China’s humanoid push is moving into production
Output, capacity, and supply chains are starting to take shape across the stack
Chinese humanoid companies are talking in the language of output, capacity, commissioning, and delivery, and the same shift is showing up underneath them across the supplier stack.
At the OEM layer, the signals are pretty clear. AgiBot said in late March that it had rolled out its 10,000th robot across all form factors. Leju is tied to an automated humanoid production line in Foshan with reported annual capacity above 10,000 units. Unitree separately said its humanoid shipments exceeded 5,500 units in late 2025, with total humanoid mass-production output above 6,500 units. None of that proves the market is mature, but it does show the conversation moving away from prototype novelty and toward repeatable industrial throughput.
Underneath that, a large and highly competitive manufacturing base is starting to gear around humanoid production.
Dexterous hands are one example. Xynova has said its Flex 1 hand secured orders for more than 10,000 hands from several top-tier customers, while the company targets annual capacity above 10,000 hands and 200,000 miniature electric cylinders. The customer list is still not clearly public. Even so, it is a supplier organizing for volume before the full downstream picture is visible.
Source: Xynova product line
Linear motion components show a similar shift, but with a different balance of evidence. Shuanglin said it delivered first samples of inverted planetary roller screws in June 2025, built a pilot line with actual annual output of 1,500 sets, and is preparing a first-phase mass-production line with capacity for 100,000 sets scheduled to start production in June 2026. It also said its miniature ball screws for dexterous hands had been sampled by several customers. These products were still in validation and had not yet secured formal orders. Xynova is an order and capacity signal with weak customer transparency. Shuanglin is a manufacturing-prep signal with clearer capacity language but weaker commercial confirmation.
A more grounded component example comes from Zhongding. Through its subsidiary Anhui Risibot, the company has been linked with LimX Dynamics, Chery Robotics, EngineAI, Fourier Intelligence, and Efort. More importantly, it has said it secured mass-production orders for humanoid core components including harmonic reducers, force sensors, and lightweight skeletons. Its humanoid project is designed for annual capacity of 3,000 units. It is one of the clearer examples of an established manufacturer moving from generic robotics exposure into direct humanoid component demand.
Perception is further along still. RoboSense said it sold 303,000 robotics LiDAR units in 2025 and named partners including AgiBot, Unitree, UniOne, Galbot, DOBOT, NEURA, Pudu Robotics, and China Mobile.
Source: Korthos relation graph
This is still a small sample of a much larger and more competitive Chinese supplier base. More of these companies are listed on Korthos. What stands out is the direction of travel. More of the ecosystem is now behaving as if the harder problem is no longer whether humanoids can attract attention, but whether the supply chain can manufacture, support, and absorb them at scale.
Other recent signals
Minth launched a joint-venture factory with Leaderdrive for humanoid joint modules.
ENCOS said joint-module shipments topped 100,000 units in 2025.
Fanzhou Precision said it secured first orders for harmonic reducers worth 5 million to 10 million yuan.
EYOU expanded its humanoid joint push with new product lines and a fully automated robot-joint industrial base.