As humanoid robots move from research labs to factory floors, the industry keeps talking about artificial intelligence: better reinforcement learning, bigger world models, more sophisticated vision systems. Yet the hardest problem in physical AI may have nothing to do with the model and everything to do with the machine. The limiting factor is hardware, specifically the compact actuators that move joints, and inside those actuators, the rare earth magnets that generate torque.
Actuators account for an estimated 40% to 60% of a humanoid robot's bill of materials, according to consulting analysis cited in a recent industry commentary. That is a startling concentration of cost. A humanoid robot is not a laptop with arms; it is a dense bundle of precision mechanical components, and the motor-gearbox assembly at every joint is a small factory of its own. With a limited pool of qualified manufacturers, supply is tight. New production capacity takes years, not quarters, to bring online because the tolerances, testing requirements and certification cycles are severe.
The viewpoint comes from an executive with a direct stake in the supply chain. He co-founded a manufacturing sourcing platform that was later acquired by a Japanese components distributor for $350 million, and he now runs the distributor's AI unit. His argument is that the scarcity is structural. The supply chain has not built up in anticipation of humanoid deployment. The current wave of humanoid enthusiasm is newer than the multi-year capital investment cycle needed to build actuator plants. As a result, any company planning to scale from thousands to millions of robots will quickly confront a procurement crisis before they ever reach an AI training bottleneck.
Where the cost sits: gearboxes
Within an actuator, the most expensive and most constrained component is the gearbox. The gearbox accounts for roughly 30% to 50% of the actuator's cost. In high-torque joints, designers typically use harmonic drives or strain wave gearboxes. These devices provide high reduction ratios in a compact package, with low backlash, making them ideal for robot joints. But the market is not wide. Harmonic Drive and Nabtesco are among the few trusted producers, and component qualification cycles are long. A humanoid manufacturer cannot simply switch suppliers overnight. The gearbox must be validated for durability, repeatability, backlash, load tolerance and thermal behavior, often across thousands of hours of testing.
The dependence on a small number of gearbox suppliers amplifies the risk. If demand for humanoid robots surges, these manufacturers will allocate limited capacity to their most reliable customers. New entrants cannot easily break in because they need robot makers to design them into the product, and robot makers need proof of reliability before they commit. This chicken-and-egg problem is common in precision engineering, but it is especially acute in robotics because the performance demands are so high.
The magnet underneath everything
Below the gearbox and the copper windings lies an even harder dependency: the permanent magnet. Every high-torque joint relies on neodymium magnets, which generate the magnetic field that converts electrical current into motion. Neodymium is a rare earth element, and while the mining is relatively diversified, the processing stage is not. China accounts for roughly 69% of rare earth extraction, but it controls about 90% of the processing capacity for rare earth magnets. The refining process, including the separation of individual rare earth elements and the production of neodymium-iron-boron alloy powders, is technically demanding and environmentally sensitive. Most of the world's magnet-making capacity sits in China, which means that a robot assembled in Europe or North America is still dependent on Chinese refined material.
This concentration is not new. Rare earths have been a geopolitical issue for more than a decade, with export restrictions and supply chain disruptions forcing manufacturers to keep buffer stocks. For humanoid robots, the exposure is sharper because the robot needs many small, powerful magnets, not just one large magnet in a wind turbine generator. The magnet content is spread across every joint, multiplying the supply chain challenge.
Europe writes the rules
Europe has recognized the vulnerability and written it into law. The Critical Raw Materials Act includes light and heavy rare earths on its strategic raw materials list. It sets a 2030 target that no single third country supplies more than 65% of the European Union's consumption of those materials. That is a deliberate cap on dependence on China, but it is a policy target without an immediate industrial solution.
The law also reaches directly into the robot supply chain. Articles 28 and 29 name industrial robots explicitly. From 2028, permanent magnets weighing more than 0.2 kilograms must carry a label, be accompanied by a digital record, and report a recycled content figure. That label is not a trivial requirement. It forces robot makers and magnet suppliers to trace the origin of the magnet, measure the recycled fraction, and register the product in a digital system. For a component that has been historically treated as a commodity, the administrative burden is substantial.
The law's intention is to create a transparent market and to incentivize recycling and diversification. But it also creates tension in the short term. If European manufacturers cannot source quickly enough from outside China, the label will simply document their continued reliance. A policy that demands visibility does not by itself create refining capacity.
Schaeffler's bet on humanoids
Europe's industrial response is, for now, largely concentrated in one company: Schaeffler. The German motion technology company has invested in Humanoid, a London-based startup that raised $152 million in July. Bosch, another German industrial giant, is building the robots, bringing manufacturing expertise to the project.
Schaeffler is not just an investor; it is also a supplier. Under an agreement running to 2031, Schaeffler is the preferred supplier for more than half of Humanoid's joint actuators. The agreement covers a seven-digit number of units, meaning at least one million actuators. That gives some sense of the scale the companies expect. One million actuators is not a laboratory run; it is a manufacturing program.
Schaeffler is also putting its money where its factory is. It plans to deploy a four-digit number of Humanoid robots into its own plants by 2032, with the first deployment starting at Herzogenaurach in December. This is a textbook case of a company creating its own market. By deploying the robots internally, Schaeffler can test the actuators in real production environments, gather data, and demonstrate the technology to other industrial customers. It also gives Humanoid a credible reference site.
The numbers are large but still tiny compared to the global manufacturing workforce. A four-digit deployment across Schaeffler plants is meaningful for the robot company, but it is a fraction of the millions of jobs in the company's own supply chain. Still, the investment signals that Schaeffler sees humanoid robots not as a science project but as a business line that will use hundreds of millions of euros worth of components.
The gap between law and capacity
Here is the lingering contradiction. An actuator assembled in Herzogenaurach still contains a magnet refined in China. No amount of supplier qualification can change that fact today. The European Union has set a 65% cap that takes effect in 2030, and a labelling regime that begins in 2028. But it has almost no refining capacity of its own to meet either requirement. The gap is a policy gap rather than a purchasing gap. Even a company with deep pockets and long-term contracts cannot source what does not exist.
For the robotics industry, the strategic lesson is that the entire supply chain matters. AI models improve quickly, and software capabilities double or triple within a few years. But the physical supply chain—magnet refining, gearbox machining, motor winding—moves at the pace of chemical plants, precision foundries and factory floor audits. That is measured in decades, not release notes.
Europe may still win on deployment, even if China controls the hardware chain. The combination of Schaeffler's manufacturing, Bosch's production know-how, and Humanoid's software could create a strong European ecosystem. But winning on deployment only matters if the hardware can actually be produced. The actuator bottleneck is the constraint that will decide whether humanoid robots become a mass-market product or remain an expensive demonstration. And the magnet sits at the bottom of that bottleneck.