Magnets (NdFeB)

Magnets are the rare-earth story, and they break the pattern of every other link here: on raw tonnage, humanoids do not bind magnets. A humanoid carries about three-and-a-half kilograms of sintered neodymium-iron-boron¹, so even a worldwide fleet of forty-seven million robots by 2040 needs only about one hundred and sixty-five kilotonnes of magnet in total, a small new slice of a global market already running near two hundred and forty to two hundred and sixty kilotonnes every year⁴. Run the four-series gap and the volume crossover simply never arrives: cumulative global production, and even China’s alone, stay comfortably ahead of the fleet’s cumulative need across the whole horizon (the global and China crossover years are both null²). The bind is a different kind entirely. It is concentration (China makes roughly ninety percent of finished magnets³ and processes about ninety-nine percent of the heavy rare earths dysprosium and terbium that high-temperature magnets need⁷); it is the upstream NdPr-oxide feedstock, near-balanced today and tipping into deficit by 2030¹¹; and it is policy: China’s April-2025 export-license regime already throttled Tesla’s Optimus line⁸. Where the tonnage view does bite is the “build it from the West” question: if the United States had to magnet-supply the entire worldwide fleet from its own factories alone, cumulative US output keeps pace through the report’s 2040 horizon and crosses the need only just beyond it⁵, and that is the optimistic reading, because it assumes every US magnet goes to robots and ignores the heavy rare earths the US still cannot separate at scale.

A sintered NdFeB magnet is the highest-energy-product permanent magnet made, the property that gives every joint and traction motor its torque density, which is why a humanoid’s roughly forty-plus motors (Optimus Gen2 runs about forty-two: rotary, linear and in the hands) all depend on it¹. We model about three-and-a-half kilograms of NdFeB per robot¹, roughly twice the two kilograms in an electric-vehicle traction motor. Crucially, small amounts of the heavy rare earths dysprosium and terbium are alloyed in to hold coercivity at high operating temperature, and those Dy/Tb additions are exactly the elements at the roughly ninety-nine-percent-China chokepoint⁷. So the magnet is not one material but a chain: iron and boron that are everywhere, light rare earths (neodymium, praseodymium) that are concentrated, and heavy rare earths that are nearly a monopoly.

Because a magnet is built into a robot and lasts its service life, magnets do not wear out and there is no replacement tail¹, the mass the fleet needs is a cumulative stock: the total number of robots ever built worldwide multiplied by three-and-a-half kilograms each, not a per-year flow. Integrating the consensus shipment ramp², the worldwide installed fleet reaches about five hundred and eighty-five thousand robots by 2030, ten million by 2035 and forty-seven million by 2040, requiring roughly two kilotonnes, thirty-five kilotonnes and one hundred and sixty-five kilotonnes of magnet respectively. In the chart below the black line is that worldwide cumulative need; the grey band is cumulative global magnet production, and the gap never opens, because global output dwarfs the robot slice. That absence of a wedge is the finding, not an error.

Production is the only thing that differs between the Global, China and US views. The worldwide humanoid fleet and the magnet mass it needs stay identical in all three; only the supply line changes, asking whether one bloc’s factories alone could equip the whole global fleet. Global sintered-NdFeB output runs from about two hundred and forty-five kilotonnes a year today⁴ toward roughly five hundred and fifty by 2040, growing the modest few percent a year that electric vehicles, wind and robots together pull¹¹. China is about ninety percent of that, the IEA puts 2024 sintered output as high as ninety-four percent³, led by JL Mag, Zhongke Sanhuan/Zhenghai and Ningbo Yunsheng, running near two hundred and twenty kilotonnes a year and growing with the global curve. US-located output is the thin line: about one kilotonne in 2025, led by MP Materials, whose Independence plant in Fort Worth began trial sintered-magnet production in late 2025 and whose announced second (“10X”) facility is expected to lift MP’s US magnet output toward roughly ten kilotonnes a year later this decade⁵, joined by Vacuumschmelze’s eVAC plant in Sumter, South Carolina (about two kilotonnes a year scaling toward twelve, roughly ninety percent committed to General Motors), Noveon Magnetics in Texas, and USA Rare Earth in Oklahoma⁶. Discounted toward realised rather than announced output, the US bloc reaches only about twenty kilotonnes a year by 2040. These post-2030 figures are a modeled extrapolation of announced lines, not a reported forecast.

Here is where magnets diverge sharply from the precision-mechanical links. The crossover, the year cumulative production drops below cumulative need², never occurs globally, and never occurs for China either. Cumulative global output runs many times the fleet’s cumulative need across the horizon⁴; the robot slice is simply too small a share of a large, growing market to overrun it. It is legitimate, and it is the point, that those two crossover years are null. The wall appears only in the US-only view, and even then it comes late: if the West had to supply the entire worldwide fleet from US factories alone, cumulative US output keeps pace across the entire 2040 horizon and crosses into deficit only just beyond it, when the durable need, by then the fleet is adding close to ten million robots a year, each carrying three-and-a-half kilograms¹, finally outruns a US line that tops out near twenty kilotonnes a year⁵. That late crossover is itself the honest verdict: the magnet problem is not a tonnage-timing problem at the press. A magnet plant is semi-elastic, brownfield sintering lines add in about two to three years⁶, and the world can press far more magnet than the robot fleet will ever need. What it cannot do is conjure the heavy rare earths those magnets require, or escape the country that controls them.

The real bottleneck sits one and two levels upstream of the magnet press, and it deepens as you descend the chain. Mining is about sixty-nine percent China⁹; NdPr-oxide separation is above ninety percent³; finished-magnet manufacture is roughly ninety percent³; and heavy-rare-earth dysprosium/terbium separation is about ninety-nine percent China⁷, the single hardest chokepoint, with no meaningful ex-China substitute and a five-to-seven-year-plus build-out. The feedstock itself is tightening: didymium (NdPr) oxide demand and supply are near-balanced today at around sixty-five thousand tonnes a year, tipping to roughly a sixteen-thousand-tonne deficit by 2030¹¹, and the NdFeB alloy/powder undersupply is projected to widen from about sixty thousand tonnes a year by 2030 toward two hundred and forty-six thousand by 2040¹². On top of geology sits policy. China’s April 2025 controls put samarium, gadolinium, terbium, dysprosium, lutetium, scandium and yttrium, plus finished magnets containing terbium or dysprosium, under export license, with case-by-case approval and processing of up to forty-five working days¹⁰. Tesla confirmed Optimus production “was impacted” and Musk sought a license⁸. An October-2025 wave expanded the list, and the November-2025 suspension lifted only that October wave, the April controls, including Dy and Tb, remain live today⁹. That is the bind: a policy hard-ceiling on a geological one, active now, categorically different from the precision-process rate-limits that cap the other links. The magnet press is not the wall; the heavy rare earths and the license regime that governs them are.

Cobots and industrial arms draw on NdFeB too. A cobot’s six frameless torque motors carry roughly 0.6 kg per arm; a traditional industrial arm’s six larger framed servo motors carry about 3.5 kg, comparable to a humanoid, so the ~6-million industrial-arm fleet adds on the order of twenty thousand tonnes by 2040 and the cobot fleet a few thousand. Both are small against the fleet’s total tonnage and neither changes the shape of the magnet story: the bind here is geopolitical (China refines ~90% of rare earths), not a volume shortfall. Quadrupeds are a lighter draw still, their twelve BLDC joint motors carry only about 0.35 kg of NdFeB in total, a fraction of a humanoid’s 3.5 kg, so even a quadruped fleet of about 5.15 million by 2040 adds well under two thousand tonnes, a modest increment that likewise rides the same rare-earth concentration rather than any tonnage limit. Drones are the surprise here: they draw nothing on the precision spine, but their four-to-eight little outrunner prop motors each carry a few grams of NdFeB, and at the drone fleet’s enormous scale (cumulative units built into the hundreds of millions by 2040) that sums to roughly ten kilotonnes, well under a tenth of the whole humanoid magnet draw, yet still the single largest non-humanoid tonnage on this link. It is still only about three-tenths of one percent of annual NdFeB output, so it does not move the tonnage picture; but it pulls the same China-refined feedstock, so the world’s biggest robot fleet quietly deepens exactly the one dependency that makes magnets a geopolitical rather than a volume story.