News · Research
Humanoid actuators and gearboxes — the supply-chain bottleneck
Source: Figure AI
Lead strip: Status: capacity race / lead times real · Availability: OEM scale ≠ reducer nameplate · All-in cues: BOM share high; “best supplier” trophies refused · Scoring: methodology
Supply-chain brief — not a sold ranking and not a trophy “best actuator supplier.” Designed reducer capacity ≠ robots shipped. Ambition unit counts ≠ qualified BOM.
Humanoid roadmaps quote thousands to millions of robots. Each biped needs dozens of precision actuators — rotary joints with strain-wave (harmonic) reducers, linear joints with roller screws or similar, plus hand motors. That multiplies demand for parts that historically shipped in industrial-arm quantities. This brief maps the bottleneck with public OEM and supplier cues — then states what buyers and investors can actually verify. Cross-read Unitree G1, Figure 03, Optimus.
Robot strips (context, not a trophy board)
- Unitree G1 · Unitree Robotics · Status: Available (research / developer) · Use: lab / EDU · Evidence: public list price; China reducer ecosystem proximity · → /robots/unitree-g1/
- Figure 03 · Figure AI · Status: Pilot (BMW Spartanburg) · Use: factory logistics · Evidence: BotQ production claims; actuator BOM — public · → /robots/figure-03/
- Optimus · Tesla · Status: Research / production-ramp · Use: factory R&D / Academy · Evidence: IR actuator architecture talk; external buy — · → /robots/optimus/
Why actuators dominate the scale problem
A conventional six-axis industrial arm uses a handful of reducers. A humanoid torso-plus-legs design commonly needs on the order of ~15–40 precision rotary stages (plus linear actuators and hand DoF), depending on generation. Tesla-oriented teardowns and secondary BOM notes repeatedly cite ~14 harmonic reducers and ~14 planetary roller screws in body architectures shown in Tesla materials — treat counts as order-of-magnitude / revision-sensitive, not a frozen production BOM (Optimus context).
Consequences:
- Volume math — 10,000 humanoids/year × ~20 harmonic reducers ≈ 200,000 precision reducers/year from one OEM program alone.
- SKU math — multiple torque classes (e.g. Tesla-shown 20 / 110 / 180 Nm rotary tiers in secondary analyses) multiply qualification SKUs.
- Cost math — actuators + reducers + screws often dominate major-component value; long-term sub-$20k–$30k humanoid BOM targets collide with today’s precision-part pricing unless suppliers scale and yield improve.
- Lead-time math — hollow-shaft / ultra-flat harmonic series and high-force roller screws have been reported with extended procurement times as robotics demand rises.
Buyer cue: when an OEM claims “we will ship N thousand next year,” ask whether qualified actuator capacity (not press-release factory floor area) matches N × joints per robot.
Supplier capacity signals (label as claims)
| Actor | Public capacity / role signal (2025–2026 coverage) | How to read it |
|---|---|---|
| Harmonic Drive Systems (Japan) | Expands Japanese production (Azumino/Hotaka-area tooling) for ultra-flat / hollow-shaft series; commercial ramp framed through late 2026 in trade coverage | Premium incumbent adding capacity — not infinite supply |
| Leaderdrive / Green Harmonic (China) | Humanoid-oriented harmonic share and multi-hundred-thousand to ~1M+ unit capacity ambitions in Chinese market research / IPO coverage; Unitree / AgiBot / Tesla-certification narratives in secondary reports | High-volume China lane — verify OEM qualification, not headlines alone |
| Shuanghuan and auto-gear entrants | RV/harmonic expansion and large annual reducer capacity plans in 2026 plant narratives | Automotive suppliers entering robotics tolerances |
| Roller-screw specialists | Planetary roller screws highlighted for Optimus-class linear joints; historically aerospace/defense volumes | Niche precision → needs civilian scale |
We do not rank these firms as “best humanoid supplier.” Capacity plans are supplier intent; ship-to-OEM yield and lifetime are separate evidence.
Sources for the table are industry/trade summaries (e.g. Harmonic Drive expansion coverage; Leaderdrive / Green Harmonic deep-dives; China supply-chain IPO features). Where a number is “annualized capacity by end-202X,” label it design/ambition, not 2026 shipped fleet.
OEM lenses
Optimus — architecture public, catalog buy not
Tesla has publicly shown rotary + linear actuator families for Optimus and frames production-ramp / Academy use inside Tesla. Secondary analyses emphasize Chinese content share for Gen-3 components and the tension between actuator BOM cost and long-term price targets. External purchase page and stable list price remain —. Page: Optimus.
Figure 03 — BotQ scale vs unpublished joint BOM
Figure’s BotQ production posts claim hundreds of Figure 03 units built and rising cadence (Ramping Figure 03 production); BMW states a Spartanburg sequencing project for Figure 03. Public part-level actuator supplier map: —. Hardware scale is a real signal; it does not disclose who machines the flexsplines. Page: Figure 03.
Unitree G1 — shoppable biped, ecosystem proximity
Unitree G1 publishes a list path from about US$13,500 (ex tax/ship). Chinese trade coverage links Unitree upper-limb harmonic procurement to Leaderdrive-class suppliers at high share in some periods. That is ecosystem evidence, not a Robotscomparison endorsement of any reducer brand — and not warehouse fleet proof.
What is not the bottleneck (usually)
- GPUs for training matter for AI labs, but multi-joint mechanical yield often gates physical fleet scale first.
- Carbon fiber covers and cosmetic shells rarely set the critical path vs precision gears and screws.
- Valuation headlines do not machine flexsplines.
Buyer / investor checklist
- Joints per robot (body + hands) × units/year = reducer/screw demand.
- Dual-source? Single-source harmonic or roller-screw risk.
- Qualified vs nameplate — ISO/TS process capability at robotics tolerances.
- Torque-class SKUs — three reducer sizes ≠ one purchase order.
- Aftermarket — field replacement lead time for a failed hip actuator.
- Refuse trophies — no “#1 actuator supplier” board on this site.
Practical takeaway
The humanoid scale story runs through precision actuation supply, not only AI demos. Japanese incumbents are adding capacity; Chinese harmonic specialists and auto-gear entrants publish aggressive unit plans; Optimus-class designs spotlight both strain-wave reducers and planetary roller screws. Figure’s BotQ and Unitree’s list-price G1 prove different commercial surfaces — neither publishes a full public BOM. Score capacity claims as supplier ambition until linked to qualified OEM deliveries. Unknowns: —.
FAQ
Are harmonic reducers the only bottleneck?
No. Planetary roller screws, frameless torque motors, bearings, torque sensors, and miniature hand actuators also constrain designs. Harmonics are the most-discussed rotary bottleneck.
Does supplier capacity expansion mean OEMs can hit million-unit roadmaps?
Not automatically. Expansion must clear qualification, yield, and lifetime at humanoid duty cycles. Million-unit OEM narratives remain ambition until matched by multi-year ship data.
Who is the “best” actuator supplier for humanoids?
We do not publish that trophy. Ask for lead time, dual-source options, and failure data on your duty cycle instead.
How should Digit / Figure buyers use this?
Enterprise buyers should add spare-actuator SLA and lead times to RaaS/CapEx diligence — beside software and safety. See also RaaS pricing.
Sources
- Trade / industry: Harmonic Drive capacity expansion coverage; Leaderdrive / Green Harmonic and China reducer IPO features; Optimus actuator architecture secondary analyses
- Figure AI: Ramping Figure 03 production
- Unitree: G1
- Robotscomparison: Unitree G1 · Figure 03 · Optimus · methodology
Exact production BOMs, dual-source contracts, and verified 2026 shipped reducer counts to each OEM: —.