For the past few years, the story of artificial intelligence has been told in software. Bigger models, longer context windows, cleverer training runs. But as AI moves off the screen and into the world â into humanoid robots, warehouse arms, delivery machines and autonomous factory equipment â the bottleneck is shifting somewhere far less glamorous: the motor, and specifically the magnet inside it.
Why magnets matter
Every robot joint that moves does so because an electric motor turns. The most compact, efficient and responsive motors used in robotics rely on permanent magnets, typically made from rare earth elements such as neodymium and dysprosium. These magnets allow a small actuator to deliver high torque without overheating or ballooning in size â exactly what you need in a robot that has to fit through a doorway, balance on two legs, and run for hours on a battery.
A humanoid robot may contain dozens of such actuators. Multiply that by the production volumes companies are promising, and the arithmetic starts to look uncomfortable. Software can be copied at essentially zero marginal cost. Magnets cannot.
A supply chain that doesn’t scale like software
The rare earth supply chain is concentrated, capital-intensive and slow to expand. Mining, separation and refining of rare earth elements are chemically demanding processes with significant environmental permitting requirements, and building new capacity is measured in years, not quarters. Processing in particular has historically been dominated by a small number of players, which makes the market vulnerable to export controls, trade disputes and price swings.
That is a very different risk profile from the one AI companies are used to managing. A model developer facing a compute crunch can rent capacity elsewhere, optimise its training pipeline, or wait for the next chip generation. A robotics company facing a magnet crunch has to renegotiate physics, chemistry and geopolitics simultaneously.
The engineering workarounds
There are alternatives, and they are being pursued seriously. Motor designs that reduce or eliminate heavy rare earth content, switched reluctance and induction motors that use no permanent magnets at all, and ferrite-based magnets all offer paths around the constraint. Each involves trade-offs, usually in power density, efficiency or control complexity â precisely the qualities that make a robot feel capable rather than clumsy.
Recycling is another lever. Magnets from retired hard drives, wind turbines and electric vehicle motors represent a growing secondary supply, though collection and separation infrastructure remains immature. Materials research into new magnetic compounds continues, but moving a promising laboratory result into mass production is a decade-scale endeavour.
What this means for the industry
The implication is not that physical AI is doomed. It is that the competitive advantages in robotics may look less like the ones that decided the language model race. Vertical integration, long-term materials contracts, motor design expertise and manufacturing know-how could matter as much as the quality of a company’s control policies or perception stack.
It also suggests a reordering of who wins. The firms best positioned may not be the ones with the most impressive demo videos, but the ones that have quietly secured a supply of the unglamorous components that make those demos repeatable ten thousand times over.
AI’s next chapter is being written in atoms as much as in code. And atoms, unlike parameters, have to be dug out of the ground. Read More
Leave a Reply