Humanoid production and supply chain — 10–18 September 2026
Company developments across robotics, humanoid production, deployments, supply chains and robot learning.
Major funding rounds
ENCOS completed a Series B of more than RMB300 million to support product development, manufacturing capacity and market expansion. Investors included CITIC Goldstone, Nice Group, Suchuang, Huarui Investment and Nanjing Transportation Group, alongside returning investors including Fosun RZ Capital and Shenzhen Capital Group.
The company develops integrated robot joints combining motors, reducers and drive electronics. Its second-generation modules introduce a modular quick-release design intended to make hardware changes and maintenance easier for robot developers. ENCOS is extending this component business into connected hardware systems: joints, batteries and communications for movement, and dexterous hands with data-collection gloves for manipulation. The financing supports that wider hardware offering as well as production of its core joint modules.
Ture Intelligent
Hangzhou Ture Intelligent completed its first financing in the tens of millions of renminbi, backed by Dongli Future Technology, Tongli Technology and industrial investors. The proceeds will support an embodied-intelligence pilot base, production lines and recruitment.
Ture provides robot design and manufacturing services, including integrated joints, motion controllers, energy-management systems and robot bodies. It also develops data-collection and training systems that record real operating conditions for robot-model development. The planned pilot base connects those activities: customers can test hardware and algorithms against physical tasks before moving to batch manufacturing for factory, warehouse and inspection applications.
Strategic investments
AIDIN Robotics completed KRW16 billion in strategic financing from HD Hyundai Robotics and Samsung Venture Investment. It plans to use the capital for production equipment, quality validation and research staff, including a force-and-torque sensor manufacturing system with planned annual capacity of 30,000 units.
AIDIN and HD Hyundai will jointly develop a heavy-duty five-finger hand for HD Hyundai’s humanoid programme and surface-treatment automation for shipbuilding. AIDIN’s sensors and control software measure contact forces so robots can adjust pressure while processing uneven hull surfaces and welded sections. The partners plan to validate the technology in Korean shipyards before expanding overseas, including the United States; the sensor capacity and joint products are planned outcomes of the investment.
Deployments
AgiBot released footage of its A3 Ultra working across car dealerships, hotels and convenience retail. The scenarios include greeting customers and explaining vehicles, making beds and folding towels, and restocking or moving goods. The company says the platform has completed thousand-unit-scale automotive-grade production validation; that phrase describes production validation and does not verify a fleet of more than 1,000 deployed robots.
A3 Ultra combines 360-degree perception, UWB and RTK positioning, 700 TOPS of onboard compute and 20-degree-of-freedom tactile hands. Autonomous charging and battery swapping are intended to support extended operation across multiple commercial sites. The demonstration shows how the service platform is being applied to different workflows, including coordination between robots and people, rather than establishing a measured reliability or autonomy rate for those tasks.
Product launches
ugo unveiled Nova, a mobile dual-arm robot designed and manufactured in Japan, with production planned for 2027. Its two seven-axis arms and articulated upper body are intended to reach workspaces from near floor level to overhead shelving, while a mecanum-wheel base allows sideways movement in narrow factory and warehouse aisles. The company specifies up to 4 kg per arm, 10 kg in combined-arm handling and six hours of operation.
Nova also provides a platform for collecting demonstrations and testing robot models. Head and hand cameras record visual input, and VR or bilateral controllers let an operator demonstrate tasks while the system synchronises joint, torque and tool-position data. NVIDIA Jetson Thor runs learned models onboard. ugo pairs the robot with Physical AI Fabric services for data preparation, model development, evaluation and implementation, connecting research demonstrations with the work needed to introduce learned behaviours into operating sites.
Agility Robotics unveiled Digit 5 with a new safety architecture intended for work near people in manufacturing, warehousing and logistics. Human detection, visible and audible movement cues, and an independent safety controller are designed to let the robot avoid people, stop or sit when required. The system integrates NVIDIA IGX Thor and the Halos safety framework.
Preliminary specifications include a 23 kg payload, 2.2 m reach, 90 minutes of operation followed by nine minutes of charging, and a target of more than 20 operating hours in a 24-hour period. Swappable grippers and Agility’s Arc software connect the robot with mobile robots and warehouse or manufacturing systems. The design draws on more than 65,000 hours accumulated across earlier Digit operations; that history is not Digit 5 deployment time.
Early access is planned for the first half of 2027 and general availability by year-end, with European and UK expansion planned. RoboFab’s designed capacity is up to 10,000 robots annually. Agility also cited more than $300 million in multi-year orders; capacity and order value are separate from robots delivered and revenue already recognised.
Production and supply chain
Samhyun is developing a second-generation robot actuator targeting torque density above 100 N·m/kg, with launch planned for July 2027 and production in the second half of that year. Its first-generation actuators are scheduled to enter production as early as the end of 2026. The second-generation specification and timetable remain development targets.
The actuator combines the components that generate and transmit movement at a robot joint. Increasing torque while reducing module weight can reduce the loads carried by neighbouring joints: a heavier wrist also requires more work from the elbow and shoulder. Samhyun is aiming to retain a mean time between failures of roughly 20,000 hours while improving torque density, using precision assembly and both standard and customer-specific configurations.
Hechuan Technology said its third-generation frameless torque motors were ready for scaled manufacturing and a rapid ramp using the parent company’s established automated lines. Orders remained too small to use the available capacity fully. The statement was made in its September 18 half-year investor meeting.
Its dexterous-hand business was at an earlier manufacturing stage, with hands still assembled manually in small batches. Management said automation could support expansion if orders increased. The update therefore describes two different production conditions within the same robotics business: motors backed by an existing industrial manufacturing system, and hands requiring further assembly scaling. Neither the ability to ramp nor potential automation represents a disclosed volume of customer deliveries.
D-Robotics reported more than 8 million cumulative shipments across its Sunrise chip family in the first half of 2026. Its Sunrise S600 embodied-AI chip had been adopted by more than 20 customers within six months of its November 2025 launch, with the company describing a majority of those programmes as already in mass production. The 8 million figure covers the wider Sunrise family, rather than S600 alone.
The company supplies robot compute, developer kits and software infrastructure for humanoids, industrial mobile systems and consumer robots. S600 users include UBTECH, Astribot, Fourier and Booster Robotics, alongside developers of robot models and perception systems. The operating update accompanied a $400 million Series C, which will expand chip options and software spanning data collection, training, simulation and deployment.
Hyundai Mobis presented two robot actuator types under development at its R&D Tech Day in Yongin, South Korea. The programme ran through September 18 and covered more than 60 technologies, with robotics components displayed alongside automotive systems. The actuator presentation was a development showcase rather than the start of commercial production.
Actuators turn control commands into physical joint movement, making them a component through which the automotive supplier can extend its engineering into robotics. Hyundai Mobis organised the event around that expansion and invited North American and Asian customers to discuss orders and technical cooperation. The announcement does not establish production volumes or customer deliveries for the two robot actuator designs.
Baolong Technology won a design nomination for a robot rotary-joint encoder. The component measures motor-side and output-side position simultaneously using a dual absolute inductive encoder and non-contact electromagnetic coupling. The September report did not identify the customer or establish production volumes.
The design addresses position measurement inside compact robot joints exposed to magnetic interference. Baolong specifies resolution up to 19 bits, operation from −40°C to 125°C and IP67 protection, with joint repeat-positioning precision of ±10 arcseconds and response below one millisecond. Drawing on its automotive-sensor manufacturing base, the company is adapting that sensing technology to robot movement control. The nomination advances a component programme; it does not establish finished robots shipped or a completed commercial deployment.
Leadshine
Leadshine said its dexterous hands and miniature joints had completed sample validation at several humanoid developers and entered small-batch delivery. Its wider robot-component business also supplies frameless torque motors, planetary and harmonic joint modules, integrated wheel modules and coreless motors. The update came in a September 10 investor meeting.
The company combines component sales with joint development for larger customers and contract assembly. Its lower-degree-of-freedom hands use linkages for lower-cost commercial and consumer applications, while higher-degree-of-freedom direct-drive hands remain under development for research data collection and industrial work. The company had studied tendon-driven designs but had not committed to launching them, citing unresolved reliability and lifetime questions. These product stages distinguish the hands already being delivered from future designs.
Fengli Intelligent
Fengli Intelligent confirmed small-batch deliveries of miniature reducers for dexterous robot hands at its half-year investor meeting. It said its harmonic, planetary and miniature reducer families had completed development into product series and were capable of batch delivery, with several robotics customers working with the company.
Reducers adapt motor output to the torque and movement required at a robot joint. Fengli’s range addresses both larger humanoid joints and the smaller mechanisms inside hands. The company said its harmonic-reducer capacity had expanded from an initial 35,000 sets annually to around 70,000–80,000, and that further capacity investment would be phased around customer validation and orders. Its August share placement had closed, but the planned factory expansion remains separate from current shipments and utilisation.
UBTECH commissioned its Liuzhou robot factory on September 12. The line is designed for a cadence of one robot every ten minutes and planned annual capacity above 10,000 units. These are manufacturing design figures; the commissioning announcement does not establish that the factory has produced or delivered that volume.
The site produces Walker S industrial humanoids and Cruzr embodied-intelligence robots, using robotic handling and digital coordination within its own manufacturing process. UBTECH and Siemens developed a digital manufacturing foundation to organise component movements, assembly and quality management. That infrastructure is intended to accommodate the large parts count, frequent transfers and changing designs involved in humanoid production.
Linghou Robotics opened a Beijing production base in Changping covering nearly 5,000 square metres. Phase one is planned for annual capacity of 6,000 embodied-intelligence robots, with total capacity targeted at 15,000 as the site expands. Those figures describe the planned manufacturing system, rather than robots already produced or delivered.
The base connects prototype fabrication, pilot production, precision inspection, final assembly and batch delivery. It will also provide prototyping and small-batch manufacturing for other robotics companies, helping them move designs through testing before larger production runs. Linghou positions Beijing as a production and service hub for northern China, with local supply-chain access intended to shorten logistics and delivery cycles for customers in the region.
Chinese reports said a Tesla team arrived in Ningbo on September 16 and began Optimus production-readiness supplier audits the following day. The reported checks concern whether component suppliers can meet quality and manufacturing requirements for a larger robot ramp. Tesla had not confirmed the visits or related orders when Yicai contacted it.
Supplier responses also left the reported audit programme unconfirmed. Tuopu said it had not received notice of a new audit, while noting ongoing cooperation on linear, rotary and dexterous-hand actuators. Sanhua said it was unaware of the new reviews, and Joyson declined to discuss customer matters. Tesla had separately disclosed installation of Optimus production lines at Fremont, with initial robots intended for training and development; the audit report does not establish a verified production quantity or delivery schedule.
Dongfeng
Dongfeng outlined a factory-development schedule for its Xiaodong humanoid at its technology innovation week. The robot is planned to enter factory training in late October, participate in trial production by the end of 2026 and move into small-batch production in 2027. Initial applications include sorting, material handling and quality inspection.
Dongfeng participates in the robot’s higher-level intelligence design, while Huazhong University of Science and Technology develops lower-level control. The company is pursuing a shared intelligence system for multiple robot forms, including possible wheeled platforms for specialist work. Its factories provide environments in which to train and evaluate those systems; the disclosed schedule is a development plan rather than evidence of completed production deployments.






