Macroplane › Blog › Unitree IPO: The $608M Humanoid Bet and the Supply Chain Behind It (2026)

Unitree IPO: The $608M Humanoid Bet and the Supply Chain Behind It (2026)

2026-06-08

Unitree's $608M Shanghai STAR IPO ships the cheapest credible humanoid (G1 at $27.3K, $8,976 BOM, 67% gross margin per SemiAnalysis). The supply chain behind it — $300124.SZ Inovance, $6324.T Harmonic Drive, $MP, $NVDA Jetson NX, $INTC RealSense, SKF / $6471.T / $TKR bearings — and where investors actually capture the trade.

More Macroplane supply-chain guides · Baskets · Investment theses · Macro trends

Unitree IPO: The $608M Humanoid Bet and the Supply Chain Behind It (2026)

Unitree Robotics has filed for a Shanghai STAR Market IPO at roughly RMB 4.2B ($608M). The pitch is the G1 humanoid at $27,300 with an estimated bill of materials of $8,976, the cheapest credible humanoid in production. The story behind the pitch is a Shenzhen-Pearl-River-Delta supply chain that iterates hardware in weeks instead of quarters, plus a verticalization push that already owns the motors and most of the gearboxes that Western humanoid OEMs still outsource.

This is the supply chain map. Where Unitree actually sits, the publicly-listed names upstream of it, the layers it cuts out by building in-house, and where investors capture the trade if Unitree, Tesla Optimus, and Figure are even half right about the next five years.

The BOM teardown, margins, and quadruped-to-humanoid cost trajectory below are anchored on the SemiAnalysis piece China's Unitree Will Dominate Global Humanoid Robotics. The supply-chain wiring and ticker mapping is our own work on top of it.

👉 See it live: the Unitree Robotics IPO Thesis on Macroplane tracks the 13 focus tickers below: motors, gearboxes, magnets, sensors, and the rare-earth feedstock layer that scales 10–20× per humanoid unit.

What Unitree actually is

Two products matter for an investor.

The Go2 quadruped, shipping $1,600–$2,800 since 2021. This is what paid for the humanoid R&D. Unitree shipped thousands of units to universities and industrial buyers, validated the motor and gearbox stack at scale, and generated cash before the humanoid race started. The price compression is its own data point: from $45,000 (Laikago, 2018) to $2,800 (Go2, current) is a 16× cost-down in seven years.

The G1 humanoid, originally launched above $50,000, currently listed at $27,300, and reportedly sold under $20,000 in some deals. SemiAnalysis estimates the bill of materials at $8,976, implying a 67% gross margin at list. Tesla Optimus and Figure are still iterating sub-1,000-unit BOMs at numbers Western chains cannot match.

The IPO filing is for Shanghai's STAR Market at roughly $608M raised. A humanoid pure-play priced on growth, into a sector where the named research-platform customers ($NVDA, $AAPL, $META) have collectively bought hundreds of units each.

Two operating numbers under that. Revenue grew 3× year-on-year into the filing. On the quadruped, gross margin walked from 42.36% to 55.49% over the same window the unit price fell roughly in half — a clean view of what verticalization is actually doing inside the income statement. The next checkpoint cited in the filing is roughly 10,000 cumulative G1 units, which is the threshold that flips humanoid from research-platform revenue to industrial-deployment revenue.

Inside the $8,976 bill of materials

SemiAnalysis broke the G1 EDU Advanced (29 DoF research SKU) down to a $8,976 median COGS at a $27,300 list ASP — 67.12% gross margin. The 29 actuators are the entire DoF count: 14 in the arms, 3 in the waist, 12 in the legs. Every joint is one BLDC motor, one planetary gearbox, one dual encoder, one FOC joint driver, plus thin-section and crossed-roller bearings. Nothing exotic. Aerospace it is not.

The shape of the BOM is what matters for the supply-chain trade:

SubsystemRangeShare of COGSWhat dominates
Legs (12 joints)$2,332–$2,634~28%Planetary gearboxes ($816–$924), BLDC motors ($624–$672)
Torso (compute + battery + structure)$2,320–$2,535~27%Nvidia Jetson NX ($700–$770), structural ($685–$730), battery ($450–$480)
Arms (14 joints)$2,266–$2,538~26%Planetary gearboxes ($728–$770), BLDC motors ($504–$560)
Head (perception)$889–$949~10%DJI Livox MID360 lidar ($557–$578), Intel RealSense D435i ($260–$285)
Waist (3 joints)$529–$599~6%Planetary gearboxes ($143–$158)
Final integration & test$168–$192~2%Assembly labor

Three reads that change how you think about the supply chain.

The 29 actuators are roughly half the BOM. Motors plus gearboxes plus drivers plus encoders plus bearings across arms, waist, and legs run to roughly $4,500–$5,200 of the $8,976. This is why the Inovance and the Chinese-actuator supply chain matter more to the trade than any single line item.

Nvidia Jetson NX at $700–$770 is the single largest semiconductor line. Bigger than every motor in the body individually. This is the line that gets pulled into US export-control risk first if Jetson-class SoCs to China get further restricted.

Sensor "verticalization" is not what it looks like on this SKU. The G1 EDU Advanced uses a DJI Livox MID360 lidar and an Intel ($INTC) RealSense D435i depth camera off the shelf — not an in-house Unitree lidar. Plausibly a research-customer concession (universities want the canonical ROS-compatible sensors), but it's worth being honest about: on this specific SKU, two well-known merchant suppliers own the entire head BOM.

The full annotated teardown — including ASP and quantity per component, the share each subsystem takes of the COGS, and the integration-labor lines — is in the SemiAnalysis piece linked above; this section summarizes the parts that change the supply-chain trade.

Where Unitree sits in the supply chain

Unitree is the OEM at the centre. Upstream it touches four layers, and each is a separately investable bucket. Every row below is a real listed name, validated against Macroplane.

1. Rare-earth feedstock and NdFeB magnets

A humanoid carries 40–50 actuators (every joint, every finger, every wrist, every ankle), each wound on sintered NdFeB magnets. That is roughly 10–20× the rare-earth content of an EV per unit. An EV uses 2–3 kg of NdFeB. A humanoid pushes well into double digits.

This is the layer where ex-China supply matters most. Beijing has been tightening rare-earth export rules, and humanoid OEMs in the US and Europe need a non-China magnet path to ship to defense or strategic customers.

The listed names: $MP (Mountain Pass, the only US mine-to-magnet vertical at scale), $USAR (Oklahoma magnet plant commissioning in 2026), $ALOY (heavy-rare-earth metallization in Pittsburgh), $LYC.AX (Lynas, the largest ex-China NdPr producer), $ILU.AX (Iluka, Australian heavy-REE refinery in build), $UUUU (Energy Fuels, uranium and REE co-processing), $TMC (deep-sea polymetallic nodules). JL Mag and Ningbo Yunsheng dominate inside China and are not yet on Macroplane.

2. Servo motors and BLDC actuators

Unitree designs its own motors. That is the single biggest line in its cost advantage. Per the SemiAnalysis BOM teardown, Unitree's BLDC motors cost roughly 30–40% of equivalent Western motors. The Maxon / Faulhaber margin is gone from the joint.

"Designs its own" is not "makes every part itself." The named upstream actuator supplier across the Chinese humanoid OEM cluster (Unitree, UBTech, AGIBot) is $300124.SZ Inovance, China's #1 servo-motor maker at roughly $22B mcap. Inovance is the single most-cited public name in Chinese humanoid supply-chain coverage. Earnings will start breaking out the humanoid contribution from EV/industrial servo motors over the next two prints, and that break-out is the cleanest listed proxy for Unitree volume.

Tuopu Group (601689.SS, not yet on Macroplane) is the rotary-actuator and dexterous-hand assembly supplier. The Western incumbents Unitree underbids: $6594.T Nidec (BLDC and coreless motors) and $MOG.A Moog (aerospace-grade actuators that some Western humanoid teams adapt for joints).

3. Strain-wave reducers (the precision bottleneck)

This is the joint-precision moat. Cycloidal and harmonic gearboxes are what convert a fast, low-torque motor into the slow, high-torque, zero-backlash motion a humanoid finger needs to hold a glass without crushing it.

The global incumbent is $6324.T Harmonic Drive Systems (Japan), the strain-wave reducer that named the category. $6268.T Nabtesco is the cycloidal RV leader for the heavier shoulder and hip joints. $SHA.DE Schaeffler sits in the bearings and planetary-screws layer and has publicly disclosed plans to supply 1,000–2,000 humanoid units to Humanoid AI by 2032. Small number, but it puts a real supplier on the board.

The Chinese challenger is Leader Drive (688017.SS, not yet on Macroplane), which ships strain-wave reducers at roughly a third of Harmonic Drive pricing and is reportedly Tesla Optimus's stated Mexico supplier. Reliability is still developing. Cost is real today.

Unitree builds its own planetary gearboxes in-house for the low-ratio joints, sidestepping this layer where it can. The rest of the humanoid market (including Optimus, Figure, and Apptronik) still buys harmonic and cycloidal stages from Harmonic, Nabtesco, and Schaeffler. If the Chinese gearbox displacement plays out, the squeeze first surfaces in Harmonic Drive's order book commentary.

4. Sensors and perception

Unitree has its own L1 and L2 lidars (used on the Go2 quadruped and the base G1), but the G1 EDU Advanced research SKU does not run them. The BOM bundles a DJI Livox MID360 lidar ($557–$578) and an $INTC Intel RealSense D435i depth camera ($260–$285) — the same canonical research sensors that ROS stacks at NVIDIA, Apple, Meta, and most university robotics labs already speak natively. On the research SKU, the entire head BOM ($889–$949) flows to two merchant suppliers, not in-house Unitree silicon. The "Unitree self-develops everything" narrative is true for motors and gearboxes. It's softer for sensors than it reads.

For Western humanoid OEMs the picture is harder. The merchant lidar play is $HSAI Hesai, the dominant Chinese-listed automotive lidar maker now expanding into robotics. Magnetic position sensors and IMUs for joint feedback are the small but durable picks-and-shovels: $ALGM Allegro (motor current sensors and magnetic position ICs), $MELE.BR Melexis (global #1 in magnetic position sensor ICs), $6762.T TDK (image sensors plus IMU through the InvenSense acquisition). Intel keeps shipping RealSense into the research humanoid layer as long as ROS stacks default to it.

5. Bearings

The BOM breaks bearings out explicitly: thin-section bearings on every joint (28 in the arms alone), crossed-roller bearings on the rotary stages, plus the linkage bar in the legs. Total bearing content runs roughly $420–$500 across the body — small per unit, but it scales linearly with humanoid volume and is exactly the kind of line item the Japanese, Swedish, and US bearings incumbents are levered to.

The merchant names that supply this layer at scale: $SHA.DE Schaeffler (already cited above for planetary screws), SKF (Sweden, mass-market thin-section bearings into industrial and robotic arms), $6471.T NSK (Japan, high-precision motion-control bearings), $TKR Timken (US, ball and roller bearings — sector is literally that). For Chinese OEMs the domestic chains are stepping in, but the precision thin-section and crossed-roller specifications used in humanoid joints are still where Schaeffler, SKF, and NSK win on cycle-life.

6. Edge compute

$NVDA (Jetson edge SoCs and the GR00T vision-language-action foundation models) and $QCOM (RB6 robotics platform) supply the on-board brain. The Jetson NX module alone is $700–$770 in the G1 EDU Advanced BOM — bigger than every individual motor, bigger than the battery. That is the single largest semiconductor line in the robot, and the one most exposed to a future tightening of Jetson-class export licenses to China.

The Nvidia line is also the one that surprises some readers: Nvidia is itself one of Unitree's largest research-platform customers, alongside Apple and Meta. The supplier-customer relationship runs both directions.

The edge: why Unitree is structurally cheaper

Three things, and the order matters.

In-house motors at 30–40% of Western prices. Unitree designs and builds its own BLDC motors and the low-ratio planetary gearboxes. That removes the Harmonic Drive and Maxon margin from the BOM on every joint that does not need strain-wave precision. The cost saving compounds across 40–50 actuators per unit.

Shenzhen-PRD proximity to suppliers. This is the part Western humanoid teams cannot replicate by spending money. Prototype samples arrive next-day. Iteration cycles are weeks, not quarters. Western teams sit 6–18 months behind on every BOM revision because their suppliers are in Germany, Japan, or California, not a 90-minute high-speed-rail ride away. Compounded over five hardware cycles, the gap becomes structural.

Quadruped scale paid for the humanoid R&D. Unitree didn't bet the company on Optimus-style ambition. It built profitable $1,600–$2,800 quadrupeds for four years, scaled the motor and gearbox supply chain, and pivoted that exact stack into the G1. Tesla, Figure, and Apptronik all started at the humanoid, with no cash-generating intermediate product to fund the iteration loop.

The risks

  • IPO timing. A $608M raise into a sector with concentrated research-platform buyers and limited industrial revenue is priced on growth, not earnings. Any slip past the 10,000-unit industrial deployment threshold compresses the multiple immediately.
  • Export controls. Both rare earths (China cutting magnet exports) and edge compute (US restricting Jetson-class SoCs to China) flow upstream into Unitree's BOM. A symmetric trade-war move hurts Western humanoid OEMs more, but Unitree still loses access to $NVDA, $QCOM, and the Western sensor IC layer.
  • $TSLA Optimus and AGIBot Walker. Optimus has Tesla's manufacturing depth: giga-scale lines, automotive supply chain leverage, and the cash to lose. AGIBot has Chinese capital and is already shipping Walker units. Both compress Unitree's margin path even where Unitree wins on cost.
  • Verticalization cuts both ways. Building its own motors and lidars means the structural cost edge is real, but Unitree's dependency on the merchant layer shrinks. The upside for $ALGM, $MELE.BR, and $HSAI from "Unitree adoption" is muted vs. the upside from Optimus, Figure, and Apptronik adoption. Investors who buy the merchant supply chain are betting on the Western humanoid wave, not Unitree itself.

The catalysts to watch

CatalystWhy it matters
Shanghai STAR Market IPO pricing and first-day tradingSets the comp for every public humanoid pure-play, including the KOID humanoid ETF and KSTR STAR Market ETF.
G1 industrial-deployment unit milestonesSub-$30/hour tote-handling cost was the stated viability threshold. Past 10,000 units is the next reset.
$300124.SZ Inovance humanoid revenue splitThe cleanest listed proxy for Unitree volume once the segment break-out lands.
$6324.T Harmonic Drive order book commentaryIf the Chinese gearbox displacement plays out, the squeeze surfaces here first.
$TSLA Optimus production volume guidePulls demand into $MP, $ALGM, $MELE.BR, $HSAI: the merchant-layer beneficiaries who are NOT Unitree suppliers.
Rare-earth export policy updatesThe most likely lever from Beijing; directly drives $MP, $USAR, $ALOY, $LYC.AX.

How to think about it as an investor

The picks-and-shovels of Unitree are not Unitree shares. They are:

  • The rare-earth feedstock layer ($MP, $USAR, $ALOY, $LYC.AX, $ILU.AX, $UUUU, $TMC), which scales 10–20× per humanoid unit and benefits from any humanoid OEM winning, not just Unitree.
  • The listed Chinese servo-motor names ($300124.SZ Inovance), which already supply Unitree by name and will start breaking out humanoid revenue.
  • The Japanese gearbox incumbents under cost pressure ($6324.T Harmonic Drive, $6268.T Nabtesco). Long if the precision bottleneck holds, short pressure if Leader Drive displaces them.
  • The Western sensor IC vendors ($ALGM, $MELE.BR, $6762.T TDK, $HSAI). The merchant layer that wins on every humanoid that is not vertically integrated. Better levered to $TSLA Optimus, Figure, and Apptronik than to Unitree.
  • The edge-compute layer ($NVDA, $QCOM). Already in the position; humanoid is incremental, not the thesis.

Unitree itself is a STAR Market IPO that most US-domiciled investors will not be able to buy directly. The humanoid ETFs (KraneShares KOID, the STAR-market ETF KSTR) are the indirect route; the focused thesis above is the direct one.

FAQ

When is the Unitree IPO? Unitree filed for a Shanghai STAR Market IPO at roughly RMB 4.2B / $608M in March 2026. The listing date depends on STAR Market review and is expected in H2 2026.

Can US investors buy Unitree stock? Not directly. STAR Market shares are not available on US brokers without an A-share routing arrangement. Indirect exposure: KraneShares Humanoid Robotics ETF (KOID) and the KraneShares STAR Market 50 ETF (KSTR), both of which expect to include Unitree on listing.

What is the Unitree G1 bill of materials? SemiAnalysis estimates roughly $8,976 of parts in a G1 humanoid at the current $27,300 list price, implying a 67% gross margin.

Which public companies are in Unitree's supply chain? $300124.SZ Inovance (servo motors), $6324.T Harmonic Drive (strain-wave reducers, though Unitree builds its own low-ratio gearboxes), $NVDA (Jetson NX at $700–$770 per G1, the largest semiconductor line), $QCOM (RB6 robotics platform), $INTC Intel (RealSense D435i depth camera on the G1 EDU Advanced), $HSAI Hesai (merchant lidar peer; G1 EDU Advanced uses DJI Livox MID360, which is private), $ALGM Allegro, $MELE.BR Melexis, $6762.T TDK (sensors and IMUs), $SHA.DE Schaeffler, SKF, $6471.T NSK, $TKR Timken (bearings — thin-section + crossed-roller on every joint). The rare-earth magnet layer feeds every actuator and is anchored by $MP, $USAR, $ALOY, $LYC.AX.

Does Unitree compete with Tesla Optimus? Yes, on price and on Chinese vs. US humanoid leadership. Unitree's G1 ships at ~$27,300 today; Tesla has targeted Optimus at $20,000–$30,000 at production volume but is still pre-volume in 2026.

Why does Unitree have a structural cost edge? In-house BLDC motors and low-ratio planetary gearboxes (30–40% cheaper than Western equivalents), Shenzhen-PRD supplier proximity (weeks not quarters per iteration), and a profitable Go2 quadruped business that funded the humanoid R&D.

This is not investment advice. Do your own research and confirm the supply-chain wiring on Macroplane before sizing a position.

Related reading

  • Rare Earth Stocks: The Complete Investor Guide (2026)
  • The AI Supply Chain Map: Every Public Company in the AI Buildout (2026)
  • Unitree Robotics IPO Thesis on Macroplane
  • Robotics & Industrial Automation trend

Source

  • SemiAnalysis, China's Unitree Will Dominate Global Humanoid Robotics — primary source for the G1 EDU Advanced bill of materials, the 67.12% gross-margin estimate, and the quadruped margin trajectory. The BOM teardown and per-subsystem dollar ranges in this post come from their analysis.

Referenced on this page

  • Unitree Robotics IPO Thesis
  • SKF
  • Rare Earth Stocks: The Complete Investor Guide (2026)
  • The AI Supply Chain Map: Every Public Company in the AI Buildout (2026)
  • Robotics & Industrial Automation trend