The humanoid moment has arrived — but the race has quietly split in two: the West leads the brain, China is winning the body. Pictured: Ameca, a humanoid platform by Engineered Arts. Photo: Willy Jackson (CC BY-SA 4.0).

For three years the humanoid-robot story was a highlight reel: a Tesla bot folding a shirt, a Figure machine sorting packages, a Unitree humanoid doing a backflip. It was a demo war, and demos are easy to fake and easy to admire. That phase is ending. The contest now is manufacturing, and manufacturing has a way of sorting marketing from reality.

When you sort it, the race splits cleanly in two. On one axis, the brain, the AI models, the foundation software, the capital, the United States is ahead and probably comfortably so. On the other axis, the body, the actuators, the gearing, the magnets, and the sheer ability to ship units, China is winning, and not by a little.

It helps to separate two questions that usually get blurred. First: who can make a robot that is smart and capable? Second: who can actually build them, by the thousand, at a price someone will pay? The first is an AI and capital question, and the West leads it. The second is an industrial question, and by the only metric that is hard to fake, units that left a factory, China is far ahead.

Over the last 18 months, MavSource’s founder tracked and invested around major AI-related names like Micron +100%, Nvidia +74%, Sandisk +130%, Western Digital +74%, TSM +22%, Broadcom +27%, Okta +35% and Lam Research +39% — an average return of approximately +63% across the group.

Now, he’s bringing the sources and ideas behind those insights to you in one daily email digest. MavSource aggregates the most important AI updates from newsletters, podcasts, company news, AI labs, funding rounds, and more — then summarizes what matters in a simple 5-minute brief.

Subscribe For Free Now

Past performance does not indicate future results. Informational only; not investment advice.

Start with where the cost lives, because it explains everything downstream. In a humanoid robot, the expensive, difficult part is not the computer. It is the motion system. By most estimates the actuators and structure together account for 40 to 60% of the bill of materials, with actuators alone sometimes pegged above half. An actuator is a small marvel, a brushless motor, a precision reducer, a ball or roller screw, and a force sensor, and a humanoid needs dozens of them.

Estimated share of a humanoid's bill of materials. The motion system dominates; the AI compute is a relatively small slice. Source: BofA, Morgan Stanley estimates. Chart: Silicon & Steel.

This is the crux. The part of a robot that is hardest and costliest to build is precisely the part that rewards decades of manufacturing experience in motors, gears, and magnets, exactly the base China spent twenty years constructing. The intelligence, by contrast, is increasingly a downloadable commodity.

Now the scoreboard. In 2025, China is estimated to have shipped roughly 90% of the world's humanoid robots. Two firms tell the story. Unitree built more than 6,500 units and shipped over 5,500. AgiBot claimed more than 5,100. Each of those companies, individually, out-shipped every Western humanoid maker combined. By contrast, Tesla built what Elon Musk described as "hundreds" of Optimus units, against an original target of around ten thousand, and admitted in early 2026 that none were yet doing useful work.

Humanoid units shipped or built in 2025. The ~90% figure is an estimate, and the Unitree-versus-AgiBot ranking is contested, but the gap to Western output is not. Source: company figures, Omdia. Chart: Silicon & Steel.

A caveat worth stating: "shipped" covers a wide range of machines, from research platforms to early commercial units, and the number-one ranking between Unitree and AgiBot is disputed since both are self-reported. But you do not need to resolve that to see the shape of it. The unit gap between China and the West is an order of magnitude, not a rounding error.

Price makes the gap concrete. Unitree's G1 humanoid has a base price around $16,000, and it is for sale now. Tesla's stated mass-production target for Optimus is $20,000 to $30,000, a target, not a price, and one that assumes a volume it has not reached. Estimates of what an Optimus actually costs to build today run closer to $50,000 to $60,000. In other words, a Chinese competitor is already selling a humanoid for less than it currently costs Tesla to make one.

A shipping price versus an aspirational target versus today's build cost. Targets assume volumes not yet achieved. Source: company figures, BofA/Morgan Stanley estimates. Chart: Silicon & Steel.

None of this means the G1 and Optimus are the same machine, they are aimed at different uses, and capability matters. But on the dimension that decides whether robots become a mass product, cost at volume, the company that is already there is Chinese.

Dig into the body and you find the same chokepoints that run through the chip and rare-earth wars. Each humanoid needs roughly 1.3 kilograms of neodymium-class magnet material, and China manufactures an estimated 90% of the world's NdFeB magnets and processes about 90% of heavy rare earths. When China tightened rare-earth export licensing in 2025, it directly disrupted Optimus's timeline, a live demonstration that the leverage is real.

China's estimated share across key parts of the humanoid body. It dominates magnets, processing and assembly, but Japan still leads on the most precise motion components. Chart: Silicon & Steel.

But the picture is not total, and honesty requires the nuance. The most precise motion parts are still Japanese. Harmonic Drive Systems of Japan holds an estimated 85% of the high-end strain-wave-gear market, and Japan's THK and NSK together hold roughly 90% of premium ball and roller screws. China is climbing fast and leads the budget tier, but it does not yet own the top of the precision stack. The body is mostly China's, not entirely.

Tesla remains the name that moves headlines, so it deserves a clear-eyed read. Optimus is a genuinely ambitious program, with a Gen-3 design and a plan to convert a Fremont Model S/X line to robot production, low-volume late 2026, higher volume in 2027. But its volume forecasts have been the least reliable numbers in the entire sector. There is no public 2026 unit target. The right posture is to treat every Optimus production figure as aspiration until a factory proves otherwise.

Where the West unambiguously leads is capital and valuation. Figure AI raised more than $1 billion in September 2025 at a roughly $39 billion post-money valuation, about fifteen times its value eighteen months earlier, and unveiled its Figure 03 with a factory designed to scale from 12,000 toward 100,000 units a year. Those capacity numbers are targets, and the gap between a factory design and a factory humming is exactly where this sector's promises go to be tested. But the money and the AI talent are real, and they are concentrated in the US.

The contrast with Unitree is instructive. Unitree turned its first annual profit in its 2025 fiscal year, around $90 million on roughly $250 million of revenue, and filed for a Shanghai STAR Market IPO in March 2026, set to be the sector's first major public listing. One ecosystem is funded by venture capital betting on the future; the other is funded by selling robots today.

For the investor
The most defensible exposure may be the body, not the brand: the suppliers of actuators, reducers, precision screws and magnets, who sell to every robot maker regardless of nationality. Be skeptical of single-number market forecasts, the credible range runs from Goldman's ~$38 billion by 2035 to Citi's ~$7 trillion by 2050, which are not even measuring the same thing. Size positions to that uncertainty.

For the manufacturer
Supply chain is strategy here. A Western humanoid built without Chinese magnets and motors could cost materially more, directionally cited as up to several times, though the precise multiple is disputed. Securing actuator, reducer and magnet supply, or reshoring it, is not a sourcing detail; it is the difference between a product and a press release.

For the policymaker
Humanoids concentrate exactly the dependencies, rare earths, magnets, precision components, that recent industrial policy has tried to address for chips and EVs. The same magnet and rare-earth chokepoints that constrain defense and autos now constrain robots. Treating robotics supply chains as a separate problem misses that it is the same problem.

1. Goldman Sachs, the global humanoid market could reach $38B by 2035 [6 min]

2. Morgan Stanley, the humanoid market and the "Humanoid 100" value chain [7 min]

3. SCMP, Unitree shipped 5,500+ humanoids in 2025 [5 min]

4. McKinsey, turning humanoid supply-chain constraints into wins [12 min]

Most founders know their product. Few know how to get it in front of the right people. In this hands-on session, Clay + HubSpot for Startups walk you through ICP definition, prospect list enrichment, and AI-personalized outreach. You launch your first sequence before the session ends. June 18. 11am ET / 4pm GMT.

Register Before June 18

Actuator: the powered joint of a robot, a motor plus a gear reducer, a screw or linkage, and usually a force sensor. A humanoid carries dozens; together they are most of its cost.

Harmonic reducer (strain-wave gear): a compact, high-precision gear that turns a fast motor into slow, strong, accurate joint motion; a Japanese specialty at the high end.

Ball / planetary roller screw: the part that converts a motor's spin into precise linear push and pull; premium versions are dominated by Japan's THK and NSK.

Degrees of freedom (DoF): the number of independent ways a robot can move; a human hand has about 27, and matching it is one of the field's hardest problems.

NdFeB magnet: a neodymium-iron-boron magnet, the strong permanent magnet inside the motors that drive robot joints; manufacturing is concentrated in China.

BOM (bill of materials): the itemized cost of every physical part in a product; in humanoids it is dominated by the motion system, not the computer.

Edge inference: running an AI model on the robot itself rather than in the cloud, needed for fast, reliable physical reaction.

Sourced facts. The ~90% China shipment share (Omdia and industry estimates), Unitree's 6,500+ output and 5,500+ shipments, AgiBot's 5,100+ claim, Tesla's "hundreds" built against a ~10,000 target and the early-2026 admission, the Unitree G1 ~$16k price, the Optimus $20-30k target and ~$50-60k current BOM (BofA, Morgan Stanley), the ~40 to 60% actuator-and-structure share, the ~1.3 kg magnet per robot, China's ~90% NdFeB and heavy-rare-earth shares, Japan's ~85% harmonic-reducer and ~90% premium-screw shares, Figure's ~$39B raise, and Unitree's first profit and 2026 IPO filing come from company filings, sell-side research, and trade reporting.

What is estimate, and what is interpretation. Shipment and component shares are estimates and move quarter to quarter; the Unitree-versus-AgiBot ranking is contested and self-reported. Tesla's volume targets are the sector's least reliable figures and I treat them as aspirational. The three market forecasts are not comparable and I deliberately did not place them on one axis. The "~3x more without Chinese parts" idea is directional, not a precise number. Framing the race as a brain-versus-body split is my interpretation; the counter-case, that AI capability and capital ultimately dominate units, is real and worth holding alongside it.

Silicon & Steel Intelligence Desk, Supply Chain Strategy & Semiconductor Analysis. Nothing here is investment advice. Please check out more offerings on https://siliconandsteel.co/. Corrections & coffee: [email protected]