Introduction: Although the robotics industry is still in its early stages, players are rushing onto the path of vertical integration.
Unitree and its peers are trying to bring key links of the robotics value chain under their control.
Recently, Unitree released the new generation dexterous hand Dex5-S, featuring 22 degrees of freedom (DOF) per hand, 1:1 scale of a human palm, with a starting price of CNY 39,900.
But what is more noteworthy than the product itself is the way Unitree went about doing it.
All 22 motors in the Dex5-S are independently developed, supporting direct drive, backdrivability, and dual-encoder control. Each joint features a built-in extreme impact torque protection structure, and the control frequency reaches up to 1000Hz under a Gigabit Ethernet configuration. More importantly, from its inception, this dexterous hand has been embedded in the ecosystem of Unitree's independently developed UnifoLM-WLA-1.0 foundation model. The model is open-source, the hardware is independently developed, and a single model coordinates 64 tasks.
From the "brain" to the "hand," Unitree is trying to keep the entire chain in its own hands.
LinkerBot, a leading domestic supplier of dexterous hands, recently invested in the algorithm company Zhixing Embodied and simultaneously established a technology joint venture with Qianwei Yangchu. Hechuan Technology has expanded from joint modules to complete machines, building a full-chain product matrix of "motor-joint-complete machine".
Complete machine manufacturers are moving upstream, and component manufacturers are also moving upstream. This choice feels strangely familiar.
The Lesson of Batteries
Five years ago, China's NEV industry was at the starting point of explosive growth, but all automakers faced the same anxiety: batteries.
At that time, the domestic power battery market was highly concentrated, with top suppliers occupying more than half of the market share. Automakers could decide the appearance, smart cockpit, and marketing scripts of a car, but the core variables determining whether the car could be built, how high the cost would be, and how fast the delivery would be, were held in the hands of others.
An executive from an automaker once stated that power batteries once accounted for 40% to 60% of the cost of NEVs, meaning automakers were essentially working for battery manufacturers.
In 2022, the price of lithium carbonate surged to CNY 600,000 per ton, and the proportion of battery costs to the whole vehicle climbed from 30% to over 40%. Automakers' profits were squeezed from both ends by the upstream. William Li of NIO publicly stated that the price increase of lithium carbonate brought tens of thousands of CNY in cost pressure per vehicle.
The result was: starting in 2022, leading automakers rushed into self-developing batteries. GAC established Inpow Battery Technology, investing CNY 20 billion; NIO established a battery technology company; Li Auto and Sunwoda formed a joint venture to build a factory; Great Wall Motor incubated SVOLT. By 2026, the market share of leading battery suppliers had fallen by more than 10 percentage points from its peak.
Although automakers' self-developed batteries might not comprehensively surpass professional battery manufacturers in performance, they at least achieved one thing: avoiding supply chain chokeholds.
The price of this lesson was expensive. But today, the robotics industry is experiencing a highly similar structural moment.
The "Battery Moment" of the Robotics Industry
In the BOM cost of humanoid robots, joint modules account for over 50%. Among them, core components such as planetary roller screws, harmonic reducers, and frameless torque motors have long been partially dependent on imports. The price of dexterous hands has dropped from the CNY million level in the early years to the range of CNY 10,000 to CNY 50,000, but it remains one of the biggest challenges for cost reduction.
This is exactly the same situation automakers faced five years ago: the most expensive components were not made in-house, and there were only a few suppliers making these components.
The choice of leading complete machine manufacturers is highly consistent with the response path of automakers back then.
Unitree is the most typical case. Starting from quadruped robots, Unitree has taken a path of independent research and development of core components—motors, reducers, and controllers are all independently developed and manufactured. With the Dex5-S dexterous hand, this path has extended from the main body to the end effector, forming a closed loop with the independently developed embodied foundation model. The prospectus shows that the company plans to raise CNY 4.2 billion, of which 85% of the funds will be invested in technical research projects such as embodied models and robot main body R&D, with the remainder used for the construction of manufacturing bases.
Tesla has gone even further. All 28 body actuators of Optimus Gen3 are independently developed and customized. Musk's goal is very clear: use automotive-grade mass production experience to drive down component costs, reducing the cost per unit from USD 20,000 to USD 30,000 down to USD 2,000.
Leading domestic enterprises such as Zhiyuan Robot and Galbot are also extending upstream in different ways. According to industry statistics, leading complete machine manufacturers have achieved independent R&D or deep binding of over 80% of core components through investment, M&A, or direct independent R&D.
Component suppliers are also integrating backwards. LinkerBot has not stopped at the stage of "selling hands," but has invested in algorithm companies, attempting to bind hardware capabilities with operational intelligence. Hechuan Technology started with servo motors and encoders, extending all the way to complete machine solutions. Inspire Robots' micro servo electric cylinders and integrated joints are all independently developed, with an annual production capacity of over 50,000 units.
Complete machine manufacturers are eating upstream, and component manufacturers are also eating upstream. The entire industry is doing the same thing: bringing key links of the value chain under its control.
The underlying logic is exactly the same as automakers self-developing batteries back then: when the most expensive components are not in your own hands, you are forever working for others.
But "Independent R&D" Is Not a Panacea
However, there is a key difference that cannot be ignored.
Batteries are highly standardized industrial products. The chemical systems, packaging forms, and interface protocols of battery cells all have mature industry standards. For automakers self-developing batteries, the core challenge is engineering capability and economies of scale.
But the core components of robots are far from reaching this stage of standardization. In the field of dexterous hands, the DOF design ranges from 11 to 42, and drive solutions include multiple parallel routes such as motor direct drive, tendon-driven transmission, and pneumatic artificial muscles, while tactile sensing has not yet converged. The technical routes for joint modules are also running in parallel with multiple solutions, and have not yet been finalized into a few mature configurations.
This means that for robotics companies self-developing core components, they not only need to solve the problem of "whether to do it or not," but also answer the question of "which route to take." A wrong choice of route could be more costly than relying on external suppliers, as it not only spends R&D funds but also locks the company into a wrong technological path.
Another reality is that not all companies have the resource endowments of Unitree and Tesla. Tesla can amortize the R&D costs of Optimus into the AI and manufacturing systems of the entire group; Unitree has accumulated over CNY 240 million in R&D investment and plans to raise CNY 4.2 billion. But for a large number of small and medium-sized robotics companies, simultaneously self-developing dexterous hands, joint modules, and algorithm large models is a bet far beyond their financial capabilities.
A deeper issue is: vertical integration reduces reliance on suppliers, but it also means that all risks are concentrated on oneself. If the shipment volume of complete machines fails to meet expectations, the fixed costs of self-developed components will instead become a heavy burden.
This is exactly the pitfall that some automakers fell into when self-developing batteries back then: production capacity was built, but battery performance could not keep up; or batteries were produced, but whole vehicle sales did not keep up, resulting in an insufficient capacity utilization rate, and the unit cost was actually higher than external procurement.
In the first half of 2026, global shipments of humanoid robots reached approximately 19,100 units, with Chinese manufacturers accounting for 97%. JPMorgan Chase predicts that by 2030, global shipments are expected to exceed 1.75 million units. But at the current shipment scale, the "economies of scale" of self-developed components are far from arriving.
The choice of Unitree and its peers is correct; without mastering core components, there will never be pricing power.
Five years ago, automakers used hundreds of billions of CNY as tuition to give a phased answer: lose money first, and wait for scale. Today, it is the turn of robotics companies to make the same choice. The difference is that the story of batteries has at least turned to the page with an answer, while the story of robots has not even had its chapters defined yet.