Deep Dive: China's Cost Advantage in Humanoid Robots
Alina Collins
Humanoid robot prices dropped from nearly ¥1 million to under ¥10,000 in a single year — driven by supply-chain reuse, 90% domestic-sourcing rates, and deliberate engineering trade-downs, not any single breakthrough. The decline is far from over.
Where does the cost actually go inside a humanoid robot?
Joints are the single most expensive component. On Tesla's Optimus, reducers alone eat 30%–40% of total cost; frameless torque motors take 16%, planetary roller screws 19%. Core joint parts account for sixty to seventy percent of the bill.
Unitree's G1, once torn down, shows a ¥41,600 materials cost — of which ¥27,500 sits in the core joints. This means → the battle for lower prices is not fought in the shell or the software; it is fought in the joints.
In plain terms = the cost structure of a humanoid robot is fundamentally a "joint problem." Drive joint prices down, and the whole machine follows.
How did component prices fall by an order of magnitude?
Harmonic reducers — the precision gear sets that let joints rotate accurately — dropped from over ¥10,000 apiece to the low thousands. Coreless cup motors — tiny motors inside each joint — fell from several thousand yuan to under ¥1,000.
Same design, different supply chain: Tesla's Optimus costs $131,000 in materials on a U.S. supply chain. Switch to Chinese suppliers, and that drops to $46,000 — roughly a three-times gap.
This reflects a first-order insight: the main cost lever is not design innovation — it is swapping one supply chain for another. Same blueprint, different vendors, two-thirds of the price gone.
Why is supply-chain density the real moat?
China's manufacturing cost edge does not come from cheap labor. Motor, reducer, and sensor production lines are highly automated; labor's share of cost is already small.
The real moat is geographic density: key suppliers of reducers, motors, sensors, battery cells, and AI chips all sit within a three-hour high-speed-rail radius. The Yangtze Delta alone hosts over 130 humanoid-robot-related firms; Shenzhen has 93 in one city.
This means → a Chinese maker can complete a component test-and-iterate cycle within a single day. An overseas maker customizing one part across borders often needs six months. The speed gap has moved from "somewhat faster" to a structural divide.
How far has component standardization gone?
Third-party joint-module shipments hit 280,000 units in 2025. Component-level scale arrived ahead of whole-machine scale — costs were already pushed down one round before entering the final assembly.
Harmonic reducer specs are converging to a few dominant models; planetary reducer interfaces are unifying. In plain terms = final assembly is moving toward a "build-your-own-PC" model — pick the spec, plug in the interface, power on.
This signals a deeper inflection: once hardware standardizes, the axis of brand competition shifts from "whose joints are better" to software and AI algorithms.
What timeline does the learning curve point to?
Manufacturing rule of thumb: every doubling of cumulative output cuts cost by 15%–20%. Global humanoid shipments in 2025 sit at roughly 18,000 units; 2026 forecasts range from 100,000 to 200,000.
After three doublings, costs can compress 40%–50% from today's base. A Unitree G1 currently selling at ¥85,000 could fall to under ¥40,000. This means → the leap from 18,000 to 144,000 cumulative units is expected in 2027–2028.
At that point, a robot that runs eight stable hours and carries over 10 kg will cost less in materials than one year's wages for a Chinese factory worker. In plain terms = the decision to buy a robot will flip from "strategic positioning" to pure arithmetic.
Can humanoid robots become China's next "super export category"?
Global shipments are projected to approach 5 million units by 2030. China already accounts for 90% of world output in 2025. As standardization and scale accelerate, that share is unlikely to be diluted soon.
Whether humanoid robots follow EVs and solar panels as China's third supply-chain super-export hinges on one variable: whether software and AI capabilities can keep pace with hardware cost declines.
This reflects the through-line of the entire story — hardware costs are already racing ahead; whether software can match them determines where the ceiling of this industry sits.
Content is for reference only, not financial advice.