Deep Tech · Paradigm Analysis
Technology Analysis Blog
Paradigm
Shift Lab
Breaking barriers through materials & structural innovation
Skip to main content

The Hidden Bottleneck Inside Every Humanoid: Who Makes the Joints?

◆ Robotics  ·  Supply Chain

The Hidden Bottleneck Inside Every Humanoid:
Who Makes the Joints?

Everyone argues about the robot brain. But the thing actually standing between us and a million humanoids isn't AI — it's a fistful of precision gears and screws that almost nobody on Earth knows how to mass-produce.

Here's a question that sounds simple and isn't: when Tesla, Figure, and a dozen Chinese startups all promise millions of humanoid robots, where do the joints come from?

Not the AI. Not the cameras. The joints — the shoulders, hips, knees, and elbows that let a 60-kilogram machine balance on one leg and set down an egg without crushing it. We spend all our breath on the robot's mind, because minds are sexy and gears are not. But ask any engineer actually shipping these machines what keeps them up at night, and they won't say "the neural net." They'll say the actuators. Let me show you why the most boring part of a humanoid is quietly the most important.

A Humanoid Is a Walking Bag of Joints

Strip away the sci-fi and a humanoid robot is, mechanically, a stack of motors. Tesla's Optimus runs on 28 actuators in its body — the electromechanical "muscles" that move each joint. For comparison, your body has roughly 600 skeletal muscles driving 83–95 degrees of freedom. So a humanoid is already a radical compression of you, down to a couple dozen motors doing the work of hundreds of muscles. Each of those 28 has to be small, strong, precise, and durable all at once — a combination that turns out to be genuinely brutal to manufacture.

And here's the part that surprised me most when I dug in: a single robot joint isn't one part. It's a tightly nested assembly — a motor, a reduction gear, a position encoder, and a structural housing — all aligned to tolerances measured in ten-thousandths of an inch. Industry machinists describe holding concentricity to around 0.0005" on the bearing bore. That's the width of a few human cells. Get it slightly wrong and the joint whines, wears, or fails.

What's Actually Inside One Humanoid Joint
Frameless motor spins fast, low torque + Harmonic reducer trades speed for torque + Encoder measures the exact angle + Precision housing aligns it all to ±0.0005"
One rotary joint = four nested precision parts. Multiply by ~28 per robot, then by a million robots. Tolerance spec per RivCut / Industry Week, 2026.

The 56% Problem

Now for the number that reframes the entire humanoid industry. According to widely-cited Morgan Stanley estimates, the actuators inside Tesla's Optimus account for roughly 56% of the robot's total bill-of-materials cost. Not the battery. Not the AI computer. Not the fancy cameras. More than half the cost of the machine is in its joints.

Drill in and it gets sharper. Optimus uses around 14 planetary roller screws — the heavy-duty linear actuators in the load-bearing lower body — and Morgan Stanley pegs each one at $1,350 to $2,700. That single component type alone is estimated at about 19% of the whole robot's cost. Fourteen screws costing as much as a used car.

56%
Actuators' share of Optimus BOM cost
~19%
Roller screws alone (14 units/robot)
28
Body actuators (vs ~600 human muscles)

Sources: Morgan Stanley estimates (via Optimus hardware analyses) · IDTechEx Humanoid Robots 2026–2036

Where does all that money pool? In the legs. Per circulated Morgan Stanley estimates of the Optimus Gen 2 build, the most expensive regions are the load-bearing ones — hands, waist/pelvis, thigh, calf, and feet — precisely because that's where the big, high-torque actuators live.

Where the Cost Hides — Optimus Gen 2 by Body Region (est., $K)
Hands $9.5K Waist & Pelvis $7.8K Thigh $7.3K Calf $7.3K Feet $6.7K Elbow $2.6K Forearm $2.2K Head $2.1K
Selected major line items, estimated Optimus Gen 2 BOM (Morgan Stanley estimates). The lower body dominates — that's where the high-torque actuators concentrate.

This is why Tesla's entire $20,000 price target is, at its core, a bet on crushing actuator costs. If you can't make the joints cheap, you can't make the robot cheap. Full stop. The AI is basically free to copy once it works; the gears have to be physically built, one precise unit at a time.

So Who Actually Makes These Things?

Here's the answer that should make you sit up: a shockingly small club of companies, most of them Japanese or Swiss, sitting on decades of head start. When you buy a humanoid, you're really buying their parts. Let's meet the gatekeepers.

⚙️
Harmonic Drive Systems
Japan  ·  Strain-wave (harmonic) reducers

The undisputed king of the harmonic reducer — the compact, zero-backlash gearbox that lets a small fast motor produce big precise torque in a rotary joint. If a humanoid has a wrist or shoulder that moves smoothly, odds are a strain-wave gear is doing the work, and odds are it traces back to this company's patents.

The default choice for rotary joints — and a single point of dependence for the whole industry.

⚙️
Nabtesco
Japan  ·  RV / cycloidal reducers

Nabtesco dominates the RV (cycloidal) reducer market, the heavy-duty cousin of the harmonic drive used where shock loads are higher. Between Harmonic Drive Systems and Nabtesco, two Japanese firms have effectively owned precision robot gearing for the entire industrial-robot era — and humanoids inherited that dependency.

Cycloidal specialist — the muscle for higher-load joints.

🔩
SKF / Ewellix · Rollvis
Europe  ·  Planetary roller screws

The planetary roller screw is the heart of a linear actuator — it turns a motor's spin into a powerful, durable push, perfect for hips and knees that carry the robot's weight with every step. McKinsey flags this as the most acute bottleneck of all: a high-load, high-precision niche with a narrow supplier base, long lead times, and almost no substitutes. Switzerland's Rollvis and SKF's Ewellix unit are among the few names that can make them at robotics grade.

The scarcest, hardest-to-scale part in the whole machine.

Kollmorgen · Maxon
USA / Switzerland  ·  Frameless & coreless motors

America's Kollmorgen leads in frameless torque motors — the high-torque electric heart of large joints. Switzerland's Maxon dominates the tiny coreless motors used in dexterous hands, where a single hand can need around twenty of them. The motors aren't the rarest piece, but the compact, high-power-density versions humanoids demand still narrow the field fast.

The electromagnetic core — common in principle, specialized in practice.

There's a pattern here that's impossible to miss. The companies that decide whether the humanoid era happens on time are not the flashy robot startups in Silicon Valley. They're century-old precision-mechanics firms in Japan, Switzerland, and Germany — the kind of companies that win awards for gear tooth geometry. Honestly, that's my favorite thing about this whole story: the future of robotics is gated by the least futuristic businesses imaginable.

Why It's a Bottleneck, Not Just a Shopping List

Knowing the suppliers is one thing. The real problem is that none of them were built for the volumes humanoids will demand. These parts were made, for decades, in the thousands — for industrial robot arms bolted to factory floors. Humanoids want them in the tens of millions.

Consider the math. One million robots × 28 actuators each is 28 million precision actuators a year — and that's a single manufacturer's ambition. The current global supply chain for strain-wave gears and roller screws simply was not designed for that order of magnitude. The strain shows up as lead times: industry reporting in 2026 put harmonic-drive gearset waits at around 26 weeks. Six months for a gear. And per machinists in the field, the true chokepoint isn't even raw machining capacity — it's the painstaking, assembly-matched integration of motor, gear, encoder, and housing to those cellular tolerances.

The uncomfortable reality: every humanoid roadmap quietly assumes the actuator supply chain will scale 100× or more this decade. IDTechEx put it plainly — high-precision components like screws, bearings, and actuators remain a critical constraint, because the supply chains "are not yet optimized for mass-volume humanoid production." The robots can be designed. Whether they can be sourced is the open question.

The Rare-Earth Wrinkle

It gets one layer deeper, and this layer is geopolitical. Those high-torque motors need powerful permanent magnets — specifically neodymium-iron-boron (NdFeB) — to hit the power density a humanoid joint requires. And here's the kicker from McKinsey: China controls roughly 69% of global rare-earth mining and around 90% of magnet processing and refining. Recent export-licensing changes have already injected volatility into that supply.

So even the "easy" part — the motor — has a chokepoint, and it runs through Beijing. The interesting wrinkle is that this turns a manufacturing story into a national-security one. You can design the world's best robot in California and still be one export rule away from not being able to build it.

China's Counterattack

Which brings us to the most dynamic front in this whole war: China isn't just sitting on the magnets. It's coming hard for the high-value parts the Japanese and Swiss have monopolized — and competing the way China usually does, on price.

Per reporting from Chinese outlet 36Kr, domestic harmonic-reducer makers like Green Harmonic are selling at roughly 40–60% of the price of Japan's Harmonic Drive. Roller-screw maker Wuzhou Xinchun is reportedly pricing at about half of Switzerland's Rollvis. One firm, Xinjian Transmission, kicked off a project to build a million planetary roller screws a year. And motor supplier Buko said a single humanoid customer's order leapt from the hundreds-to-thousands range to the "tens of thousands" in a year.

Even Tesla — the most American of robot makers — leans on this. In October 2025, per 36Kr, Musk reportedly placed a roughly $685 million order with Sanhua for rotary actuator assemblies, with delivery from a Mexican factory starting in 2026. The quiet irony of the humanoid age: the all-American robot may run on substantially Chinese joints.

The New Entrants Smell Blood

When a market has a screaming bottleneck and a trillion-dollar prize on the other side of it, big industrial players don't stay on the sidelines. The most telling move of the past year: Germany's Schaeffler — an automotive Tier-1 giant — unveiled an "all-in-one" humanoid actuator at CES 2026, integrating motor, gearbox, and controller into a single unit for hips and knees. Schaeffler isn't a robot company. It's trying to become the Bosch of robot joints — the indispensable supplier that sells the muscle to everyone, no matter whose logo is on the robot.

That's the real strategic prize hiding in this article. Whoever cracks high-volume, low-cost, reliable actuator manufacturing doesn't just win a contract — they own a toll booth on the entire humanoid economy. My honest bet: ten years from now, the most valuable company in robotics may be one that never builds a single complete robot.

What About the Hands?

You may have noticed I've mostly skipped the hands — and that's deliberate, because hands are a bottleneck so brutal they deserve their own story. A dexterous hand can pack 20-plus tiny actuators into a space the size of, well, a hand, and getting them small, strong, and survivable is arguably the single hardest mechanical problem in the field. Tesla's latest design even relocates the hand motors up into the forearm and pulls the fingers with tendons, just to fit everything in.

The Joint Supply Chain at a Glance

ComponentWhat It DoesWho LeadsBottleneck Risk
Harmonic reducerCompact, zero-backlash torque for rotary jointsHarmonic Drive Systems (JP)High — concentrated supplier base
RV / cycloidal reducerHeavy-load rotary gearingNabtesco (JP)Moderate–High
Planetary roller screwHigh-force linear push for hips/kneesRollvis, SKF/Ewellix (EU)Severe — narrow supply, long lead
Frameless torque motorElectromagnetic core of large jointsKollmorgen (US)Moderate (robotics-grade)
Coreless motorTiny motors for dexterous handsMaxon (CH)Moderate
NdFeB magnetsPower density for all motorsChina (~90% refining)Geopolitical — export risk

Sources: McKinsey (Apr 2026) · 36Kr (2025–26) · Morgan Stanley Humanoid 100 · RivCut / Industry Week (2026)

So What Happens Now?

The honest read: 2026–2028 is when this gets decided. If the precision-parts industry scales fast enough — through Japanese incumbents expanding, Chinese challengers undercutting, and new Tier-1s like Schaeffler muscling in — then the million-robot promises have a fighting chance. If it doesn't, the robots will exist, but in the thousands, gated not by intelligence but by inventory.

So the next time you watch a slick humanoid demo and marvel at the AI, let your eye drift to the shoulders and hips instead. That smooth, quiet rotation is the hard part. That's the part a handful of companies on the other side of the world spent fifty years learning to make — and the part that will decide, more than any neural network, how many robots actually show up.

The Brain Gets the Headlines. The Joints Decide the Future.

It's tempting to think the humanoid race is a software race. It isn't — not yet. The intelligence is improving exponentially and copies for free. The joints improve slowly, cost a fortune, and have to be physically machined one at a time by a small club of specialists.

Watch the actuator suppliers, the lead times, and the rare-earth headlines. That's where the real bottleneck lives.

"The robot you'll meet in 2030 is being limited today — not by how smart it can be, but by how fast someone can build its knees."

Comments

Figure, Tesla Optimus, 1X NEO — What's Actually Changed in 2026

Figure, Tesla Optimus, 1X NEO — What's Actually Changed in 2026 | Paradigm Shift Lab ◆ Robotics  ·  2026 Update Figure, Tesla Optimus, 1X NEO — What's Actually Changed in 2026 Not demos. Not concept renders. Real factories, real production lines, real dollar signs. Here's the honest 2026 scorecard for the humanoid robot race — and why this year is genuinely different. Figure AI Tesla Optimus 1X NEO Boston Dynamics Atlas PSL Editorial  ·  May 2026  ·  ~7 min read Let's be honest for a second. For the past three years, "humanoid robot updates" meant the same thing: a carefully choreographed demo, a funding announcement, and a CEO posting a slow-motion video of their robot picking up a box. Then everyone moved on. 2026 feels different. Not because the robots suddenly got perfect — they didn't. But ...

Glass Interposer vs Glass Core Substrate: TSMC CoPoS vs Intel EMIB Explained

Glass Substrate Series | Advanced Packaging Glass Interposer vs Glass Core Substrate: TSMC CoPoS vs Intel EMIB Explained They're both called "glass substrates" — but they solve completely different problems. The race to define AI chip packaging's future is playing out between two very different bets. ⬡ Paradigm Shift Lab · Advanced Packaging Deep Dive · 2026 ⚡ Key Takeaway Glass interposers replace the silicon bridge between GPU and HBM dies. Glass core substrates replace the entire organic base layer of a chip package. TSMC is betting on the former. Intel is pushing the latter. Same material — completely different strategy. Context Why Glass Now — The Limits of Organic Substrates The AI accelerator era has pushed semiconductor packaging requireme...

Boston Dynamics' Atlas Is Training for the 2026 World Cup — And I'm Not Sure How to Feel

Boston Dynamics' Atlas Is Training for the 2026 World Cup — And I'm Not Sure How to Feel (Part 1) ◆ Robotics  ·  Part 1 of 2 Boston Dynamics' Atlas Is Training for the 2026 World Cup — And I'm Not Sure How to Feel I genuinely believed robots doing fine, expressive, human-like movement was decades away. Then Boston Dynamics dropped a video this week. I had to revise everything. Boston Dynamics FIFA World Cup 2026 School of Football PSL Editorial  ·  May 2026  ·  ~5 min read Hyundai × FIFA World Cup 2026 Official Partner. That figure in the background isn't a player. It's Atlas. There are things you quietly file away as "not in my lifetime." Not impossible — just comfortably distant. Fusion energy. Mars colonies. And robots doing things that require genuine physical expressiveness. Haircuts. Dance. Spo...