Article
← All posts
robotics

The Robot Is 44 Motors in a Trench Coat

July 11, 2026 — Darwin Frederic

Reading time: 8 minutes

TL;DR

  • A humanoid robot needs roughly 28 to 44 actuators, one for every joint. At today’s low volumes, the actuators alone can cost $16,000 to $40,000 per robot, with mass production expected to cut that by 50-70%
  • The mechanical systems, the “brawn,” account for roughly 50% of a humanoid’s total production cost. The AI is not the expensive part
  • Fewer than 10 suppliers worldwide currently make high-precision, high-torque actuators at the required spec, and the harmonic gears inside them demand sub-micron tolerances
  • Every actuator spins on rare earth magnets, and China produces 94% of the world’s sintered permanent magnets. Its harshest export controls are suspended until November 10, 2026, a date every robot program is planning around
  • Humanoid shipments are projected to jump from roughly 2,000 units in 2024 to more than 50,000 in 2026, a 700% surge that TrendForce calls a watershed year
  • The supply chain for millions of robots per year does not exist yet. The encouraging part: it is visibly being built, from in-house actuator programs to new magnet capacity, and the suppliers who solve it will own one of the decade’s great industrial prizes

1. The Demo Is the Brain. The Bill Is the Body.

Watch any humanoid robot demo from the past year and the story is always about intelligence. The robot understands language. The robot learns tasks from video. The robot improvises.

Now look at the bill of materials, and the story inverts. Per Barclays Research, the mechanical systems that give a humanoid its “brawn” account for roughly 50% of total production cost. Not the AI chips, not the sensors, not the software. The motors, the gears, and the magnets inside them.

That means the most important companies in the humanoid race are mostly ones you have never heard of. There are far fewer of them than the keynote presentations suggest, and the race to multiply them is where much of the industry’s real progress is happening.

2. What a Joint Actually Is

A humanoid needs between 28 and 44 actuators depending on the design, one for each degree of freedom: hips, knees, ankles, shoulders, elbows, wrists, fingers, neck, and waist. An actuator is not just a motor, it’s a stack of precision components that has to fit inside a joint the size of your fist.

Inside One Humanoid Joint

  A SINGLE ROTARY ACTUATOR, EXPLODED
  ────────────────────────────────────────────

   [ Controller ]   drives the motor, talks to the brain
        │
   [ Encoder ]      measures position, thousands of
        │           times per second
   [ Motor ]        rare earth magnets spin here
        │           (NdFeB, ~12+ kW/kg torque density)
   [ Reducer ]      harmonic gear, 30:1 to 320:1,
        │           zero backlash, sub-micron tolerance
   [ Bearing ]      carries the load
        │
   [ Output ]       the actual joint movement

  × 28 to 44 of these stacks per robot
  ────────────────────────────────────────────

  WHERE THE ACTUATORS LIVE
        (o)  neck: 2-3
       /═══\
      ║ ▓▓▓ ║   waist: 2-3
   ═══╣ ▓▓▓ ╠═══
   arms: 7 each   hands: 6-12 each
      ║     ║
   legs: 6 each (hip 3, knee 1, ankle 2)
      ▐▌   ▐▌

The component that makes or breaks the whole stack is the reducer, usually a harmonic drive: a strain wave gear that converts a fast, weak motor spin into slow, powerful, precise joint movement, with effectively zero backlash. Each humanoid uses 20 to 40 of them. They are machined to tolerances under 1 micron, roughly one hundredth the width of a human hair, and even experienced manufacturers see defect rates above 5% in high-volume runs. Their unit costs run 2 to 3 times conventional gearing.

This is why the actuator stack alone costs $16,000 to $40,000 per robot at current volumes. The good news is baked into the same analyses: mass production is expected to cut that by 50-70%, and as the next section shows, the industry is attacking the problem from several directions at once.

3. Fewer Than Ten Companies, and Growing

Here is the number that should anchor every humanoid forecast you read: fewer than 10 suppliers worldwide currently produce high-precision, high-torque actuators at the spec these robots demand.

At the high end, two Japanese firms, Harmonic Drive Systems and Nabtesco, have anchored precision gear supply for decades. It’s not a coincidence that “harmonic drive” is both a product category and one company’s name. China has been closing the gap fast: Leaderdrive, the country’s largest harmonic reducer maker, holds an estimated 30-40% of China’s market per J.P. Morgan, and its 2025 net profit more than doubled while revenue rose 47%, a direct readout of how much real production activity is building behind the demos, and proof that supply responds when demand shows up.

The strain is real. Industry analyses of the reducer market report raw material shortages in high-strength alloys and flex splines pushing lead times up 20-30%. But so is the response, and it is exactly what a healthy industrializing sector looks like. Robot makers are pulling components in-house the way carmakers once did: Tesla has stated publicly, in earnings calls and investor presentations, that it designs Optimus actuators internally, and Figure built its BotQ factory partly to control this exact dependency. Vertical integration is what companies do when they want direct control over the parts that gate their scaling, and in this industry it doubles as new global actuator capacity.

Geography shapes the race too. Per SVRC Research, China holds roughly 26% of global actuator supply, and a Western prototype iteration that takes 12 weeks can be done in Shenzhen in 10 to 14 days. That density is a formidable head start, and it’s also a template: it shows the component problem is solvable at scale, because someone is already solving it.

4. The Magnet Under Everything

Go one layer deeper than the gears and you reach the most concentrated link in the chain: the magnets.

Every high-performance actuator motor depends on neodymium iron boron (NdFeB) permanent magnets. For dynamic humanoid joints, which demand torque density above 12 kW per kilogram in a package the size of a fist, there is currently no rare-earth-free substitute. Ferrite and other alternatives work for fans and pumps, not for a robot catching itself mid-stumble.

The supply picture, per the IEA: China produces 94% of the world’s sintered permanent magnets, up from about 50% two decades ago, and performs roughly 91% of global rare earth refining and separation. The EU imports 98% of its rare earth magnets from China. The largest US alternative, MP Materials, has magnet capacity of about 1,000 tonnes per year, roughly 1% of global NdFeB output. Small against the total, but it exists, it is growing, and it’s the template Western capacity will be built on.

This layer also carries a policy timer. In April 2025, China placed seven medium and heavy rare earths, plus NdFeB magnets containing terbium or dysprosium, under export licensing, with license reviews commonly running 60 to 120 days. A far stricter second wave announced in October 2025, extending Chinese licensing to products made anywhere containing as little as 0.1% Chinese-origin rare earth inputs, was suspended as part of a US-China trade truce, but only until November 10, 2026. Every robot program on Earth is planning around that date, and it’s the strongest single argument for the magnet capacity buildout now underway outside China.

Then there is the long-run volume math, which is best read as a measure of the opportunity. Adamas Intelligence calculated that the most expansive robot futures discussed in the industry, on the order of 10 billion humanoids by 2040, would require a 93-fold increase in NdFeB magnet capacity for robots alone. That capacity does not exist today. Building even a fraction of it is one of the largest identifiable industrial projects of the next two decades, and the investment case behind every magnet plant now being announced.

5. The Scale-Up Gap, Honestly Measured

Put the pieces together and measure the distance between ambition and supply chain.

Humanoid shipments are growing explosively: roughly 2,000 units in 2024, somewhere between 13,000 and 15,000 in 2025 depending on whose count you use, and forecasts of 50,000 to 60,000 in 2026, which TrendForce calls a 700% jump and a watershed year. Omdia projects 2.6 million units annually by 2035. Morgan Stanley sketches a billion robots by 2050.

Now run the component math. Sixty thousand robots in 2026 means roughly 1.8 to 2.6 million actuators and a similar count of precision reducers, from an industry where fewer than 10 firms currently make the good ones. Million-unit annual production, a scale that multiple companies across the US and China have publicly set as their ambition, means 30 to 44 million actuators per year, an order of magnitude beyond today’s entire global precision reducer output for all robotics combined.

That gap is real, and it should be read for what it is: not a verdict, but a construction schedule. Closing it is the industry’s shared engineering challenge, and the closing is already visible in public data: reducer makers doubling profits, robot companies building actuator lines in-house, component suppliers pre-building capacity, and magnet plants breaking ground on three continents. The intelligence improves on software timelines. The joints improve on machining, magnet processing, and export license timelines. The width of that gap, more than any demo, is what will set the industry’s real pace, and every quarter it’s visibly narrowing.

The Bottom Line

The humanoid robot is best understood not as an AI product but as 28 to 44 precision actuators wearing a trench coat, each one a stack of sub-micron gearing and rare earth magnets from one of the most concentrated supply chains in modern manufacturing. Half the cost is brawn. Fewer than ten companies make the critical parts today. One country makes 94% of the magnets inside them, with its strictest export controls paused until November 2026.

None of that makes the humanoid future doubtful. It makes it earned. The brain will scale like software; the body has to scale like a machine shop, and machine shops scale the old way: capacity built, suppliers qualified, tolerances held, one production line at a time. The most credible signal in this entire story is that the building has begun, and the companies doing the unglamorous work at the bottom of the stack are the ones who will make the demos deliverable.

Sources

  1. Barclays Investment Bank Research, “10 things you need to know about humanoid robots,” 2026 (mechanical systems ~50% of production cost; shipment trajectory 2,000 to 15,000 to 60,000 units)
  2. RoboZaps, “Humanoid Production Economics,” 2026 (28 to 44 actuators per robot; $16,000 to $40,000 actuator cost at low volume; 50-70% cost reduction at volume; fewer than 10 qualified suppliers)
  3. IDTechEx, “Humanoid Robots 2026-2036: Technologies, Markets, and Opportunities,” 2026 (precision component bottlenecks: screws, bearings, actuators; battery and thermal constraints)
  4. PW Consulting / harmonic reducer market research, 2026 (20 to 40 harmonic reducers per humanoid; supplier landscape including Harmonic Drive Systems, Nabtesco, Cone Drive, Laifual)
  5. Robot harmonic drive market analyses, 2026 (sub-micron tolerances; 5%+ defect rates in high-volume runs; 2 to 3x unit costs; lead times up 20-30% on alloy and flex spline shortages)
  6. Humanoid Guide and J.P. Morgan estimates, April 2026 (Leaderdrive 30-40% China market share; FY2025 profit and revenue growth)
  7. SVRC Research, “China Robotics Market 2026” (China ~26% of global actuator supply; Shenzhen 10 to 14 day iteration cycles vs 12 week Western cycles)
  8. IEA, “With new export controls on critical minerals, supply concentration risks become reality,” 2025 (China 94% of sintered permanent magnet production, ~91% of refining and separation; 2024 magnet export volumes)
  9. China Briefing (Dezan Shira) and European Parliament Research Service, November 2025 to January 2026 (April 2025 controls in force; October 2025 extraterritorial 0.1% rule suspended until November 10, 2026)
  10. Arnold Magnetic Technologies, February 2026 (60 to 120+ day export license reviews; ongoing disruption planning through 2026)
  11. Adamas Intelligence, “Humanoid robots and the future of motors and NdFeB markets” (93-fold magnet capacity increase implied by 10 billion robots by 2040)
  12. PatSnap NdFeB supply chain analysis, April 2026 (MP Materials ~1,000 tonnes per year vs ~100,000 tonnes global NdFeB output); 2026 magnet industry reporting (12+ kW/kg NdFeB torque density; no rare-earth-free substitute for dynamic joints)
  13. TrendForce, December 2025 (2026 shipments above 50,000 units, 700% YoY growth); Omdia via industry reporting (13,318 units in 2025; 2.6 million by 2035)