China holds 40.4% of the 255-company humanoid universe by name and a moderate-looking country HHI of 2,468 — the numbers of a leader, not a chokepoint. The chokepoint is one node down: China supplies roughly 100% of the rare-earth sintered magnets every joint motor needs (94% of NdFeB by mass), 72.7% of precision screws and 63.6% of reducers, and since April 2025 its dysprosium and terbium export licences sit on exactly those parts — according to InfraMosaic's Humanoid Value-Chain database.
A national average that hides the weapon
Concentration indices are built to reassure. The humanoid universe's country HHI of 2,468[ledger] sits only just inside the "moderately concentrated" band — below semiconductors, below EV cells. China's 40.4% share by company count[ledger] looks like dominance you could engineer around: the United States holds 23.9%, Japan 14.5%, and twelve other countries carry a name apiece. The headline number is the number of a crowded, contestable market.
That average is the wrong lens. A humanoid robot is not 255 companies in proportion — it is a bill of materials in which a handful of parts are load-bearing for everything else. When you re-weight the same universe by where the value chain actually narrows, the picture inverts. China does not own 40% of every node. It owns ~100% of the nodes that cannot be substituted, and a contestable slice of everything that can. That is not a broad lead. It is a surgical one.
The three nodes that decide everything
InfraMosaic scores every one of the 25 value-chain nodes on a 1–5 criticality scale and a China-dependence band. Four Body nodes carry the maximum criticality of 5 and a "High" or "Very High" China-dependence band — and they are the four that move every limb a humanoid has. Three are the irreplaceable trio:
| Value-chain node | Crit. | China share | Node HHI | Export-control lever |
|---|---|---|---|---|
| Rare-earth magnets | 5 | ~100% | 10,000 | Dy / Tb feedstockCONTROLLED |
| Screws & linear motion | 5 | 72.7% | 5,703 | roller-screw alloy & grind |
| Reducers & precision gears | 5 | 63.6% | 4,545 | harmonic / RV gearing |
| Actuators & servo motors | 5 | 55.6% | 4,012 | inherits magnet constraint |
Notice what the HHI column is doing. The universe HHI is 2,468 — moderate. But every one of these nodes is two-to-four times more concentrated than the headline, and the magnet node maxes the scale at 10,000: a single-country monopoly. The "moderate" national average is an arithmetic artefact of averaging a monopoly node against software nodes the West still owns. Strip out the parts a robot can live without, and the concentration of the parts it can't lives at the very top of the scale.
Why magnets are the keystone — and why Dy/Tb is the lever
A humanoid carries roughly 3.5 kg of NdFeB sintered magnet across its 28-plus frameless joint motors. There is no non-magnetic substitute at the torque density humanoids need, and China produces about 94% of global NdFeB by mass. That alone is a chokepoint. What makes it a policy chokepoint is the heavy rare earths that make those magnets survive a hot joint: dysprosium and terbium. Both sit under China's export-licence regime introduced in April 2025, and both are roughly 90% China-supplied.
The price tells you the regime is binding. Dysprosium oxide assesses at about $230/kg inside China against roughly $800/kg in Europe — a 3.5× export-control premium on the same oxide. Terbium oxide runs to about $1,600/kg. These are not mined-scarcity prices; they are licence prices. The lever is precise: it does not touch the 40% of the value chain China merely competes in. It touches the dopant that turns a commodity magnet into one that can run inside an actuator — which is to say, it touches all of it, one molecule deep.
Rystad asks you to trust the trace. InfraMosaic lets you check it.
Every headline figure in this article resolves to a hash-chained record in the Publication Ledger — source files, SHA-256 inputs, the formula reference and the AQuA sign-off. Don't take the desk's word for it. Open the record.
The lever lands where the cost — and the bottleneck — already concentrate
If the export regime sat on a cheap, abundant part, it would be a nuisance. It sits where money and scarcity already pile up. On InfraMosaic's reference bill of materials, rebased on Morgan Stanley's Exhibit 54, the drivetrain China dominates is most of the robot: screws 20%, motors 20%, reducers 13% — and the motors run on the magnets China controls outright. Roughly half the hardware cost of a humanoid flows through nodes where China's share runs from 56% to ~100%.
Supply tightness points the same way. InfraMosaic's node-by-node order book has planetary roller screws running at 770% of projected 2030 capacity — the single tightest mechanical bottleneck in the build — and the magnet node flagged "deficit-risk" not because of factory capacity but because the rare-earth input is the binding constraint. The two places the value chain is physically tightest are two of the three places the export lever already sits. The West's most exposed pure-plays — Japan's Harmonic Drive (InfraMosaic exposure score 92) and Nabtesco (84) — make the reducers, not the magnets that feed them. The keystone is upstream of the West's best cards.
What to watch
- The Dy/Tb premium spread. The $230-vs-$800 dysprosium gap is the cleanest real-time read on how tightly the licence regime is being enforced. A widening spread is the lever being pulled; a narrowing one is reshoring starting to bite.
- Ex-China magnet tonnage. US capacity is barely 6,000 tonnes by 2027 against China's ~300,000 today. Until that number moves an order of magnitude, "magnet diversification" is a press release, not a supply chain.
- Roller-screw localisation outside China. At 770% of 2030 capacity, screws are where a Western or Japanese line could most change the map — watch new precision-grind capacity, not unit-price headlines.
- Whether the headline HHI moves. It won't fall meaningfully while the magnet node sits at 10,000. The number to track is node-level concentration, not the reassuring 2,468 national average.
The investable mistake is to read 40.4% and HHI 2,468 as a market you can engineer around. The data says the opposite. China's lead is narrow in coverage and total in depth — it owns the few parts that gate the many. The export-control regime didn't create that asymmetry; it weaponised the part of it that was already irreplaceable. Diversification will come, node by node, screw by screw. The magnet keystone goes last.
Method. Figures are drawn from InfraMosaic's Humanoid Value-Chain database (build 2026-06-13, methodology v1.0) — a 255-company universe mapped to 25 value-chain nodes with criticality, China-dependence, HHI and a node-by-node supply/demand order book. Headline KPIs kpi.china_share_pct (40.4%) and kpi.overall_hhi_country (2,467.6) resolve to hash-chained, AQuA-reviewed records in the Publication Ledger (363 records, 134 integrity tests passing). Rare-earth pricing reflects 2025-Q4 assessments; Dy/Tb export controls dated April 2025. InfraMosaic is independent research; nothing herein is investment advice.