looking at the manufacturing arrangements of AgiBot, Yushu Technology, and Youbixuan together, the divergence is clear.
AgiBot follows an asset-light ecosystem: it masters algorithms and whole-machine design itself, outsourcing a large amount of structural parts, modules, and final assembly to external partners, while locking the supply chain with capital, cumulatively establishing 17 joint ventures and investing in 65 ecosystem enterprises. The cost of this system is the dispersion of bargaining power, in exchange for a roll-off speed of 20,000 units in three years.
Yushu Technology follows vertical self-production: over 90% of core components are independently developed and produced, unit costs are compressed to 62,200 CNY, and the gross profit margin of the humanoid robot business reaches 62.91%. Most new OEMs neither have this amount of cash nor this supply chain depth. Therefore, vertical self-production is the result of capability, not the starting point of the path.
Youbixuan follows M&A integration: mainly self-production, while spending 1.665 billion CNY to acquire a 43% equity stake in Fenglong Gufen to make up for manufacturing shortcomings.
Putting the three paths together, the conclusion is very clear: vertical self-production is the privilege of a few heavy-capital players. Most OEMs lack manufacturing assets, and entrusting contract manufacturing is their preferred path during the volume growth stage.
Data source: Public information such as company annual reports/prospectuses, compiled by MIR DATABANK
02. The Top Five Apple Supply Chain Players Are at the Forefront, and Actual Capacity Has Already Widened
The main force building robots for robot companies consists of five manufacturers that have crossed over from the consumer electronics Apple supply chain, and each of them already has verifiable shipment or mass production actions.
LINGYI iTECH completed the assembly of over 5,000 units/sets of humanoid robot hardware and whole machines in 2025, and established a joint venture, Lingzhi Chuangxin, with AgiBot. Lansi Keji shipped over 3,000 humanoid robots, and is triply bound with AgiBot through whole-machine contract manufacturing, joint venture, and strategic investment; it also undertook an order for 10,000 quadruped robots from Yuejiang Robot as the exclusive assembler. Luxun Jingmi shipped about 3,000 units in 2025, and the 2026 target has been upgraded to an expected annual delivery of over 10,000 units, adopting the OEM/JDM/ODM model. Everwin Precision entered the Figure AI supply chain, becoming a Tier-1 direct supplier of components. Gongyefulian disclosed in its annual report that it already has the scaled production capacity for key robot components and whole machines, and the relevant production lines will be introduced in 2025.
A more certain business than whole machines lies in joint modules. Joint modules account for 45% to 70% of the cost of a humanoid robot whole machine, and professional module manufacturers have begun to scale up: Quanzhibo shipped over 120,000 joint modules in the first half of 2026, and Yiyou Keji delivered 95,000 sets in 2025.
03. The Seats in the Contract Manufacturing Market Are Not Yet Fully Occupied; Depending on Their Origins, Their Strategies Vary
Since 2025, a large number of players have entered the humanoid robot manufacturing segment.
The first category is the spillover of automotive-grade manufacturing: Automotive parts manufacturers transfer automotive-grade quality systems and precision manufacturing capabilities into the robot industry. Zhongding Gufen and Zhuji Dongli jointly established Dingli Kexing to enter the whole-machine contract manufacturing of humanoid robots. Judgment: Quality systems and consistency management are hard advantages, suitable for modules and whole machines.
The second category is Electronic Manufacturing Services (EMS): EMS and ODM manufacturers output one-stop capabilities from design to mass production. Jabil has become the global manufacturing partner for the Apollo humanoid robot under the US company Apptronik; the leading mobile phone ODM manufacturer Huaqin Jishu is also promoting one-stop humanoid robot solutions externally. Judgment: Scale and yield rate are the strongest, naturally suitable for whole machines, but weak in robot electromechanical know-how.
The third category is equipment upward exploration: Automation equipment manufacturers use production lines and engineering capabilities to reversely cut into whole machines. Dongfang Yuanqi, a joint venture between Dongfang Precision Science and Leju Robot, has built an automated production line with an annual designed capacity of 10,000 units, which will be put into production in March 2026; Puzhi Weilai, a joint venture between Junpu Zhineng and AgiBot, has already undertaken whole-machine manufacturing. Judgment: Production lines and engineering capabilities can be directly reused, positioning at whole machines plus testing.
The fourth category is extension within the industry chain: Component and module enterprises already in the humanoid robot supply chain extend to whole-machine contract manufacturing by leveraging existing customer positioning. The representative is Taihu Robot, which extends from a joint module and dexterous hand supplier to whole-machine contract manufacturing; similar ones include Linghou, Guangyang, etc. Currently, many enterprises are in the early validation stage. Judgment: Customers and technologies are readily available.
The capability genes and binding methods of each path vary greatly, which also directly determines whether they can get orders and which segment of orders they can get.
04. Contract Manufacturing Opportunity Map: The Gap Is Not About Whether There Are Factories, But About Segments and Pace
By checking the manufacturing postures of major OEMs other than AgiBot, Youbixuan, and Yushu one by one, a universal phenomenon can be seen: having own factories does not mean manufacturing everything by themselves. The real gap in contract manufacturing lies in the segment position and volume growth pace. The essential role of contract manufacturing is, after the finalization of whole-machine design, software, and the 'brain', to use external production lines in parallel with own production lines to accelerate the roll-off speed and scenario landing.
Even if most manufacturers already have bases, the external manufacturing segments still truly exist, just falling in different positions. The Suzhou base of Xingdong Jiyuan has been built, and the model is 'components manufactured by others, whole-machine assembly independently', with component processing and production handed over to the supply chain; Yinhe Tongyong already has a signed base in Suzhou and three joint venture industrialization platforms in Wuxi, Suzhou Industrial Park, and Taizhou. The S1 heavy-duty robot is still mass-produced by Fenda Keji and delivered into the Ningde Shidai production line; Rongyao has no manufacturing assets, and 159 core structural parts are delivered in batches by LINGYI iTECH; Leju's whole machine is undertaken by the joint-stock joint venture Dongfang Yuanqi, and 82.82% of joint modules are purchased from Quanzhibo; Zibianliang's core components are fully independently developed, and the whole-machine manufacturing layout has not been publicly disclosed.
The level of capacity demand is ultimately determined by the order pace. Currently, the industry as a whole is in the early stage of order ramp-up, and the contract manufacturing volume is still at the scale of hundreds to thousands of units: Zhuji Dongli entrusts the joint venture Dingli Kexing to contract manufacture whole machines, with an actual contract manufacturing volume of 60 units in the first half of 2026; the Wuxi joint venture production line of Yinhe Tongyong plans an annual capacity of 1,000 units and will be put into production in the second half of 2026, while the company's cumulative orders have reached thousands of units. As orders are converted into delivery plans, the capacity demand for parallel multi-production lines will be simultaneously amplified, which is exactly where the incremental space for contract manufacturers lies.
Running in parallel with this is the expansion of self-built production lines: TARS signed a contract for the Changzhou Wujin R&D and manufacturing base in August, positioning it as the first scaled mass production center; the Zhongqing Shenzhen Honghualing base has been put into production, and the 10,000-unit-level production line planning falls in Zhengzhou. Parallel expansion of self-built and contract manufacturing is the norm during the volume growth period of this industry.
One more thing: binding does not equal lifetime. In addition to joint venture contract manufacturing, Zhuji Dongli is also simultaneously planning its own capacity. The internalization swing is not a hypothesis, but something that is happening. Overseas leading manufacturers generally adopt vertical integration, leaving limited external opportunities.
Conclusion
In the 10,000-unit era of humanoid robots, the apparent competition is about algorithms and products, but breaking it down one layer, it is a competition of manufacturing capabilities. OEMs are answering 'whether to build it ourselves or find someone to build it', while contract manufacturers need to answer 'why it should be me'.
Due to space limitations, this article focuses on sharing the conclusion level. The complete list of 30+ contract manufacturers, etc., is included in the MIR DATABANK 'Humanoid Robot Contract Manufacturer Analysis Report'.