A Humanoid Robot Has Dozens of Joints. Who Makes the Parts Inside Them?
China produced about 95% of the world's roughly 20,000 humanoid robot shipments in 2025, according to BofA Global Research figures reported by Reuters.
That is an impressive number.
But building a humanoid is one thing. Building one cheaply enough to become a useful worker is another.
To understand why, look past the AI and open the robot.
A large part of the hardware bill is simply the cost of making its body move.
Open the Robot's Hardware Bill
A humanoid needs cameras, sensors, batteries and computers.
It also needs a lot of joints.
According to McKinsey, actuation can account for roughly 40% to 60% of a humanoid robot's hardware bill, depending on the design.
Actuation simply means the system that creates and controls movement.
Think about a robot picking up a cup.
Its AI can decide:
Pick up the cup.
But that decision does not move the arm.
The shoulder has to turn. The elbow has to bend. The wrist has to move into position. The fingers have to close around the cup.
Each movement needs mechanical hardware.
Now imagine a company wants to build one million humanoid robots a year.
This is only an illustrative example, not a production forecast for any specific company.
The company does not just need one million AI computers.
If each robot contains dozens of powered joints, it may need tens of millions of precision joint systems and the parts inside them.
That is where another industry begins.
One Joint Is Already a Supply Chain
A robot joint is not simply a motor.
Open one and you can find several important components.
The motor creates motion.
The gearbox takes the motor's fast rotation and turns it into slower, stronger and more controlled movement.
The encoder tells the robot the exact position of the joint.
The bearing helps the joint rotate smoothly while carrying weight and force.
Drivers, controllers and other mechanical parts connect everything together.
Among these parts, the gearbox can be especially important.
McKinsey estimates that gearboxes can account for roughly 30% to 50% of the cost inside an actuator itself.
That distinction matters.
Actuation may represent 40% to 60% of the robot's hardware bill. The gearbox figure refers only to the cost inside the actuator, not to the cost of the entire robot.
The engineering problem is easy to understand.
Making a motor spin fast is not enough.
A humanoid needs to move with strength and precision inside a small, light joint. It may need to lift an object, stop at the right angle and repeat the same movement thousands of times.
That is why precision gear systems such as strain-wave and cycloidal gears matter.
Some of the Companies Behind Humanoids Are Not New
The humanoid industry may be new.
Many of its suppliers are not.
Consider Japan's Nabtesco.
The company has spent years making precision reduction gears for industrial robots, machine tools and other equipment. Its RV precision reduction gears have passed 14 million units in cumulative production.
Now the same type of expertise is moving toward humanoids.
Nabtesco has introduced smaller products including RVmini and Monocrank. The company specifically lists humanoid shoulders and hip joints among the possible applications for Monocrank.
The path looks something like this:
Industrial robots → Precision gears → Smaller and lighter joint systems → Humanoids
A new robot does not require an entirely new industrial base.
In many cases, old manufacturing knowledge is being adapted to a new machine.
Why an Auto-Parts Company Wants to Build Robot Joints
Germany's Schaeffler offers another example.
The company is well known for bearings and components used in cars and industrial machinery.
It is now developing products for humanoids, including bearings, motors, sensors, precision gears and actuators.
In August 2026, Schaeffler announced a new manufacturing process for strain-wave gearboxes designed for humanoid robots.
Traditional production can require highly precise machining.
Schaeffler is using a forming process instead.
According to the company, the new process can reduce manufacturing costs by more than 25% and material use by more than 75%.
Some important production steps can fall from minutes to seconds.
Series production is planned to begin in 2027.
Those numbers point to one of the biggest challenges in humanoid robotics.
It is not enough to make a good joint.
If humanoids are ever produced in very large numbers, manufacturers will need to make good joints quickly, repeatedly and cheaply.
AI cannot solve that manufacturing problem by itself.
China's Advantage Did Not Start With Humanoids
This also helps explain China's position in humanoid manufacturing.
China already has a huge electric-vehicle and industrial manufacturing supply chain.
Many of the components and production skills needed for EVs are also useful in humanoids.
McKinsey estimates that China accounts for about 90% of global permanent-magnet production capacity relevant to this supply chain.
Its estimated share is about 40% for precision bearings, 40% for encoders and position sensors, and 35% for motors.
These are not glamorous parts of a humanoid.
But a robot cannot move without them.
China's humanoid advantage therefore cannot be explained only by AI models or robot startups.
Behind those companies sits an existing industrial system built around motors, magnets, bearings, batteries, power electronics and precision manufacturing.
Humanoids can build on infrastructure that already exists.
Shipping a Robot Is Not the Same as Creating a Worker
There is an important reason not to get carried away by the production numbers.
Reuters reported that China produced about 95% of the world's roughly 20,000 humanoid shipments in 2025.
That does not mean 20,000 economically useful workers entered factories.
Reuters visited humanoid facilities in China and found robots that were still slow, struggled with unexpected situations and, in some tasks, remained far less productive than humans or traditional automation.
At one training center visited by Reuters, robots performed simple tasks at about 20% of the speed or productivity of a human worker.
Cost remains another problem.
Morgan Stanley estimates cited by Reuters put the low-volume hardware bill for Tesla's Optimus Gen 2 at roughly $50,000 to $60,000.
Government support is also helping create early demand in China.
Reuters reviewed around 1,000 government procurement documents and found that Chinese central and local government bodies spent at least $230 million during the first half of 2026 on humanoids, related equipment, training systems and demonstration projects.
That does not make the industry a bubble.
It does mean shipments should not automatically be treated as proof of commercial demand.
A factory eventually has to answer a simpler question:
Does using this robot save enough money to justify buying it?
The Other Race Inside Humanoid Robotics
Most discussions about humanoids focus on intelligence.
Which robot learns faster? Which AI model is better? Which machine can understand the real world?
Those questions matter.
But moving from demonstrations to mass production creates another race.
Manufacturers need millions of motors, gearboxes, bearings and encoders. Those parts need to become smaller and cheaper. Factories need to make them faster. And they need to work reliably after thousands or millions of movements.
Building a smarter robot is one problem.
Building millions of precise mechanical parts cheaply enough is another.
The next bottleneck may not be decided by software alone.
BEYOND THE OBVIOUS.
Sources
Reuters — China's Humanoid Robots Aren't Smart Enough to Take Your Job Yet
China's share of 2025 humanoid shipments, robot productivity, government procurement and cost estimates.
https://www.reuters.com/investigations/chinas-humanoid-robots-arent-smart-enough-take-your-job-yet-2026-08-27/
McKinsey & Company — Turning Humanoid Supply-Chain Constraints Into Billion-Dollar Wins
Humanoid hardware costs, actuator share of BOM and the cost structure inside actuators.
https://www.mckinsey.com/industries/industrials/our-insights/turning-humanoid-supply-chain-constraints-into-billion-dollar-wins
McKinsey & Company — China's Humanoid Robot Edge
China's estimated production capacity in permanent magnets, precision bearings, encoders, position sensors and motors.
https://www.mckinsey.com/featured-insights/charts/chinas-humanoid-robot-edge
Nabtesco — Precision Reduction Gears for Robotics
Precision reduction gears, cumulative production and applications for newer compact gear systems.
https://www.nabtesco.com/en/products/robot/
Schaeffler — Humanoid Robotics
Schaeffler's bearings, motors, sensors, gears and actuator technologies for humanoid robots.
https://www.schaeffler.com/en/technology-innovation/humanoid-robotics/
Schaeffler — New Manufacturing Technology for Humanoid Gearboxes
Schaeffler's new strain-wave gearbox manufacturing process, including cost, material and production-time reductions and its 2027 series-production plan.
https://www.schaeffler.com/en/media/press-releases/press-releases-detail.jsp?id=88210241