Why rare earth magnets could decide the humanoid robot race

An electric motor spins because electromagnets push and pull against permanent magnets. Stronger magnets give more torque (turning force) from a smaller, lighter motor: FAI notes that motors using cheaper ferrite magnets have to be about 30% heavier, and usually bigger, to deliver the same torque. In a robot that matters a lot, because each motor sits inside a limb that other motors have to lift.

FAI says robot actuators (the motor and gear units that move each joint) depend on NdFeB for now. The magnets also have to stay strong when they get hot, so makers add "heavy" rare earths such as dysprosium and terbium. Commodity analysts at CRU say humanoids may need up to 70 degrees of freedom, meaning separate directions their joints can move. Magnet motors are the preferred choice for most of those joints, though CRU notes that low-power ones like fingers could in theory use other motor types.

How much magnet goes into one humanoid?

No robot maker publishes this figure, so every number here is an outside estimate. CRU and FAI both land on a few kilograms of NdFeB per robot, with published estimates running from 2 to 4.5 kilograms.

That sounds small until you multiply it. In CRU's base case, the world is building well over 100 million humanoids a year by 2040, and those robots alone would need about one and a half times as much magnet as the entire market makes today. It's a forecast built on adoption assumptions, not a tally of orders.

For a nearer example, Musk said in April 2025 that he was confident Tesla could make a million Optimus robots a year by 2030. At CRU's 2 to 3 kilograms each, that alone works out to 2,000 to 3,000 tonnes of magnets a year (our arithmetic). Treat the million-unit figure as a goal. Musk said on the same call that nobody can predict a new product's production ramp with precision.

How much of the supply does China control?

Think of it as bread. Mining is growing the wheat. Refining, which separates mixed ore into pure elements, is milling the flour. Making magnets is the baking. In 2024 China grew most of the wheat, milled nearly all the flour and baked 94% of the bread, according to the International Energy Agency.

The IEA calls magnet making "the main bottleneck" for countries trying to diversify. To meet demand outside China by 2035, it says, the rest of the world would have to bake six times as much bread as it does now, at a cost of around $60 billion over the decade.

What did China's export controls actually do?

On April 4, 2025, Beijing began requiring export licenses for seven medium and heavy rare earths, including dysprosium and terbium, along with products made from them, such as permanent magnets. The rules allow 45 working days to process a license, but approvals took longer than that in practice.

The effect was quick. That waiting is what idled the Ford plant in Chicago and Suzuki in Japan.

In October 2025, China moved to add five more elements. It also wanted to require licenses for some goods made abroad with Chinese rare earth inputs or technology. After the US and China reached a deal, the White House said on Nov. 1, 2025 that China would suspend the October rules and issue general licenses for rare earth exports to US users. China paused the October measures until Nov. 10, 2026, but its April 2025 licensing rules stayed in force.

What has Tesla said about Optimus?

On that April 2025 call, Musk admitted Optimus production had already been hit. Tesla was asking Beijing for permission, and he said he hoped it would "get a license to use the rare earth magnets".

Tesla hasn't said publicly whether it got that license. On the July 22, 2026 call, Musk sounded less like a man waiting on one part and more like one missing all of them. "With Optimus, there is no supply chain," he said, and he called the robot "the hardest product to scale manufacturing that we've ever made at Tesla". That transcript doesn't mention magnets or China.

Who is building a supply chain outside China?

MP Materials (MP) runs the Mountain Pass rare earth mine in California. In July 2025 the Pentagon agreed to buy $400 million of its preferred stock, guaranteed a decade-long minimum price for MP's NdPr (the neodymium and praseodymium mix used in magnets) and promised that every magnet from a new MP plant would find a buyer.

That plant, called 10X, will be built in Northlake, Texas. MP says it'll invest more than $1.25 billion and begin commissioning in 2028, which would bring its total US capacity to about 10,000 tonnes a year. Its existing Fort Worth plant started making finished magnets in 2025. General Motors and Apple are customers.

Australia's Lynas Rare Earths (LYC on the ASX) runs the world's largest rare earth refinery, in Malaysia, and now separates dysprosium and terbium there. It has $258 million in US defense funding for a heavy rare earth plant in Texas, but TIME reported in April that the deal was still under negotiation. Higher prices helped Lynas lift its fiscal 2026 net profit roughly 27-fold.

Washington has also bought a 10% stake in USA Rare Earth (USAR) for $1.6 billion, TIME reported.

The bottom line

A humanoid robot is only as available as its magnets, and most of those still come from China. The US projects are real, but MP's planned 10,000 tonnes a year would cover less than 3% of what CRU's 2040 robot forecast calls for (our arithmetic). Watch Nov. 10, when China's pause on the October rules runs out, and watch whether Tesla ever says where Optimus gets its magnets.

Musk says there's no supply chain for Optimus yet. Should Washington keep spending billions to build one, or should robot makers design their way around rare earths altogether?