Humanoid robots could require 271,000 to 449,000 tonnes of NdFeB magnets by 2040, equivalent to 120–200% of forecast 2026 demand [S2].
Each robot typically carries 2–4.5 kg of sintered NdFeB in 30 or more permanent-magnet motors, requiring an estimated 0.57–1.3 kg of NdPr and 35–79 g of dysprosium [S1][S3].
How joints drive magnet consumption
Humanoid robots require up to 70 degrees of freedom to achieve human-like movement, and permanent-magnet motors are generally preferred for these joints, except for lower-power applications such as fingers [S2].
A typical humanoid contains 2–4.5 kg of sintered NdFeB. The rotor carries a ring of alternating magnetic poles, while the copper stator creates a rotating field and the reducer or screw converts rotation into joint torque [S3].
Sintered NdFeB provides high magnetic energy density, allowing a compact motor to produce useful torque without excessive limb mass [S3].
The demand calculation starts with magnet mass, not motor count. Published estimates put magnet use at 2–4.5 kg of NdFeB per humanoid, compared with about 1.4 kg in an electric car [S1][S3].
NdPr adds volume, dysprosium adds exposure
Light rare earths Nd and Pr provide bulk magnetization, while dysprosium and terbium improve heat resistance by replacing some NdPr [S1].
For 1 kg of H-grade NdFeB, the US Department of Energy composition data cited by FAI gives 28 g of dysprosium and 272 g of NdPr. SH-grade material contains 42 g of dysprosium and 258 g of NdPr per kilogram [S1].
Most NdFeB magnets contain less than 5% heavy rare earth by weight. Dysprosium and terbium are produced at negligible quantities outside the Chinese sphere of influence, while terbium is approximately 2.5 times as effective per gram for heat resistance and commands more than four times the price of dysprosium [S1].
At 1 million humanoids per year, additional demand would be roughly 7–15% of EV magnet demand and 35,000–79 tonnes of dysprosium. That compares with about 2,700 tonnes of dysprosium sold worldwide in 2024, making the heavy rare earth component more constrained than NdPr alone [S3].
Three uptake cases define the demand range
CRU’s low case reaches 271,000 tonnes of humanoid NdFeB demand, or 120% of forecast 2026 demand. Its base case reaches 363,000 tonnes, or 160%, and its high case reaches 449,000 tonnes, or 200% [S2].
The base case equates to 151 million humanoid robots by 2040 and 363,000 tonnes of NdFeB. The high case adds another 86,000 tonnes of magnet demand above the base case, requiring proportionately more NdPr, dysprosium or terbium depending on magnet grade [S2].
The range reflects robot-unit uncertainty as much as material efficiency. Forecasts cited by FAI range from 700,000 robots per year by 2035 to 10 billion by 2040, implying radically different supply requirements even if the per-robot magnet load falls [S1].
CRU does not expect significant humanoid magnet demand before 2035, as the technology matures. The forecast therefore represents a medium-term capacity issue rather than an immediate 2026 demand shock [S2].
Material options trade torque for mass and heat
NdFeB permanent magnets offer higher energy density than conventional magnet materials, helping keep high-torque humanoid actuators compact and efficient [S5].
Ferrite is a non-rare-earth alternative, but the same torque requires approximately 30% more motor mass. In a walking machine, that penalty raises the load on the next joint, which can require a larger motor and additional cooling capacity [S3].
NdFeB also has a thermal limit because coercivity falls as temperature rises. Nominal maximum operating temperatures rise from 80 °C for N grade to 100 °C for M, 120 °C for H, 150 °C for SH, 180 °C for UH, 200 °C for EH and 220–230 °C for AH grade [S3].
Grain-boundary diffusion can reduce the dysprosium needed in some magnet grades, but the resulting composition and thermal performance must be qualified for each actuator design. Copper, current and cooling can substitute for some magnet grade, yet they cannot remove the mass penalty without changing motor size and system load [S1][S3].
Humanoids are material for EV and wind supply
EVs already consume as much heavy rare earth annually as the production of 26 million humanoid robots, so the humanoid market would not be the first large customer for these materials [S1].
That comparison does not remove the bottleneck. Wind turbines and electric vehicles are expected to drive magnet rare earths into deficit independently of humanoid adoption, with the robotics surge adding pressure from 2035 onward [S2].
Humanoids also shift the mix toward light NdPr, for which Western production capacity is more substantial, while preserving demand for controlled heavy rare earths. China produced about 94% of global sintered permanent magnets in 2024, so geographic concentration remains the central procurement risk [S1][S3].
That concentration makes recycling and non-China refining capacity relevant to the robot supply chain. The comparison with EV traction motors is covered in EV traction motor magnet supply risk: NdFeB dependence in 2026.
Procurement should track three measurable signals
They should also separate NdPr offtake from dysprosium and terbium procurement, because the 2.5-to-1 dysprosium-to-terbium heat-resistance equivalence does not make their supply risks interchangeable [S1].
They should also verify actuator magnet grade, weight and motor count for each robot platform. Moving from H to SH grade changes the cited elemental composition from 28 g to 42 g of dysprosium per kilogram of magnet, despite only a 30 °C increase in the nominal temperature class from 120 °C to 150 °C [S1][S3].
The clearest scale signal is annual humanoid output. One million units per year would add 2,000–4,500 tonnes of NdFeB demand, equivalent to 7–15% of EV magnet demand [S3].
The more consequential signal is supply-chain concentration: China held about 94% of global sintered permanent-magnet production in 2024, while projected 2026–2040 demand scenarios reach 271,000–449,000 tonnes of humanoid magnet requirements [S2][S3].