A $5-trillion industry is being built one actuator at a time — and three-quarters of the supply chain runs through one country. A map of who makes the robots, what they cost, and where the choke-points are.
The humanoid robot has stopped being a demo and started being a supply chain. In 2025 the first paid pilots landed on factory floors — Figure at BMW, Apptronik at Mercedes, Agility at GXO — while Chinese makers shipped robots by the thousand. Behind every one of them sits a value chain of actuators, reducers, roller screws, rare-earth magnets, sensors and AI compute that, until now, nobody has mapped at the level the money requires.
This is that map. InfraMosaic tracks the public-market humanoid value chain end to end. The findings below are drawn from the live database; the numbers update when the data does.
Every humanoid resolves into three layers. The Brain — AI models, compute, vision, simulation — is where the West and Taiwan lead. The Body — the electro-mechanical stack of motors, gears, screws, sensors, magnets and batteries — is where China has built an almost unassailable position. The Integrators assemble it all into a working robot. The economics live in the Body: joint drive modules alone are 60–70% of a humanoid's cost.
Nodes flagged as bottlenecks — concentrated supply and hard to substitute — ranked by criticality and China dependence.
At base-case 2030 volumes those choke-points translate into hard numbers: demand for planetary roller screws and integrated actuators runs several times today's assessed production capacity, while LiDAR and batteries sit in glut. The full supply/demand order book — component demand from the robot ramp against assessed capacity — is tracked live in the InfraMosaic platform.
The reducer, the roller screw and the rare-earth magnet are to the humanoid what the lithium cell was to the electric car: small, unglamorous, and the whole game.
Share of each segment's tracked companies headquartered in China. The Body is the exposure; the Brain is the hedge.
The risk is not abstract. In 2025 China placed export licences on dysprosium and terbium — the heat-resistance dopants in the high-temperature magnets inside every frameless torque motor. Those two elements sit directly on the highest-criticality nodes in the chain. A humanoid built entirely outside the Chinese supply chain costs roughly three times as much. That spread is the single most important number in the industry, and the strongest case for a price-reporting agency to exist.
Reference-robot cost, indexed to January 2024 = 100. A modelled back-cast pending a live monthly assessment panel.
The reference humanoid — an Optimus-Gen-2-class machine of ~28 degrees of freedom — carries a material bill of materials of about . Counted by component, sensors are the single largest line — about 37% of the bill, driven by the six-axis force-torque sensors in the hands and feet; motors and roller screws are roughly a fifth each, reducers an eighth. The planetary roller screw, fourteen to a robot, is the most-watched cost-down lever in the industry, already cut from $3,000 to $800 at Tesla's negotiated scale — but the force-torque sensor is the quieter choke-point, near-monopolised by ATI and Novanta.
A selection from the full program directory. Price = announced, target or third-party listing.
The deployment data tells the real story: most automotive "fleets" are still low-unit pilots, while China's open-market sales — Unitree and AgiBot each claiming 5,000-plus units in 2025 — are where volume actually lives. The gap between a vendor-claimed swarm and an audited deployment is exactly the signal a buyer pays an intelligence service to draw.
Published forecasts span two orders of magnitude. InfraMosaic's base case sits deliberately conservative on hardware units.
Morgan Stanley sees a billion humanoids and a $5-trillion market by 2050. Elon Musk says ten billion. Somewhere between the spreadsheet and the dream is a number you can underwrite — and that number is the product.
InfraMosaic sources its company universe independently and uses Morgan Stanley's Humanoid 100 only as a cross-check. Exposure scores are forward-looking assessments against a documented rubric. Prices are tagged HARD (sourced) or ASSESSED (triangulated). Concentration is measured by the Herfindahl-Hirschman Index of supplier countries. The cost index is a modelled back-cast pending a live, audited assessment panel — the next milestone on the path to a contract-grade reference. Full methodology and provenance accompany the dataset.