Humanoid Robots Have a Materials Problem

Humanoid robotics is not only an AI and dexterity story. If the category moves from pilots to fleets, the scarce layer may be the physical stack: actuators, motors, magnets, batteries, copper, charging infrastructure, and the suppliers that can qualify those parts at scale.

What changed

The humanoid robot debate is starting to move downstream from model intelligence and upstream into the physical supply chain.

Goldman Sachs has lifted its public humanoid market outlook to a $38 billion addressable market by 2035, with projected shipments of 1.4 million units. The important detail is not only the larger number. It is the reason the forecast moved: faster AI progress, lower component costs, more supplier options, and a clearer path toward industrial commercialization.

That turns humanoids into more than a software adoption curve. A robot fleet is a dense physical system made from actuators, motors, reducers, sensors, batteries, wiring, power electronics, magnets, tooling, and serviceable parts. If deployments scale, the bottleneck may not be only whether the model can generalize. It may be whether the hardware stack can be sourced, manufactured, qualified, and maintained at fleet volumes.

Why it matters

A humanoid robot looks like an AI product in a demo video. At scale, it behaves more like an industrial manufacturing program.

Every additional unit pulls on several constrained layers:

• actuation and precision motion
• permanent magnets and motor materials
• copper wiring and power electronics
• battery packs and charging systems
• thermal management and reliability engineering
• high-precision manufacturing equipment
• quality control and field-service loops

That matters because humanoid demand does not arrive in an empty supply chain. It competes with AI data centers, grid upgrades, EV batteries, defense modernization, industrial automation, and electrification. Robotics may become another claim on the same copper, magnet, battery, and power-infrastructure base that already has multiple demand curves stacked on top of it.

The materials layer

The most useful way to frame the issue is not “robots need metals.” It is more specific: humanoids are actuator-dense machines.

Actuators and motors

Humanoid robots need many degrees of freedom across legs, arms, wrists, hands, and torso systems. That pushes demand toward compact motors, precision reducers, bearings, encoders, thermal envelopes, and often rare-earth permanent magnets.

The magnet layer is important because high-performance motion systems commonly rely on NdFeB magnets, which in turn link to neodymium and praseodymium supply chains. Rare-earth mining, separation, alloying, and magnet production are not evenly distributed globally. China remains a central node in the downstream rare-earth magnet supply chain.

Copper and power

Copper shows up in motor windings, wiring harnesses, power electronics, charging systems, and broader electrical infrastructure. A single robot is not the issue. Fleet deployment is the issue.

If humanoids remain a low-volume industrial pilot category, the incremental copper load is manageable. If robots become a real fleet product in factories, logistics sites, warehouses, and eventually homes, the copper story becomes part of the same power-infrastructure conversation already happening around AI compute and grid modernization.

Batteries and charging

Humanoids also require batteries, battery-management systems, charging stations, and duty-cycle planning. The battery layer ties robotics to lithium, nickel, cobalt, graphite, and alternative chemistries depending on design choices.

The open question is not simply pack size. It is utilization. A robot that works long shifts in an industrial setting needs uptime, predictable charging windows, safe battery behavior, and field-replaceable systems. That turns energy storage into an operating constraint, not just a bill-of-materials line.

The demand collision

The most important interpretation is that humanoid robots are emerging into a crowded physical economy.

The same decade may include:

• AI data-center buildouts that need power, copper, transformers, cooling, and grid connections
• EV and battery supply chains still competing for energy materials
• industrial automation upgrades in manufacturing and logistics
• defense and reshoring demand for secure supply chains
• grid modernization driven by electrification and compute load
• humanoid robots adding another motion-and-power-heavy hardware category

This does not mean every aggressive robot-unit forecast should be treated as base case. The 2030s scenarios range from conservative industrial adoption to extremely aggressive projections of hundreds of millions or billions of humanoids. Those high-end scenarios are useful stress tests, not underwriting assumptions.

The base case is simpler: even modest humanoid volume growth creates a new demand vector in components that already have strategic value.

Interpretation

The robotics supply chain should be tracked like a bottleneck chain, not only like an AI adoption curve.

The first wave of value capture may not sit only with humanoid OEMs. It may also sit with companies that become qualified into the physical stack:

• actuator and joint-module suppliers
• motor, reducer, bearing, and encoder suppliers
• magnet and rare-earth processing layers
• battery, power-electronics, and charging suppliers
• precision manufacturing equipment vendors
• test, inspection, and quality-control systems
• industrial customers that provide real deployment data

The key distinction is qualification. A supplier mention in a robotics deck is weak evidence. A supplier that survives engineering validation, safety testing, reliability cycles, cost-down pressure, and repeat fleet orders is much stronger evidence.

Market read-through

For Robotics Radar, the market read-through is not “buy metals because robots.” That is too broad and too sensitive to adoption assumptions.

The better research path is to map which physical layers become harder to replace as robots move from pilot to fleet:

• Which actuator designs are winning, and how material-intensive are they?
• Which OEMs are vertically integrating motors, actuators, batteries, or hands?
• Which suppliers are being designed into repeat deployments rather than one-off prototypes?
• Which rare-earth and magnet supply chains are exposed to robot-motor demand?
• Which companies benefit if humanoids need more power infrastructure, charging, and field service?
• Which bottlenecks can be designed away through different architectures?

This is also where the robotics story overlaps with the broader AI-infrastructure story. Data centers are the visible power demand. Humanoid fleets would be a distributed physical-AI demand layer: smaller per site, but potentially widespread if deployment works.

Layers to watch

Actuation stack

The highest-signal hardware layer. Track motor architecture, reducer choice, integrated joint modules, hand actuation, torque density, thermal behavior, serviceability, and production yield.

Magnets and rare earths

Track NdFeB magnet demand, rare-earth separation and refining capacity, China exposure, substitution attempts, recycling, and design choices that reduce rare-earth intensity.

Copper and electrical infrastructure

Track robot wiring, motors, charging stations, factory electrical upgrades, transformer availability, and whether robot fleets become another small but repeatable load on already constrained power systems.

Batteries and uptime

Track runtime, charging time, pack replacement, safety, battery chemistry, and whether industrial duty cycles force designs toward swappable packs or managed charging workflows.

Manufacturing equipment and quality

Track high-precision grinding, machining, testing, end-of-line inspection, yield learning, and the equipment needed to make motion systems reliable at volume.

Open questions

• How much copper, magnet material, and battery capacity does a production humanoid actually require?
• Will leading OEMs converge on similar actuator architectures, or fragment across different material-intensity profiles?
• Can humanoid designs reduce rare-earth dependence without sacrificing torque density and efficiency?
• Does the first commercial wave create real incremental metals demand before 2030, or is this primarily a 2030s story?
• Which suppliers move from prototype support into qualified fleet production?
• Does China’s supply-chain depth become a decisive advantage in early humanoid scale-up?
• Are metals and power infrastructure a true bottleneck, or merely a second-order constraint until robot demand is much larger?

The near-term robotics question is still deployment evidence: uptime, interventions, task ROI, and repeat customer orders. But if those numbers start working, the next question becomes physical scale.

Humanoid robots do not just need better brains. They need a supply chain that can build bodies by the thousands, then millions.

Not investment advice. Research notes only.