Humanoid Robot Supply Chain: Companies & Value Chain
The robotics value chain starts with rare-earth magnets and ends with a machine that either does a paid shift or doesn't. In between sit the precision reducers, joint actuators, force sensors, and edge compute that decide whether a humanoid stands up at all. Here are the listed companies actually shipping parts into that chain in 2026 — plus an honest read on which stages already earn money and which are still running on slide decks.
Industry Status (2026)
Robotics is two industries wearing one trench coat. Industrial arms and cobots have been billing customers for forty years; humanoids have been billing them for about eighteen months. These bars separate the two so the chain below reads honestly — they describe where the industry stands, not what anyone should do about it.
The Eight Stages at a Glance
Every stage below has at least one listed company in it, so the useful question isn't "who is here" but "what is holding this stage back, and what would tell you it's clearing." That's what this table is for — the detail follows underneath.
| Stage | Maturity | Bottleneck | What would confirm progress |
|---|---|---|---|
| 01 Magnets & materials | Scaling | China refines and sinters the overwhelming majority of NdFeB | Qualified magnet output from non-China plants, not announced capacity |
| 02 Precision reducers | Scaling | Qualification inside a specific robot design, not factory capacity | Named production wins at volume pricing |
| 03 Actuators & joints | Scaling | Integrating motor, gear, encoder and sensing into one safety-rated module | Repeat purchase orders rather than first orders |
| 04 Sensing & perception | Early | Still in design-win phase; humanoid unit volumes are tiny | Humanoid order lines growing an order of magnitude |
| 05 Compute & AI models | Developing | Running a capable model on-robot inside a real power budget | Vision-language-action models running on production fleets |
| 06 Robot batteries | Early | Energy density and safety in a pack that stands next to people | 2027 mass-production dates holding, then a named robot program |
| 07 Robot builders | Mature Early | Industrial arms are mature; humanoid unit economics are not | Humanoids shipping to customers who aren't the manufacturer |
| 08 Deployment & demand | Early | Return on investment per task, measured against a human doing it | Pilots converting to paid multi-site rollouts |
Stage 07 carries two badges on purpose: FANUC and Yaskawa have been billing customers for forty years, and the humanoid makers next to them have been doing it for roughly eighteen months. Averaging those two into one label would hide the only thing that matters about this stage.
1 Rare-Earth Magnets & Critical Materials
Strip the marketing off a humanoid and you find a bag of magnets. Industry estimates put roughly 2–3 kg of NdFeB magnet in a single humanoid — several times the magnet content of a typical EV — because torque density across 25 to 30 joints is what decides whether the robot lifts a tote or just poses for the camera. China refines and sinters the overwhelming majority of that magnet supply, which is why this stage stopped being a mining story and became a policy story: from January 1, 2027, US defense sourcing rules extend restrictions on Chinese-origin NdFeB across the full supply chain.
Listed names mapped to this stageMP Materials (NYSE MP), Lynas Rare Earths (ASX LYC), USA Rare Earth (NASDAQ USAR), JL MAG Rare-Earth (SZSE 300748)
2 Precision Reducers & Gearing
Here is the part almost nobody outside the industry can name and almost every robot depends on. A reducer sits between motor and limb, trading speed for torque; a humanoid needs somewhere between 14 and 28 of them, and they are typically the single most expensive line in the bill of materials after compute. The structure held for decades — Harmonic Drive owned the strain-wave lane, Nabtesco owned the RV cycloidal lane, and both printed money. Then Nabtesco sold down the roughly one-fifth of Harmonic Drive it had held since 2008, and a group of Chinese challengers arrived priced 30–40% below the Japanese incumbents. The duopoly is now a race.
Listed names mapped to this stageHarmonic Drive Systems (TSE 6324), Nabtesco (TSE 6268), Leaderdrive (SSE 688017), SPG (KOSDAQ 058610), Nidec (TSE 6594)
3 Actuators, Screws & Joint Modules
The industry is quietly moving on from "buy a motor, a gear, an encoder and a sensor, then bolt them together." The unit of purchase is becoming the integrated joint module: frameless torque motor, reducer, encoder and thermal sensing in one housing. More of them now ship with functional safety ratings (Cat 3, PLe, SIL 3), because these machines work within arm's reach of people. It's also the stage where the automotive supply chain walked in and made itself at home: if you can mass-produce a million electric power steering units, a robot joint is a familiar problem.
Listed names mapped to this stageSchaeffler (XETRA SHA), THK (TSE 6481), Sanhua Intelligent Controls (SZSE 002050), Hyundai Mobis (KRX 012330), HL Mando (KRX 204320), Robotis (KOSDAQ 108490)
4 Sensing & Perception
Two philosophical camps, both funded. Tesla's Optimus is camera-only, in keeping with house doctrine; nearly everyone else pairs cameras with lidar because unstructured environments punish vision-only stacks in ways a demo video never shows. The more interesting frontier is touch: force, torque and tactile sensing are what separate a robot that moves from a robot that can pick up an egg, and a single humanoid may carry dozens of distributed tactile nodes. Watch the order sizes here closely — they are small, which is precisely the point. This stage is in design-win phase, not volume phase.
Listed names mapped to this stageHesai Group (NASDAQ HSAI), Ouster (NASDAQ OUST), Vishay Precision Group (NYSE VPG), Novanta (NASDAQ NOVT), Aeva Technologies (NASDAQ AEVA)
5 Compute & Robot Foundation Models
This is the stage that turned robotics from a mechanical engineering discipline into an AI one. Vision-language-action models let a robot be told what to do in a sentence instead of scripted in a teach pendant, and the compute has to run on-board, in real time, on a power budget the battery can actually carry. The tell of the year: at GTC 2026 the chip company published an open reference humanoid — a Unitree H2 Plus chassis, Sharpa tactile hands, Jetson Thor compute — and handed it to university labs. When your silicon vendor ships the reference body, the body is no longer the moat.
Listed names mapped to this stageNVIDIA (NASDAQ NVDA), Alphabet (NASDAQ GOOGL), Qualcomm (NASDAQ QCOM), Samsung Electronics (KRX 005930), Klobot (KOSDAQ 466100)
6 Power: Batteries Built for Robots
A robot pack is not a shrunken car pack. It has to push high current out of a tight, oddly shaped volume, and it has to survive thousands of shallow cycles a week. Then comes the uncomfortable part: it does all of that while standing next to a person. That combination is why the "dream battery" conversation moved from cars to robots faster than anyone expected, and why this stage is the thinnest in the chain: two Korean cell makers, a handful of named programs, and roadmaps pointing at 2027–28.
Listed names mapped to this stageSamsung SDI (KRX 006400), LG Energy Solution (KRX 373220)
7 Robot Builders: Humanoids, Cobots & Industrial Arms
Two very different businesses share this stage. Industrial arm and cobot makers have decades of real revenue and installed base. Humanoid makers have order books, and the gap between a committed order and a shipped robot is where this industry has historically gone to disappoint people. Omdia put 2025 global humanoid shipments at 13,318 units — up nearly 480%, and still a rounding error against 4 million industrial robots already installed worldwide. The other structural change: ABB agreed to sell its robotics division to SoftBank for $5.375bn, expected to close in mid-to-late 2026, taking one of the big four industrial robot makers off the public market entirely.
Listed names mapped to this stageTesla (NASDAQ TSLA), Hyundai Motor (KRX 005380), UBTech Robotics (HKEX 9880), Rainbow Robotics (KOSDAQ 277810), Doosan Robotics (KOSPI 454910), FANUC (TSE 6954), Yaskawa Electric (TSE 6506), LG Electronics (KRX 066570)
8 Deployment & End Markets
Deployment is where announcements either graduate or quietly die. Amazon passed one million warehouse robots and then pulled its multi-armed Blue Jay system within months of unveiling it — a useful reminder that the company with the most robot experience on earth still kills programs on cost. Meanwhile the humanoid pilots that matter are unglamorous and specific: Figure robots at BMW Spartanburg, then Leipzig; Agility's Digit at Schaeffler, GXO and Toyota Motor Manufacturing Canada; Atlas sequencing parts at Hyundai's Georgia Metaplant. The number to track isn't robots announced. It's robots doing paid work for someone other than their maker.
Listed names mapped to this stageAmazon (NASDAQ AMZN), Symbotic (NASDAQ SYM), BMW (XETRA BMW), Toyota (TSE 7203), Intuitive Surgical (NASDAQ ISRG)
What to watch
Follow-on orders, not first orders. A $1M humanoid booking at a sensor maker or a 320% robotics growth rate at an actuator supplier tells you a design was won. The signal that matters next is the second purchase order at production volume — component suppliers report these long before robot makers report deployments.
Whether the reducer duopoly holds. Chinese harmonic reducers priced 30–40% below Japanese equivalents are the single biggest lever on a $20,000 robot target. Watch share disclosures and qualification wins, not capacity announcements.
Two listings that set the price of everything. Unitree's post-IPO trading gives the market its first pure-play humanoid comparable, and Agility's SPAC close would create the first US-listed one. Until then every valuation in this sector is triangulated from private rounds.
Policy, not just physics. The January 1, 2027 US defense sourcing rules on Chinese-origin NdFeB, and China's export controls in the other direction, now shape the magnet and motor supply chain more than any engineering roadmap does.
Paid third-party work. Internal fleets are R&D. The threshold this industry has not yet crossed at scale is thousands of robots doing productive work for customers who aren't the manufacturer — and it's the cleanest test of every 2027 target on this page.
Where the Revenue Actually Sits Today
Read the eight stages back to back and one pattern does most of the explaining: the money in this chain currently moves upstream, not at the robot. Magnet makers, reducer makers, actuator suppliers and the compute vendor get paid when a purchase order is signed. Robot builders get paid when a robot completes a shift someone is willing to pay for — and that second event is still rare enough to make headlines when it happens.
That's why the sequence of disclosures usually runs in this order: a component supplier books a small design-win order, then a bigger one, then a robot maker announces a deployment, then a customer expands it. The BMW expansion from Spartanburg to Leipzig matters more than any unveiling, because the customer initiated it. Blue Jay matters for the same reason in reverse — Amazon has more robot experience than anyone alive and still cancelled a flagship system on cost.
The obvious counterweight: for almost every supplier on this page, humanoid robotics is a small slice of total revenue. Harmonic Drive sells into semiconductor equipment. Sanhua sells into HVAC. Vishay Precision Group sells strain gauges into steel mills and crash-test rigs. Exposure to this theme is real but diluted, and an order book can be cancelled as easily as it was announced. Both facts are worth holding at once.
FAQ
What does the humanoid robot supply chain actually look like?
In order: rare-earth magnets and critical materials, precision reducers and gearing, actuators and integrated joint modules, sensing and perception, on-board compute and robot foundation models, batteries, the robot builders themselves, and finally the factories, warehouses and hospitals that deploy them. Value is concentrated at the two ends — magnets and compute — with reducers and actuators the most contested middle.
Which listed companies make humanoid robot parts today?
On reducers and gearing: Harmonic Drive Systems, Nabtesco, Leaderdrive, SPG and Nidec. On actuators and motion: Schaeffler, THK, Sanhua, Hyundai Mobis, HL Mando and Robotis. On sensing: Hesai, Ouster, Vishay Precision Group, Novanta and Aeva. On compute: NVIDIA and Qualcomm. On batteries: Samsung SDI and LG Energy Solution.
Why are precision reducers considered the industry bottleneck?
Each humanoid needs roughly 14 to 28 of them, they are among the most expensive components after compute, and only a handful of firms can make them to arc-second precision with a service life measured in tens of thousands of hours. Capacity is being built quickly — but qualification, not capacity, is the gate. A reducer has to be validated inside a specific robot design before it ships in volume.
Can you buy shares in a pure-play humanoid robot company?
Barely. UBTech Robotics has traded in Hong Kong since 2023, and Unitree priced its Shanghai STAR Market IPO in August 2026. Agility Robotics announced a SPAC merger in June 2026 and expects to trade as AGLT once it closes. Figure AI and Apptronik remain private, and Tesla's Optimus and Hyundai's Boston Dynamics are line items inside much larger companies.
Which part of the humanoid robot supply chain is closest to real revenue?
The component stages — precision reducers, actuators and on-board compute — plus the industrial robot makers sitting alongside the humanoid ones. Those businesses are paid on purchase orders and already have decades of non-humanoid demand underneath them. Humanoid builders and their deployment partners sit at the other end: real order books, but revenue that depends on robots completing paid work for third parties, which remains the industry's unproven step.
How many humanoid robots are actually working right now?
Omdia put 2025 global humanoid shipments at 13,318 units, up nearly 480% year on year, and Figure reported over 350 units delivered by late April 2026. Set against roughly four million industrial robots already installed worldwide, humanoids remain a rounding error — which is exactly why component order books are a better near-term indicator than robot headcount.
- NVIDIA newsroom and GTC 2026 announcements
- Boston Dynamics and Hyundai Motor Group newsroom (CES 2026)
- Hesai Group SEC filings and quarterly results
- Vishay Precision Group quarterly results and investor presentations
- Churchill Capital Corp XI SEC filings (Agility Robotics)
- Shanghai Stock Exchange filings and Unitree IPO prospectus
- ABB and SoftBank Group transaction announcements
- McKinsey humanoid supply chain research; IDTechEx and Omdia market data
- Korea Herald, KED Global, The Robot Report, Nikkei and company disclosures