The humanoid industry likes to argue about brains. Whose VLA transfers better, whose world model learns faster, whose teleoperation fleet collects cleaner data. All of that debate sits one floor above a quieter dependency that nobody can train their way out of: the electric motor. A humanoid is, before it is anything else, a stack of 40 to 100 synchronized actuators, and the torque that lets a 78-kilogram machine swing its own mass through a gait cycle comes from permanent magnets. Specifically, from sintered neodymium-iron-boron, the highest-energy-density commercial magnetic material in existence, at roughly 400 kJ per cubic meter. And the supply chain that turns rare earth ore into finished humanoid-grade magnets runs, at 94 percent of global manufacturing capacity, through China. That figure is not a lobbyist’s talking point. It is the International Energy Agency’s count for 2024, and it is the reason a component most robotics executives never mention by name is now on the agenda of a superpower summit.
The forty-motor machine
Start with the physical reality, because it explains why this is a structural dependency rather than a procurement detail. Published teardowns and supplier engineering guides converge on the same arithmetic: an Optimus-class platform of the Gen 2 generation carried 16 rotary actuators, 14 linear actuators and two dexterous hands with six coreless motors each, roughly 42 magnetic motors in total. Newer hands push the count far higher, with more than 50 actuators per hand on the most dexterous platforms. Estimates of total NdFeB content per robot converge on a range of 2 to 4.5 kilograms, which is two to three times the magnet load of a typical electric vehicle per MP Materials’ own framing.
The magnet is not incidental to the actuator. It is the actuator’s reason for existing. Permanent-magnet synchronous motors deliver the torque density that legged locomotion demands, and engineering analyses put NdFeB-based designs at 8 to 12 Nm per kilogram against 2 to 3 Nm per kilogram for ferrite alternatives. Axial-flux topologies built around NdFeB reach 15 to 25 Nm per kilogram at over 90 percent efficiency, which is why they have become the default choice for weight-critical joints. A robot that swapped to ferrite to escape the rare earth problem would, at constant torque, gain meaningful mass in every joint it changed. On a machine where actuators already account for more than half of body weight and every kilogram of metal fights every kilogram of battery, that trade is usually unacceptable. IDTechEx’s materials work projects humanoid materials demand growing at a 47 percent compound annual rate through 2036, passing 112,000 tonnes. Most of the value in that stack flows through the magnet supply chain.
The heat tax: dysprosium and terbium
Raw neodymium is only half the story, and the less strategic half. Joint motors run hot, surrounded by windings, gears and their own waste heat in sealed joint cavities. Standard-grade NdFeB demagnetizes as temperature rises. To hold performance at operating temperature, magnet makers substitute heavy rare earths, dysprosium or terbium, into the crystal structure, trading raw strength for thermal stability.
The quantities involved are small and the dependency they create is total. United States Department of Energy supply chain tables show a kilogram of H-grade magnet requires 28 grams of dysprosium against 272 grams of neodymium-praseodymium; SH-grade raises that to 42 grams. Modern grain-boundary diffusion techniques can cut dysprosium content by half for a given grade. Run the numbers, per the Foundation for American Innovation’s analysis, and a single humanoid contains roughly 0.57 to 1.3 kilograms of NdPr and 35 to 79 grams of dysprosium. Neodymium and praseodymium are light rare earths, mined and refined at meaningful scale outside China, at Lynas in Australia and Malaysia and MP Materials in the United States. Dysprosium and terbium are not. They are produced in negligible quantities outside the Chinese sphere of influence, and they are the specific elements Beijing has chosen to regulate.
That is the first thing to understand about the magnet choke point: it is not the element you have heard of. The neodymium headline conceals the dysprosium dependency.
One company in Baotou
Concentrate the map further and the supply side resolves to a single company. JL Mag Rare Earth, headquartered in Baotou, Inner Mongolia, near the Bayan Obo mine, became the world’s largest high-performance magnet producer by output and sales in 2024, with an estimated 10 to 20 percent of the global market. Its capacity is more than ten times that of its closest American competitor, its three closest rivals are also Chinese, and between 2019 and 2025 its annual capacity more than quadrupled. The company targets 60,000 tons of capacity by next year and plans to add another 20,000 tons in Baotou by 2028.
Two data points make JL Mag the single most important company in the humanoid story that almost no robotics publication names. First, Daiwa Securities has identified the firm as the sole supplier of motor rotors for Tesla’s Optimus robots. Not a supplier: the supplier, for the rotors at the heart of every Optimus actuator. A second analyst estimates JL Mag provides roughly 80 percent of the magnets in Tesla’s vehicles. Second, the company launched a dedicated humanoid robotics business unit last year, reporting directly to its chief executive. The world’s magnet juggernaut has reorganized itself around your industry.
The competitive gap is not primarily geological. “The Western companies are still developing their technology, doing everything from scratch,” Daiwa analyst Derek Zhang told Bloomberg. “The Chinese companies can ramp up capacity very easily and the cost is maybe about 1/10th.” Constantine Karayannopoulos, former CEO of Neo Performance Materials, was blunter about the company he competes with: “They came out of nowhere. To me, there is JL and then there is everybody else. They are a juggernaut. They can take business at any price.” David Abraham of Materium Strata describes the operational gap: customers hand JL Mag their specifications and the company’s production lines simply produce, a maturity no Western magnet startup can currently match. And Beijing is deliberately moving the target: JL Mag and its peers are expanding from magnets into finished components like motor rotors, meaning that even a West that secured its own dysprosium could still depend on Chinese manufacturing technology, patents and scale to build the parts robots actually need.
The body count: April 2025 already ran the experiment
The dependency has already produced casualties, which is what distinguishes this risk from hypothetical industrial-policy anxiety. Tesla aimed for 5,000 Optimus units for internal use in 2025. Actual output was reportedly a few hundred units, and China’s April 2025 export controls on seven rare earth elements and the magnets made from them were cited as one of the causes. Those controls remain in place alongside a dual-use licensing regime, which is the quiet operational meaning of the whole story: every Western humanoid manufacturer shipping at scale is doing so with Beijing’s export licenses in the loop.
The regime is tightening on schedule. The New York Times reports that beyond the April 2025 controls, China has announced a broader set of restrictions scheduled to take effect in November. Chinese magnet exports to the United States fell last month. Rare earths sit at the top of the trade agenda as Xi Jinping meets Trump in Washington, with Beijing weighing additional export licenses as a bargaining chip, and Rare Earth Exchanges reports the two sides have extended their trade truce to January 10. For a robotics industry planning 2027 production volumes right now, the binding constraint is not magnet physics. It is a licensing decision in Beijing that can change with a summit communiqué.
The arithmetic says scarcity is not the problem
Here is the counterintuitive part, and the reason serious analysts should resist the panic narrative. The Foundation for American Innovation’s supply chain math, built on DOE and IEA data, shows the electric vehicle industry already consumes more heavy rare earth each year than the production of 26 million humanoids would require. In 2025 the world bought 21 million EVs carrying about 19.4 million permanent-magnet motors and roughly 30 million kilograms of NdFeB, absorbing approximately 31 percent of global dysprosium-equivalent supply, because EV drive motors use higher-temperature grades than robot joints. Bank of America projects 10 million humanoids produced annually by 2035, which would require about 12 percent of existing heavy rare earth supply. Goldman Sachs’ base case of 1.4 million units needs under 2 percent.
At any forecast volume short of Musk’s 10-billion-robot fantasy, the planet has enough dysprosium. A humanoid boom would actually tilt demand toward light rare earths, where Western refining already exists. The tonnage is not the choke point. The licensing is. As the FAI analysis puts it, a tightening of export controls on heavy rare earths could still be disastrous for Western EV and robotics manufacturing alike, concentrating both industries inside China regardless of how much material the rest of the world could technically dig up. The magnet problem is not a resource problem. It is a leverage problem wearing a resource costume.
The Western answer: ten thousand tons in Texas
The response is real, late, and an order of magnitude short. MP Materials, operator of the only US rare earth mine and processing site, began producing magnets at Fort Worth last year. After a $400 million Trump administration investment for roughly a 15 percent equity stake, the company unveiled plans for a $1.25 billion plant in Northlake, Texas that would lift its capacity to about 10,000 tons per year. That equals roughly a third of JL Mag’s 2025 output, before JL Mag’s own expansion lands. The IEA projects demand for magnet rare earths outside China rising 50 percent by 2035, with existing and planned non-Chinese projects meeting less than a fifth of it.
The timing mismatch matters more than the tonnage gap. Tesla’s Fremont pilot line is reportedly targeting initial Optimus production this quarter, with a 7 million square foot dedicated plant rising at Giga Texas behind it. Chinese sources describe an unconfirmed 50,000-unit 2026 Optimus plan surfacing around Tesla’s September 17 supplier audits in Ningbo, where teams examined Tuopu Group, Sanhua Intelligent Controls and Joyson Electronics, with some production functions potentially shifting toward Chinese suppliers rather than away from them. The strategic paradox deepens by the week: Washington bans new Chinese humanoid imports while the flagship Western humanoid program audits Chinese suppliers and depends on a Baotou company’s rotors to hit volume. JL Mag, for its part, was largely unfazed by the last round of controls and told investors in its annual report that Chinese magnet makers will remain the “dominant players.”
What the choke point does to the roadmap
For robot designers, the magnet constraint is now a first-order engineering input, not a sourcing afterthought. The available levers are real but each one spends performance. Grain-boundary diffusion thrifts dysprosium by half and is already assumed in most estimates. Terbium substitutes for dysprosium at 2.5 to 1 effectiveness, but commands over four times the price and is equally controlled. Ferrite-backed or mixed designs work for non-critical joints at a third of the torque density. Motor topology redesign, axial flux, high pole counts, Halbach arrays, squeezes more torque per gram of magnet, which is why every actator supplier’s roadmap now reads like a magnet-efficiency program.
For everyone else, the metric that matters was framed crisply by Rare Earth Exchanges: not robots announced, but robots shipped, multiplied by magnet content, multiplied by magnet chemistry, multiplied by country of origin. Until a Western humanoid maker discloses all four variables, its volume targets are undated promises sitting on top of a licensed supply chain.
The volume race between Chinese and Western humanoids is usually narrated as a contest of factories, capital and AI. It is also a contest being decided in Baotou, in presses that run at a tenth of Western cost, behind export licenses that expire when geopolitics do. China’s humanoid makers get their magnets domestically, at scale, without a license queue. The West’s makers get theirs through a choke point that the November restrictions could narrow further. Watch the November effective date, the January 10 truce deadline, and MP’s Northlake timeline. The first company to build a humanoid actuator supply chain that survives a full export cutoff, not a summit, will be the one that actually broke the dependency.
Sources: Bloomberg via Mining.com, B2BNN / IDTechEx, Foundation for American Innovation, Rare Earth Exchanges, New York Times, IEA Rare Earth Elements report.