Joint actuators account for >30% of humanoid robot BOM (Interact Analysis 2025); reducer/transmission sub-assembly is estimated ~30-40% of actuator cost, implying reducers alone ~10-15% of total humanoid BOM and the single largest cost item within each actuator
High: estimated 8-15% of humanoid robot BOM at current prices (~14 PRS units × $200-300 each per Optimus unit); expected to compress with scale but remains top-3 BOM line item alongside motors and reducers
High-precision 7-axis internal-thread CNC grinding machines (内螺纹磨床) are the critical choke…
~12% of robot BOM at $20K target cost (36KR); frameless torque alone ~16% of component value (TechBuzzChina); coreless adds ~8%; combined motor module is 2nd costliest category after lead screws; reference classification: med
Ultra-precision winding technology (sub-0.05mm copper wire) for coreless cup motors limits…
med (~10–15% of humanoid robot BOM; typically 4 six-axis FT units per robot at $200–2,000 each, 28+ encoder channels integrated in actuators, and 2–4 IMUs; FT sensors are the highest unit-cost sub-segment; sensors overall estimated at ~10% of ~$50K unit-scale BOM per humanoid.guide)
No scalable MEMS 6-DOF force-torque sensor IC exists in production volume as of Q1 2026—fo…
~31% of humanoid robot BOM (high-DOF configurations; broker consensus 25–31%; ~17% in Tesla Optimus Gen 2 per Morgan Stanley; largest single module in humanoid BOM, driven by coreless motors ~50%, tactile sensors ~30%, micro-transmission ~12% of hand sub-BOM)
Miniaturized coreless (hollow-cup) motor supply constrained by European import dominance (…
med (perception sensors estimated at 15–25% of humanoid robot BOM; second-largest cost center after actuators; LiDAR units alone at current ASPs represent the single largest per-unit line item in the perception stack)
~8% of humanoid robot BOM (2025) as silicon/compute line item; part of broader 10-20% perception+compute category; expected to decline to ~5% by 2035 as scale drives unit cost down (BofA/RoboZaps estimates)
TSMC advanced-node (3nm/2nm) wafer capacity and CoWoS advanced packaging sold out through …
med; ~3–8% of humanoid unit BOM (pack $300–1,100 on a $30–50k robot); actuators/joints dominate at >30%; battery is a mid-tier cost contributor today but share rises if solid-state cells command premium pricing
Energy density constrained by weight budget: humanoids can allocate only ~1/8 of total mas…
med (~5-8% est. of humanoid robot total BOM; NdFeB magnets are core to servo motors which account for ~20% of robot BOM, and magnets represent a meaningful fraction of each motor's cost)
China controls ~90% of global sintered NdFeB production. April 2025 MOFCOM export controls…
② 形态 BOM 拆解 — 各类机器人的成本构成与典型供应商
人形（双足） conf: med
BOM 总成本 $30,000 – $80,000
Range spans China-sourced full-featured humanoid (~$30K at 5,000+ units/yr, Unitree G1-class) to Western OEM industrial humanoid (~$50K–$80K at 500–2,000 units/yr, Tesla Optimus / Figure 03 / Apptronik Apollo class). BofA (Apr 2025) pegs a mostly-China-sourced BOM at ~$35K at end-2025 scale; early low-volume Western production runs $80K–$120K. Long-term target at 100K+ units/yr is $13K–$20K per BofA/DIGITIMES projections.
22+ DoF hands require 16–20 hollow-cup micro-motors (6–8 mm dia). Swiss Maxon and German Faulhaber dominate micro-motor supply at $50–$80/unit imported; Chinese entrant Aoyi Technology offers alternatives at ~¥100/unit. High-DoF designs push hand share toward 25–30% of BOM. Tesla Optimus Gen 3 targets 50 hand actuators total. Inspire Robots shipped 10K hands in 2025, targeting 50–100K in 2026.
Stack: multi-camera RGBD arrays, 6-axis IMUs, joint-level force/torque sensors, tactile skin patches, microphones. Most industrial humanoids use vision-only (no LiDAR) to reduce cost. Force/torque sensors at wrists and ankles add ~$500–$1,500/unit. Domestic tactile sensors in China at ¥199 vs. $1,000+ for Tekscan/SynTouch imports.
计算与控制电子
8%
NVIDIA (US, Jetson Orin/Thor) · Texas Instruments (US, motor drivers) · Infineon (DE, power ICs) · TSMC (TW, chip fab) · Tesla (US, custom FSD chip for Optimus)
AI inference SoC (NVIDIA Jetson AGX Orin at 275 TOPS dominant; Jetson Thor at 2,070 TOPS emerging) + motor controller ICs + embedded safety MCUs. NVIDIA holds >80% compute market share in humanoid platforms. Tesla Optimus uses in-house FSD chip, leveraging automotive supply chain. 48V architecture convergence with automotive is reducing power electronics costs 15–20%.
Torso and legs use aerospace-grade aluminum alloy (6061/7075); arms and covers use CFRP for weight reduction; titanium at high-stress joints. Fabrication and CNC machining are predominantly China-based OEMs. Frame cost is relatively low but material selection heavily influences total robot mass (target ~50–70 kg) and payload capacity.
2–4 kWh 48V Li-ion pack; target ~4 hr continuous operation. Hot-swappable packs (Apptronik Apollo design) enable near-continuous factory uptime. CATL supplies 4680 cells to Tesla for Optimus. EV-derived battery tech is the key cost reduction lever; humanoid packs benefit from EV scale economies. BMS and DC-DC power distribution add ~$500–$800.
High-flex cable harnesses, miniature waterproof connectors, power rails, signal integrity shielding. Labor-intensive integration step; predominantly assembled in China. Weight and routing constraints in humanoid form factor make this disproportionately expensive vs. industrial arm robots. Accounts for ~$1,500–$4,000 in materials plus significant assembly labor.
供应商格局：The humanoid bipedal supply chain is structurally bifurcated: Chinese OEMs (Unitree, Fourier, UBTECH) achieve $16K–$30K BOM by vertically integrating domestic reducers, motors, and battery cells, while Western OEMs (Tesla, Figure, Apptronik) pay a 30–50% premium for Japanese/European precision components, putting their BOMs at $50K–$100K at current volumes. Japan holds a strategic chokehold on precision harmonic reducers — Harmonic Drive Systems commands ~58% global market share — though China's Leaderdrive and Laifu are scaling rapidly and have passed 60% domestic substitution rate. The single most acute supply-chain risk as of mid-2026 is China's April 2025 export licensing regime on heavy rare earth elements (Dy/Tb) used in NdFeB magnets: US-bound magnet shipments fell 93% in May 2025, directly threatening actuator production for all non-China humanoid manufacturers, including Tesla Optimus which sourced exclusively from Beijing Zhongke Sanhuan.
Wide spectrum by tier. Consumer/research CN (Unitree Go2 Air/Pro, 1,000+ unit volume): ~$800–1,500 BOM; Go2 EDU+ research class: ~$2,500–4,500. Industrial CN heavy-duty (Unitree B2 / DEEP Robotics X20-class, 100+ units/yr): ~$8,000–18,000. Western premium low-volume (Boston Dynamics Spot at ~75K MSRP, ~10–15× CN cost): ~$20,000–35,000 BOM. Unitree's IPO disclosures confirm gross margins of 44–60%, implying BOM ~40–56% of MSRP. Subsystem percentages below are calibrated to a mid-tier industrial CN quadruped (~$10,000–16,000 BOM, B2/X20 class). No hands or dexterous end-effector—actuator share is lower than humanoid.
子系统
%
典型供应商
说明
执行器（关节电机 + 减速器 + 驱动 PCB）
42%
Unitree Robotics (in-house, CN) · T-Motor / TMOTOR (CN) · RobStride (Shenzhen, CN) · Maxon Group (CH, used by Boston Dynamics) · Dongguan Getech Motor (CN)
12 identical Quasi-Direct-Drive BLDC outrunner motors (3 per leg), each with planetary reducer (~6:1), magnetic encoder, and FOC driver PCB. Go2 retail replacement motor (GO-M8010-6) lists at $369/unit; OEM/volume cost likely $80–200 CN, $300–600 Western. No dexterous hands means fewer total actuators vs humanoid, but actuators still dominate BOM. Unitree self-manufactures motors, reducers, and encoders; purchased parts = only 14–18% of COGS.
LiDAR is the single most expensive individual part: CN domestic units (Robosense, Hesai, LIVOX, Unitree proprietary) cost ~$200–600 each; Western equivalents (Velodyne/Ouster) run ~$1,500–3,000. IMU: $50–150. Depth/RGB cameras: 2–4 units × $50–150 each. Foot contact sensors: $80–200/set. CN platforms omit side cameras to cut cost. Hesai JT-series hit 100K cumulative deliveries within 5 months, indicating fast commoditization.
计算模块（运动控制 MCU + AI/自主 SoC）
12%
NVIDIA (US, Jetson Orin Nano / NX for EDU and research tiers) · Rockchip (CN, RK3588 SoC in base tiers) · Texas Instruments (US, real-time motor control DSP) · Unitree (proprietary ARM-based SoC in Air/Pro variants)
Dual-layer compute: a real-time motion-control MCU (~$50–200) plus an optional AI inference SoC. Go2 Air/Pro use a proprietary ARM SoC; Go2 EDU adds Jetson Orin Nano 8 GB (40 TOPS); Go2 EDU+ uses Jetson Orin NX 16 GB (100 TOPS, module ~$500–700). Industrial CN heavy class adds up to 3 Jetson Orins (~50 W each). Compute is 10–15% of BOM; grows with autonomy requirements.
电池与电源系统
10%
CATL (CN, cylindrical 18650 / pouch cells) · BYD (CN) · Molicel (CA, used in Spot Enterprise pack) · Samsung SDI (KR, used in Spot Explorer pack) · Unitree (CN, proprietary BMS PCB)
Go2 uses an 18650 lithium-ion pack (28.8 V, 8,000 mAh standard / 15,000 mAh ultra); Spot uses Molicel cells in Enterprise variant. BOM cost: ~$100–300 for Go2-class packs; $400–1,200 for heavier industrial packs (B2 / Spot class). Includes BMS PCB and charger. Unitree battery life is ~2× Western rivals at equivalent payload, partly from weight savings.
结构件与机械（机身框架、腿连杆、关节外壳）
11%
In-house CNC + die-casting (CN contract manufacturers, e.g., Shenzhen precision shops) · Foxconn Industrial (CN, aluminum die-cast) · TRUMPF / Bystronic (DE/CH, laser cutting used by Western OEMs) · Kennametal (US, tooling for machined Al alloy parts)
Primary material is 6061 aluminum alloy (CNC-machined + die-cast for high-volume CN; machined billet for low-volume Western). Body shell, hip/shoulder brackets, and leg link tubes. CN contract manufacturing brings frame cost to ~$200–600 for Go2-class; $1,000–3,000 for B2/Spot-class (heavier duty, IP67 sealing, larger cross sections). Carbon fiber composites used in premium Western legs. IP66/67 rating adds sealing cost.
电子与线束（主 PCB、功率分配板、连接器、线束）
7%
JLCPCB / PCBWay (CN, PCB fabrication) · Molex / Amphenol (US, rugged connectors) · TE Connectivity (US-CH, connectors) · STMicroelectronics (FR-IT, power management ICs) · TI / Infineon (US/DE, gate drivers, power ICs)
Includes main system PCBs, motor power distribution board, multi-stage voltage regulation (motor bus 24–48V, logic 5V/3.3V), and wiring harness. Vibration and shock tolerance is critical (flexible PCB segments used at joints). Cost: $100–400 for consumer CN class; $500–1,500 for industrial heavy-duty class. JLCPCB and PCBWay dominate CN low-volume fabrication.
供应商格局：Chinese manufacturers — led by Unitree Robotics (est. 70% of global unit shipments in 2023, filing for a ~$610M Shanghai IPO in 2026) — dominate quadruped production through extreme vertical integration: Unitree self-manufactures motors, harmonic/planetary reducers, encoders, controllers, and LiDAR modules, leaving only 14–18% of COGS as externally purchased parts, enabling gross margins of 44–60% even at retail prices well below Western competitors. The primary cost asymmetry versus Boston Dynamics and ANYbotics is in the actuator layer (QDD motors with CN-made planetary gearboxes vs. Maxon-based Western designs) and LiDAR (CN domestic at $200–600 vs. Velodyne/Ouster at $1,500–3,000); the largest single-source risk is the actuator itself, though multiple CN suppliers (T-Motor, RobStride, Getech) now offer drop-in QDD alternatives. LiDAR is rapidly commoditizing within China — Hesai's JT robot-specific series shipped 100,000 units in under five months through H1 2025 — further compressing sensor costs for CN OEMs and widening the gap with Western platforms that must comply with non-Chinese BOM requirements (e.g., Boston Dynamics).
代表产品: Unitree Go2 Pro (Unitree Robotics) · Unitree B2 (Unitree Robotics) · Boston Dynamics Spot (Boston Dynamics) · DEEP Robotics Jueying X20 (DEEP Robotics) · ANYmal D (ANYbotics)
轮式人形 / 移动操作 conf: med
BOM 总成本 $22,000 – $50,000
At 1,000–10,000 unit/year volume with China-dominant supply chain. Low end (~$22K): basic parallel gripper, Rockchip compute, Chinese components throughout, e.g. Unitree-class build. High end (~$50K): dexterous 5-finger hands, NVIDIA Jetson Thor, international sensors, low-volume production (100–500 units/year). Saves ~$5K–10K vs full bipedal (no 10–12 leg joint modules) partially offset by wheeled-base addition (~$2K–6K). International-component configurations can reach $60K–80K.
子系统
%
典型供应商
说明
臂部与躯干关节执行器
38%
CubeMars (CN) · Leaderdrive / Green Harmonic (CN) · Harmonic Drive Systems (JP) · Nabtesco (JP) · Dongli Electric (CN)
Dominant cost driver: ~20–30 integrated joint modules (BLDC motor + strain-wave reducer + encoder) for dual-arm plus torso. Unit cost $200–$600 depending on torque class; harmonic reducers remain the supply-chain bottleneck. Wheeled form skips ~10–12 leg joints, saving ~15% vs full bipedal.
Dexterous 5-finger pair $6K–$15K (Inspire RH56, Shadow); parallel gripper pair $500–$2K (Robotiq). Choice of end effector is the largest single BOM swing factor after actuators.
Form-factor-defining subsystem: differential-drive or Mecanum omnidirectional base with drive motors, motor controllers, wheel encoders, chassis. Cost $2K–$6K; well-commoditized vs bipedal legs — lower single-source risk. Navigation-grade SLAM integration adds SW cost.
Richer sensor suite needed than pure bipedal: 3D LiDAR for navigation (Livox MID360 ~$530 USD), 2–3 depth cameras (~$260 ea), wrist F/T sensors, IMU. Full stack $1.5K–$8K. Chinese LiDAR dominates cost-down path.
Budget builds: Rockchip RK3588 ~$200; mid-tier: Jetson Orin NX $500–$700; high-end: Jetson Thor (Richtech Dex, Figure AI class). Full compute stack including motion controllers $800–$3K. Silicon share shrinking as volume rises.
电池与电源电子
7%
CATL (CN) · BYD (CN) · BAK Battery (CN) · Sony Energy Devices (JP)
300–600 Wh LiFePO4/Li-ion pack + BMS + power distribution board; $700–$2,500. Hot-swap capability (e.g. Apollo) raises cost but enables continuous operation. Chinese cell makers control ~80% of robot battery supply.
结构与机械底盘
8%
Foxconn / Jabil (CN/US) — EMS assembly · Local CNC shops (CN) · Die-cast foundries (CN/JP) · Minth Group (CN) — die castings
Aluminum extrusions, die-cast joint housings, ABS/PA covers, precision bearings. Still CNC-heavy at <10K units, inflating cost; die-casting tooling payoff only at >50K units. 5–10% of BOM.
线束、连接器与集成
4%
Molex (US) · JST (JP) · TE Connectivity (US/CH) · Local harness shops (CN)
Flex PCBs, cable harnesses, connectors; labor-intensive at small scale. Typically 4–6% of BOM. Integration labor cost is an additional OPEX item not captured in component BOM.
供应商格局：Chinese suppliers control an estimated 63–70% of the wheeled humanoid component supply chain by value, with dominant positions in joint actuator modules (CubeMars, Leaderdrive), 3D LiDAR (Livox/DJI, RoboSense), battery cells (CATL, BYD), and structural manufacturing, at 30–40% cost advantages over Western equivalents. Japanese incumbents Harmonic Drive Systems and Nabtesco retain a premium position in high-precision strain-wave reducers — a bottleneck shared with bipedal humanoids — but are losing share as Chinese alternatives (Leaderdrive, Shuanghuan) reach acceptable quality at volume. The wheeled base subsystem is the lowest single-source risk area in this form factor: AMR-class drive platforms are commoditized across Chinese OEM motor suppliers and Western integrators (Clearpath, Omron), contrasting with the actuator and harmonic reducer stack where supply constraints could become acute as annual volumes cross 50,000 units in the 2027–2028 window.
Arm-only BOM (excludes end-effector tooling, integration labor, and software licensing) for a standard 6-axis, 5–10 kg payload cobot. Low end = China-assembled with domestic harmonic reducers at 500+ units/yr production volume (e.g., JAKA/AUBO/Fairino class). High end = Western/Japanese component mix (HDSI reducers, Maxon motors) at <100 units/yr or for a premium brand like UR5e/Doosan M1013. System selling price is typically 2–4× BOM cost.
子系统
%
典型供应商
说明
关节减速器（谐波/应变波）
27%
Harmonic Drive Systems Inc. (US/JP) · Nidec-Shimpo (JP) · Leaderdrive (CN) · Zhejiang Laifual (CN) · PICEA Motion (CN)
Cobots use harmonic strain-wave reducers at all 6 joints (unlike heavy industrial arms that use bulky RV reducers at the base). The single largest cost driver; HDSI/Nidec-Shimpo price 2–3× above Chinese alternatives (Leaderdrive, Laifual), which captured >35% of CN OEM purchases by 2024. Average market price ~$234/unit globally in 2024.
Six frameless BLDC motors with integrated servo drivers and multi-turn encoders; increasingly bundled as compact joint modules. CN suppliers have largely closed the performance gap at 40–60% lower cost. Servo systems collectively represent 20–28% of cobot BOM per industry analysis; critical for impedance bandwidth and compliant motion.
Real-time motion controller, ISO 13849 PL-e / IEC 61508 SIL2-certified safety CPU, teach pendant display and MCU. Safety certification adds meaningful BOM premium over non-safety-rated compute. ARM-based consolidation is reducing PCB count; includes embedded fieldbus (EtherCAT) interfaces.
力矩传感 / 力觉感知（关节级）
12%
TE Connectivity (intl) · ATI Industrial Automation (US) · Bota Systems (CH) · FUTEK (US) · SRI International (US)
Per-joint strain-gauge or Hall-effect torque sensors (6 joints) enabling compliant control, soft collision detection to <5 N, and ISO 10218-1 collaborative safety compliance. Adds BOM cost and integration complexity; some entry-level Chinese cobots substitute model-based current-feedback sensing at near-zero component cost, trading off sensitivity. Doosan and UR cobots both carry full joint torque sensing.
Six die-cast or CNC-machined aluminum arm segments, cross-roller or angular-contact bearings at each joint, internal cable harness (power + data), and connectors. Commodity subsystem in unit cost but precision machining and anodizing tolerances still favor specialized CN casting shops; igus chainflex used widely for internal cable routing.
ISO 9409-1 standard tool flange; electrical and pneumatic pass-throughs. Many cobots ship bare flange only; tool changers are sold separately. Included here at base flange cost only. EOAT (grippers, welding, etc.) is excluded from BOM scope.
供应商格局：The cobot supply chain's dominant bottleneck is the harmonic strain-wave reducer: Harmonic Drive Systems Inc. (HDSI) holds foundational precision patents and sets global pricing, while Nidec-Shimpo (JP) and two leading Chinese entrants — Leaderdrive and Zhejiang Laifual — address the mid-tier; by 2024 Chinese domestic reducers captured over 35% of unit volume in Chinese-assembled cobots at 40–60% of HDSI pricing, enabling Chinese OEMs (JAKA, AUBO, Fairino) to undercut Universal Robots by 40–60% on comparable payload/reach specs, contributing to UR's ~20% revenue decline in 2025. The remaining single-source risk is concentrated in two nodes: HDSI for high-precision reducers above ~100 Nm torque (where no CN supplier has reached equivalent angular transmission error), and IEC 61508 SIL2-certified safety controller silicon, which requires third-party validation and where only a handful of vendors (Beckhoff, Omron, a few ARM licensees) operate at volume. On the motor/driver side, the competitive landscape has effectively commoditized, with Chinese BLDC module makers closing the gap rapidly and pulling cobot BOM floors toward $5,000 for a 5 kg arm at scale.
Medium-duty warehouse AMR (300–600 kg payload, LiDAR-SLAM, no arm, auto-charging). BOM at fleet volume (50+ units, CN supply chain) ≈ $12k–$22k; low-volume or Western-component build ≈ $35k–$50k. Full system sell price is typically 2–3× BOM (integration, SW, support). Simple QR-code-nav light AMRs can go below $10k BOM; heavy-duty forklift-class AMRs exceed $60k BOM.
Navigation LiDAR is the single largest sensor cost ($800–$2,500 post-Hesai 2025 ~50% price cut); 3D LiDAR for obstacle avoidance adds cost vs. 2D-only; depth cameras and ultrasonic arrays add $200–$600. CN makers (Hesai, Livox, Slamtec) now dominate AMR nav LiDAR globally on price.
Most AMRs use 1–2 certified safety laser scanners (PLd/SIL2); SICK/Omron units run $1,500–$4,000 each and dominate certified builds; CN alternatives (Hikrobot) at $400–$900 but with less global safety certification. Safety scanner + e-stop + safety PLC together are 10–14% of BOM.
NVIDIA Jetson Orin NX (~$300–$600 module) is the de-facto mid/high-tier choice; ARM SoC from Rockchip or Horizon used in cost-tier CN builds (~$80–$200). Includes IO boards, CAN/EtherCAT motor controllers, and embedded SLAM nav compute. Total compute subsystem $1,500–$5,000.
Typically 2 powered differential drive wheels + 2–4 passive casters; CN-made BLDC hub motor+encoder+brake integrated assemblies at $300–$800 each dominate cost-tier builds; Western motors (Maxon, Yaskawa) used in premium/certified AMRs at $800–$3,000 per axis. Motor controllers add $300–$1,000.
底盘与结构件
15%
Oneway Robot (CN) · Roboct (CN) · SEER Robotics (CN) · Local fabrication (CN) · Custom sheet metal (global)
Welded steel or aluminum extrusion frame, cover panels, bumpers, and payload deck. Low software differentiation but weight and rigidity determine motor sizing. Most fabricated in CN ODM/OEM shops at $1,500–$5,000; precision docking pin/magnet assembly adds $500–$1,500.
Per-unit embedded SLAM navigation and fleet management agent license. CN vendors typically bundle into hardware price; Western vendors (MiR, Fetch) increasingly move to RaaS/subscription models. ROS2-based open-source stacks reduce this line to near zero for integrators.
供应商格局：Chinese AMR manufacturers (Geek+, Hikrobot, Hai Robotics, Quicktron, SEER Robotics) now dominate the global volume market on price, with domestic supply chains covering nearly every subsystem from LiDAR to LFP battery to hub motors — a CN-made medium AMR BOM can be 2–4x cheaper than an equivalent Western build. The biggest cost-reduction event in 2025 is LiDAR: Hesai announced a ~50% price cut targeting robotics and sub-$20k EV markets, compressing the sensor line from $2,000–$5,000 to under $1,000 for 3D navigation LiDAR. The primary single-source risk remains certified safety laser scanners: SICK (DE) and Omron (JP) together control the certified (PLd/SIL2) segment required for human-collaborative AMRs in Western markets, and no CN alternative has achieved equivalent global certification at scale as of 2026.
代表产品: Geek+ P800 Goods-to-Person AMR (Geek+ Technology) · Hikrobot Q7 Series AMR (Hikrobot (Hikvision subsidiary)) · MiR250S Collaborative Mobile Robot (Mobile Industrial Robots (MiR / Teradyne)) · OMRON LD-250 Autonomous Mobile Robot (OMRON Robotics) · OTTO 600 Industrial AMR (OTTO Motors (Rockwell Automation))
③ 分模块详解（供应链环节）
整机 OEM (人形/机器狗) conf: med
Humanoid-specific global market; 2026 base ~$6.24B growing to ~$165B by 2034 at ~50% CAGR (Fortune Business Insights). Goldman Sachs projects $38B by 2035 with ~70% CAGR (revised 6x upward). China alone targets ~50,000 humanoid shipments in 2026 (Morgan Stanley). Total robotics market including quadrupeds/service robots is larger.
市场规模 $6.24B
CAGR 50.6%
BOM -
瓶颈：Precision actuators (40-60% of humanoid BOM) dependent on neodymium-iron-boron (NdFeB) rare-earth magnets, of which China controls ~90% of global processing capacity; non-China OEMs face both magnet supply risk and 3x BOM cost disadvantage ($131K vs $46K). Beyond magnets, qualifying robotics-grade high-torque actuators, harmonic reducers and dexterous-hand end effectors at mass-production volumes remains unsolved for most players; supply architecture lock-in is projected 2027-2028.
中国头部 (9)
AgiBot (智元机器人)未上市龙头Unitree + AgiBot together ≈80% of China humanoid shipments (TrendForce 2026); 2025 units ~5,168; scaled 1K→5K→10K units within months (10,000th unit Mar 2026); A2 humanoid in automotive/logistics/retail; founded by ex-Huawei engineers.
Unitree Robotics (宇树科技)未上市龙头Unitree + AgiBot together ≈80% of China humanoid shipments (TrendForce 2026 est.); 2025 units ~5,500 (roughly tied with AgiBot). G1/H1 humanoid + Go2/B2 quadruped; STAR Market IPO approved Jun 1 2026 (~$6.2B val, ¥4.2B raise); capacity to 75K humanoid + 115K quadruped/yr.
UBTech Robotics (优必选)9880.HK挑战者 · 8%Walker S2 in mass production since late 2025; 1,079 full-size units sold in 2025 (35,867% YoY); targeting 5,000 humanoids in 2026; ¥800M+ orders; listed HKEx; automotive factory deployments at BYD and others
Xpeng IRONXPEV挑战者IRON humanoid with 82 DOF including 22-per-hand dexterity; VLA 2.0 AI stack from Xpeng Motors; targeting commercial mass production end-2026 for guided-tour, retail and automotive factory use cases
Xiaomi CyberOne1810.HK挑战者CyberOne V2 with 22-27 DOF; 90% success rate in autonomous automotive assembly pilots; self-deploying on Xiaomi production lines; 2030 full-factory deployment target
Fourier Intelligence (傅利叶智能)未上市挑战者GR-2 humanoid (53 DOF, 175 cm / 63 kg) targeting mass production 2026; 12-DOF dexterous hands with six tactile sensors per finger; serves medical rehabilitation and industrial sectors
Galbot (银河通用机器人)未上市新兴Raised ¥2.5B in Mar 2026 (total ¥7B over 3 yrs, valuation ~¥20B); described as highest-valued unlisted humanoid company in China; mobile manipulation platform targeting logistics; mass production planned 2026
LimX Dynamics (逐际动力)未上市新兴P1 bipedal humanoid; strong bipedal locomotion R&D; ¥1B+ funding round early 2026; industry and logistics focus
Leju Robotics (乐聚机器人)未上市新兴KUAVO humanoid; real-time whole-body control; ¥1B+ round early 2026 pushing valuation to ¥10B+; service, entertainment and factory floor applications
国际头部 (8)
Tesla OptimusTSLA龙头Gen 3 in Fremont and Austin Gigafactories performing battery handling and logistics; volume production targeting H2 2026; $20-30K price target; potentially largest addressable production scale globally if 10K+ unit ramp succeeds
Agility RoboticsCCXI龙头Only humanoid with proven commercial revenue at scale; $2.5B SPAC merger with Churchill Capital XI (→ AGLT ticker, Q4 2026); Digit v5 deployed at Amazon, GXO, Toyota, Schaeffler, Mercado Libre; $300M+ multi-year orders
Boston Dynamics Atlas未上市挑战者Electric Atlas (56 DOF, 50 kg lift) launched CES 2026; 2026 production fully committed to Hyundai RMAC and Google DeepMind; Hyundai Mobis supplies actuators; Spot quadruped in broad commercial use; owned by Hyundai (005380.KS)
Apptronik Apollo未上市挑战者$935M Series A total ($5.3B valuation); NASA-heritage design; targeting <$50K at scale; next-gen Apollo in year-long factory testing as of mid-2026; logistics and manufacturing focus
1X Technologies (NEO)未上市新兴Norwegian startup backed by OpenAI; NEO home robot priced at $20K or $499/mo; first customer deliveries planned 2026; targets unstructured home environments rather than industrial settings
Sanctuary AI未上市新兴Canadian; Phoenix humanoid powered by Carbon general-purpose AI model; Tier-1 automotive manufacturing trials underway; differentiated by AI-first whole-body control architecture
NEURA Robotics未上市新兴German startup; 4NE-1 humanoid for human-robot collaboration; €200M+ raised; leading European humanoid OEM; focuses on cognitive robotics and intuitive safety for shared workspaces
催化剂
Tesla Optimus Gen 3 volume production launch H2 2026 — if confirmed, forces market-wide price expectation reset toward $20-30K and validates mass-market demand
Agility Robotics SPAC listing (AGLT, Q4 2026) as first pure-play public humanoid company — creates valuation benchmark and liquidity for the sector
NVIDIA Isaac GR00T N models and Cosmos world-model platform expanding to 10+ OEM platforms simultaneously — lowers AI brain development barrier and accelerates time-to-deployment
China government subsidies plus 50,000-unit annual shipment milestone in 2026 — creates domestic demand base while Chinese BOM cost advantage intensifies export-cost pressure on Western OEMs
~$2.0B total (harmonic+RV+planetary, industrial robots + nascent humanoid, 2025); 5.9% CAGR for industrial baseline; humanoid-specific sub-segment only ~$52M in 2025 but at 46% CAGR reaching ~$580M by 2032 (Intel Market Research); combined growth rate elevated to ~8.5% as humanoid volume ramps
市场规模 $2B
CAGR 8.5%
BOM Joint actuators account for >30% of humanoid robot BOM (Interact Analysis 2025); reducer/transmission sub-assembly is estimated ~30-40% of actuator cost, implying reducers alone ~10-15% of total humanoid BOM and the single largest cost item within each actuator
瓶颈：Micron-level precision tolerance requirements (arc-second backlash targets) and 18+ month OEM qualification cycles constrain capacity ramp for humanoid-grade units; Japanese leaders (HDSI, Nabtesco) face production ceiling as humanoid demand inflects; Chinese challengers still face a quality/endurance gap in high-cycle dynamic walking applications (humanoid demands >1M cycles vs industrial ~6,000 hours) limiting premium humanoid socket penetration
中国头部 (5)
绿的谐波 Leaderdrive688017.SS挑战者 · 35%China harmonic leader: >60% domestic, ~35% global; PASSED Tesla supplier validation & supplies Tesla's Mexico Optimus line (verified, multi-source); Suzhou plant 500K units/yr in 2026; targets ~60% of Optimus reducer market at 40-60% of Japanese pricing.
来福谐波 Laifu Harmonic3952.HK挑战者 · 21%China #2 harmonic reducer by units (21.4% domestic share); HKEX IPO June 2026 (3952.HK, ~HK$1B raise); FY2025 revenue +142% to 2.61B yuan; one of only two CN makers with verified volume humanoid deliveries at scale
双环传动 Shuanghuan Transmission002472.SZ挑战者China's largest listed precision gear maker; parent of RV-specialist 环动科技 (STAR Market IPO pending); confirmed supplier to Unitree H1 humanoid and Tesla Optimus across RV, harmonic, and planetary types; stock +~500% over 5 years
中大力德 Zhongda Leader002896.SZ挑战者Nidec-backed full-line reducer maker (harmonic+RV+planetary); one of few CN players with all three types in production; FY2024 precision reducer revenue 2.43B yuan; strong in collaborative robot and service robot segments
环动科技 Huandong Technology未上市新兴Dedicated RV reducer unit spun off from 双环传动 (002472.SZ); China domestic RV share grew from ~10% to ~25% (2021-2024), displacing Nabtesco in mid-range applications; STAR Market IPO in progress
国际头部 (5)
Harmonic Drive Systems6324.T龙头Invented strain-wave (harmonic) gear; global market LEADER but not a monopoly — top-5 makers ≈61% of global revenue (2025); core trio = Harmonic Drive + Nabtesco + HDSI. Faces China price pressure (Green Harmonic at 40-60% of its price).
Nabtesco6268.T龙头 · 60%World #1 in RV (cycloidal) reducers with ~60%+ global share for medium-to-large industrial robot joints; FY2025 sales 307.9B yen (+9.8%), OP +60.3%; capacity expansion underway targeting humanoid hip/leg joints
Sumitomo Drive Technologies6302.T挑战者FINE CYCLO® high-precision cycloidal reducers; ~10% global cycloidal market share; deep installed base with European and North American robot OEMs; subsidiary of Sumitomo Heavy Industries (6302.T)
Nidec Drive Technology6594.T挑战者FLEXWAVE strain-wave and precision planetary gears for robotics; Nidec parent also holds strategic stake in China's 中大力德 (002896.SZ); expanding FLEXWAVE line for humanoid joint applications
SchaefflerSHA.F新兴Unveiled integrated planetary gear actuator for humanoid robots at CES 2026 (60-250Nm, 8 product families, fully in-house); vertically integrated motor+gearbox+encoder+controller; deploying in own factories as validation testbed
催化剂
Humanoid production ramp to 50,000+ units globally in 2026 (Tesla Optimus, Figure AI, Unitree, Chinese OEMs) creates step-change demand for miniaturized high-cycle reducers in leg and arm joints
China domestic substitution acceleration: harmonic reducer domestic share in China-assembled robots rose from <5% (2018) to >35% (2025), further tightening Japanese incumbent economics and margin
来福谐波 HKEX IPO (3952.HK, June 2026) raises capital for capacity expansion and benchmarks sector valuation, drawing institutional focus to the segment
Nabtesco production capacity doubling by 2026 and Harmonic Drive FY25 order recovery (4-year order high) signal renewed upstream capex cycle tied to robot super-cycle
Total global ball screw + PRS market (all industrial applications) ~$2.5B in 2025, blended CAGR ~12% (traditional ball screw ~6-8% + humanoid PRS sub-segment ~$0.42B at ~48% CAGR pulling composite higher); humanoid-specific PRS sub-segment per QYResearch and LP Information market reports; total market per multiple research firms including MarkWide, ZionMarket
市场规模 $2.5B
CAGR 12%
BOM High: estimated 8-15% of humanoid robot BOM at current prices (~14 PRS units × $200-300 each per Optimus unit); expected to compress with scale but remains top-3 BOM line item alongside motors and reducers
瓶颈：High-precision 7-axis internal-thread CNC grinding machines (内螺纹磨床) are the critical chokepoint: these are required to grind the nut threads of planetary roller screws to P3 or better precision; global supply is dominated by German (SMS) and Japanese (Mitsui) OEMs with 1-2 year lead times and selective export restrictions to China; domestic Chinese inventory is under 5% of global total; each machine costs ~$1M imported vs ~$300K domestic equivalent (which lags in precision). Secondary constraint: miniaturizing below 8mm diameter for dexterous hand applications requires new process development not yet at volume scale.
中国头部 (6)
恒立液压 Hengli Hydraulics601100.SS龙头Reportedly won majority of Tesla Optimus body screw share; motor factory cleared Tesla audit Dec 2025; leg linear assembly module sent to Tesla for testing; batch production since Sept 2024; 500K sets annual capacity targeted; Mexico plant in planning for export supply
贝斯特 Best Inc.300580.SZ龙头Among first Chinese companies to reportedly pass Tesla on-site supplier audit; C0-precision PRS production; Anhui plant completed May 2024; Thailand greenfield factory for overseas capacity planned for 2026 ramp
秦川机床 Qinchuan Machine Tool000837.SZ挑战者Ball screw P0/P2 precision via Hanjiang Machine Tool subsidiary; reportedly in Tesla supply chain; capacity doubled to 100K sets in 2024 with 300K target; also manufactures PRS thread grinding machines (equipment accounts for ~46% of PRS line cost)
北特科技 Beite Technology603009.SS挑战者18.5亿 investment in Kunshan PRS R&D and production base established Oct 2024; production ramp targeted end-2025/2026; automotive parts background transitioning to humanoid robot supply chain
宁波双林 Shuanglin300100.SZ新兴100K-set PRS production line expected June 2026; acquired internal thread grinding machine technology; 63 ball screw and PRS product variants serving 3 humanoid robot OEM customers
国际头部 (7)
GSA AG未上市龙头 · 26%Global PRS leader; combined with acquired subsidiary Rollvis SA (2016) holds ~52% of planetary roller screw market; confirmed Tesla Optimus supplier for leg and arm linear joints
Rollvis SA未上市龙头 · 26%Acquired by GSA in 2016, operates as independent brand; historically #2 in PRS globally; five product series covering aerospace, industrial, and humanoid robotics applications
Schaeffler AG (Ewellix)SHA.DE龙头 · 14%14% PRS share through Ewellix brand (acquired 2022); presented dedicated humanoid planetary gear actuator at CES 2026; signed 5-year actuator supply deal with UK-based Humanoid Inc. May 2026; targets €100M+ humanoid order book by 2030
THK Co., Ltd.6481.T龙头Global LM guide pioneer and high-end ball screw market leader; THK + NSK combined hold ~90% of high-precision ball screw segment for industrial robots; actively developing humanoid robot-specific linear motion solutions
Bosch Rexroth未上市挑战者 · 12%12% PRS market share as Robert Bosch GmbH subsidiary; industrial-standard roller screw components using rolled-forming technology; broad linear motion portfolio for servo-driven applications
NSK Ltd.6471.T挑战者Developing dedicated humanoid robot actuator suite (rotary + linear), targeting 2028 commercialization; strong incumbent in industrial ball screws; one of global top-3 ball screw suppliers
SKF ABSKFB.ST挑战者Named alongside THK and NSK as dominant global ball screw market participant; offers full linear motion and ball screw portfolio for robotics; increasing robotics segment focus for 2026
催化剂
Tesla Optimus Gen 3 mass production targeting 50K+ units in 2026, implying 700K+ PRS demand annually (14 screws per robot) and forcing multi-supplier qualification
Chinese humanoid OEM production ramp: Agibot reached 10K cumulative units by March 2026; Unitree, UBTECH, and others in active scale-up, collectively requiring tens of millions of ball and roller screws per year by 2027
Domestic Chinese PRS production lines coming online H2 2026 (Shuanglin 100K-set line, XCC 980K-unit plan, Hengli 500K sets), expected to drive per-unit cost from current ~$200-300 toward sub-$50 within 2-3 years
China released national humanoid robot technical standards March 2026, standardizing component specs and accelerating domestic supplier qualification cycles
Humanoid-robot-specific motors (all types combined); source: QY Research 2026. Frameless sub-segment grows faster: $0.04B in 2024 at 44% CAGR to $0.58B by 2031. Coreless sub-segment alone estimated ~$0.5B in 2025. Total servo motor market (all robotics) is ~$5B.
市场规模 $0.64B
CAGR 12.4%
BOM ~12% of robot BOM at $20K target cost (36KR); frameless torque alone ~16% of component value (TechBuzzChina); coreless adds ~8%; combined motor module is 2nd costliest category after lead screws; reference classification: med
瓶颈：Ultra-precision winding technology (sub-0.05mm copper wire) for coreless cup motors limits globally qualified suppliers to a handful; domestic China winding machine capability still relies partly on Japanese/German imports; rare-earth NdFeB magnet supply and processing add a second choke; combined effect: only ~3-5 companies globally can supply production-ready coreless motors at humanoid-grade specs, creating acute allocation risk as Optimus and domestic Chinese robot volumes scale in 2026.
中国头部 (7)
鸣志电器 / MOONS' Industries603728.SS龙头Global top-3 in coreless cup motors; passed Tesla Optimus C-round certification; ¥200M+ 2026 orders confirmed; 2025 robot revenue +300%; direct-wound tech enables 40-70% price discount vs Maxon
步科股份 / Kinco688160.SS龙头Highest-volume Chinese frameless torque motor maker; 3.5万台 shipped Q1 2026 (+246% YoY); transitioned from small-batch to mass production; 4th-gen frameless validated across multiple humanoid OEMs
雷赛智能 / Leadshine Intelligence002979.SZ挑战者Full-stack: frameless torque + coreless + 21-DOF dexterous hand modules; 12万+ units shipped 2025 (20× growth); ~70% of top-tier China humanoid OEMs are customers; building 2M unit/yr capacity
卧龙电驱 / Wolong Electric Drive600580.SS挑战者Exclusive frameless torque supplier for Unitree Walker S2; 20万台 delivered Jan-Oct 2025; >80万台 backlog; mutual equity stake with Unitree; pursuing A+H dual listing
昊志机电 / Haozhi Electromechanical300503.SZ挑战者Frameless torque motors; achieved first batch production in collaborative massage robots; now in prototype testing with multiple humanoid OEMs; early-mover in domestic frameless supply chain
江苏雷利 / Jiangsu Leili300660.SZ挑战者Coreless cup motors + planetary roller screws; 3rd-gen dexterous hand product; Weilai Robotics and Unitree design partnerships; FY2024 revenue ¥3B (+21% YoY)
鼎智科技 / Dingzhi Technology873593.BJ新兴One of few domestic makers with fully automated coreless cup motor mass production; ~¥600/unit for humanoid dexterous-hand actuators; Beijing Stock Exchange emerging player
国际头部 (7)
Maxon Group未上市龙头 · 20%Global #1-2 in precision coreless motors (~CHF 600M revenue); benchmark spec for humanoid hands; Chinese challengers undercut by 40-70% at ¥1,200-2,300 vs Maxon's ¥4,000+
Faulhaber未上市龙头 · 17%Germany-based family business, ~2,300 employees; global #2 in coreless DC motors; 3mm minimum motor size; BXI integrated drive for robotics; preferred spec in EU humanoid programs
KollmorgenFTV龙头Pioneer of frameless torque motor technology (TBM2G/TB-series); among earliest globally to develop frameless; Fortive subsidiary; key reference design for humanoid rotary joint modules
AMETEK (Precision Motion)AME龙头Top-3 global coreless motor supplier per 36KR supply chain analysis; Precision Motion Group unit serves robotics/dexterous-hand market; rated high probability for Optimus supply selection
Nidec Corporation6594.T龙头 · 11%Diversified motor giant with ~11% coreless DC market share; Nidec Drive Technology expanding servo/gearing for robotics; broad capacity but humanoid-specific business still <5% of group revenue
Allient Inc. (Allied Motion)ALNT挑战者 · 9%Pioneer and large producer of frameless torque motors; ElectroFlux and ultra-slim axial-flux series for humanoid joints; showcasing at Automate 2026; ~9% global coreless share
Moog Inc.MOG.A挑战者Silencer-series frameless brushless DC motors for precision robotics; established in defense/aerospace motion control extending into humanoid; competes on reliability and certification pedigree
催化剂
Tesla Optimus Gen 3 production ramp at Fremont/Texas in 2026 could drive coreless motor demand into the millions-per-year range, pulling qualified Chinese suppliers (鸣志 C-round) into mass supply
China domestic humanoid robot volume crossing the 万台 threshold in 2026 (Unitree, AgiBot, Galaxy) converts prototype orders to sustained production contracts for frameless and coreless suppliers
Chinese OEMs completing D-round qualification with international customers accelerates import substitution: 鸣志 priced at ¥1,200-2,300 vs Maxon ¥4,000+, cutting per-robot BOM by several thousand RMB
Vertical integration trend (Unitree self-developing joint modules) may compress tier-1 motor supplier margins but creates design-in opportunities for component specialists with strong IP in winding and magnet assembly
2025E combined TAM for FT sensors, encoders, and IMUs across all robotics segments (industrial, collaborative, humanoid). Humanoid robot sensor market alone estimated at $1.84B (2025) at 25.8% CAGR to $9.2B by 2032 (Maximize Market Research); 6-axis FT sensor segment at $1.07-1.28B; humanoid encoder segment ~$250M; MEMS IMU all-applications ~$1.21B. FT sensor unit demand surged ~11x in China in 2024 on humanoid ramp.
市场规模 $2.5B
CAGR 24%
BOM med (~10–15% of humanoid robot BOM; typically 4 six-axis FT units per robot at $200–2,000 each, 28+ encoder channels integrated in actuators, and 2–4 IMUs; FT sensors are the highest unit-cost sub-segment; sensors overall estimated at ~10% of ~$50K unit-scale BOM per humanoid.guide)
瓶颈：No scalable MEMS 6-DOF force-torque sensor IC exists in production volume as of Q1 2026—foil strain-gauge bonding to elastic bodies requires skilled manual labor that cannot be automated to semiconductor scale, capping addressable supply at tens of thousands of units/year against projected humanoid demand in the millions. Current workarounds (Tesla Optimus and Figure AI use custom in-house assemblies at $200–800/unit) are not sustainable beyond early-ramp volumes. Secondary constraint: encoder miniaturization for sub-30mm joint form factors without sacrificing resolution or thermal stability.
中国头部 (5)
宇立仪器 / Sunrise Instruments (SRI)未上市龙头Pioneer Chinese 6-axis FT sensor maker based in Nanning; large-volume humanoid wrist/ankle supplier; priced ~50% below ATI equivalents; dominant domestic market position across collaborative and humanoid robot OEMs
坤维科技 / Kunwei Technology未上市挑战者Joint torque + 6-axis FT sensor specialist (founded 2018, aerospace R&D lineage); completed Series B+ in 2025; won 'Excellent Supplier' at 1st China Humanoid Robot Industry Conference; large-scale delivery commenced 2026; 30% cost below ATI
柯力传感 / Keli Sensing603662.SS挑战者A-share listed sensor incumbent; >1,500 humanoid FT sensors shipped in 2025 to 70+ robot OEM customers; monthly run-rate exceeded 1,000 units in 2026; FY2025 revenue +20% YoY driven by robot sensor ramp
汉威科技 / Hanwei Technology300007.SZ挑战者Gas sensor incumbent pivoting to robotics; flexible tactile sensors for dexterous hands (30 robot OEM partnerships as of H2 2025); IMU + 6-axis FT module win at BYD exoskeleton program; ¥200M projected 2026 volume orders
鑫精诚 / XJCSENSOR未上市新兴Launched China's first automated MEMS 6-axis FT sensor production line (Aug 2025); unveiled at World Robot Conference 2025 in Beijing; targets sub-$100 unit cost vs $2K–8K for conventional strain-gauge sensors
国际头部 (7)
ATI Industrial Automation (Novanta)NVST龙头World leader in 6-axis FT sensors (30+ yr track record, now Novanta subsidiary); gold standard for cobots and surgical robots; instruments at $2K–8K; developing humanoid-specific miniaturized designs
HEIDENHAIN未上市龙头German precision encoder incumbent; ECI/EQI inductive rotary series dominant in humanoid joint feedback; dual-encoder (pre+post gearbox) architecture adopted by leading platforms; key products showcased at Automatica 2025
RenishawRSW.L龙头UK precision metrology leader; optical and magnetic encoders widely used in cobot and humanoid joints; >5M encoder units/year; 2023 dedicated robotics encoder launch targeting dexterous manipulation applications
TDK / InvenSense6762.T龙头ICM-42688-P 6-axis MEMS IMU widely adopted for humanoid state estimation and balance control; Qualcomm Robotics RB3 platform design win; MEMS IMU accounts for ~60-65% of humanoid IMU unit demand
Bota Systems未上市挑战者ETH Zürich spinout (Switzerland); SensONE and PixONE compact 6-axis FT sensors for cobots and humanoid joints; PixONE 30% lighter than comparable sensors; SensONE T80 handles up to 80 Nm for heavy-payload cobots
OnRobot未上市挑战者Danish one-stop-shop for cobot accessories; 6-axis FT sensor fits most industrial robot arms; growing into humanoid adjacency via force-sensing grippers and wrist modules
TE ConnectivityTEL挑战者Broad connector and sensor portfolio ubiquitous in robot wiring harnesses; force, pressure, and position sensors for cobot joint modules; connectors are standard in most humanoid sensor integration stacks
催化剂
Tesla Optimus and Chinese humanoid OEMs (Unitree, AgiBot) targeting 100K+ units/year by 2026-2027, each requiring 4-6 six-axis FT sensors and 56+ encoder channels per robot, compressing prices and forcing supply chain buildout
China HEIS 2026 national standard system (released March 2026 by MIIT) specifying sensor interface and performance standards across 120+ institutions, unlocking mass procurement and accelerating domestic supplier qualification
Chinese domestic FT sensor prices now ~50% of imported equivalents; six-axis FT sensor localization rate surged from 19% (2020) to 57.8% (2024), expanding BOM affordability for all humanoid makers
MEMS FT sensor IC development programs anticipated at Bosch Sensortec and STMicroelectronics for ~2028-2030 commercial availability, expected to collapse per-unit costs 10x and eliminate the manual-assembly bottleneck
Valuates Reports (Jan 2026); total dexterous hands all types (humanoid + industrial + prosthetics); humanoid-specific multi-fingered sub-segment ~$0.09B in 2024 projected to ~$5.0B by 2031 at CAGR ~68.7% (QYResearch/Valuates); China volume: ~19,200 units in 2025 → ~70,200 units in 2026 (GGII)
市场规模 $0.815B
CAGR 40.4%
BOM ~31% of humanoid robot BOM (high-DOF configurations; broker consensus 25–31%; ~17% in Tesla Optimus Gen 2 per Morgan Stanley; largest single module in humanoid BOM, driven by coreless motors ~50%, tactile sensors ~30%, micro-transmission ~12% of hand sub-BOM)
瓶颈：Miniaturized coreless (hollow-cup) motor supply constrained by European import dominance (Maxon/Faulhaber/Portescap at ~4,500 CNY/unit vs domestic target ~1,500 CNY), compounded by immature mass-production consistency of high-density tactile sensor arrays — both limit cost deflation speed and AI-policy training data quality for reliable grasping of diverse unstructured objects
中国头部 (6)
灵心巧手 Linkerbot未上市龙头 · 80%Claims >80% of global high-DOF dexterous hand segment (company-stated, segment-specific); 10,000+ units shipped as of Apr 2025; >1,000 units/month peak output; Samsung and Siemens named as OEM customers; $150M raised total; Beijing-based
因时机器人 InTime/Inspire Robots未上市龙头10,000+ dexterous hand units delivered in 2025; featured on CCTV Spring Festival Gala 2026; fully self-developed core components; alongside Linkerbot forms China first-tier duopoly in volume delivery
兆威机电 Zhaowi Mechatronics003021.SZ挑战者Only pure-play listed A-share dexterous hand co; ZWHAND DM17 (17 DOF) and B20 (20 DOF, 600g) in mass production; dual-listed A+H (02692.HK, Mar 2026); embodied AI revenue +281% YoY in 9M 2025; signed 12 strategic OEM partnerships; micro-drive expertise from micro-transmission core business
宇树科技 Unitree Dex未上市挑战者Dex5-1P: 20 DOF (16 active), 94 tactile sensors including palm and fingertips; sold as OEM add-on to G1/H1 humanoid; benefits from Unitree's leading low-cost humanoid platform ecosystem
傲意科技 OYMotion未上市新兴ROHAND (6 active / 11 total DOF); myoelectric sensor array + AI intent recognition; 100+ core patents; Shanghai Zhangjiang-based; used in academic dexterous manipulation benchmarks; prosthetics background expanding into industrial humanoid
国际头部 (6)
Tesla (Optimus in-house)TSLA龙头Gen 3 Optimus hand: 50 actuators per forearm, tendon-driven 22 DOF; all components custom-designed in-house; not sold externally; Model S line repurposed for Optimus Gen 3 targeting 2026 ramp; largest single design-win that sets humanoid hand performance benchmarks
Shadow Robot未上市龙头20-year benchmark hand maker (UK); DEX-EE series co-developed with Google DeepMind (2024-2025); classic Shadow Hand (24 DOF, 4.3kg) is de facto research gold standard; commercial availability with AI/ML integrations
SCHUNK未上市挑战者SVH 5-finger servo-electric industrial hand; spun out dedicated humanoid hand subsidiary (Jan 2026); joint Bosch-SCHUNK industrial-grade humanoid hand showcased at Bosch Connected World (Jun 2026); German industrial precision focus
Wonik Robotics / Allegro Hand未上市挑战者Allegro Hand V5 (16 DOF, 4-finger); most widely deployed research dexterous hand globally; used by Google DeepMind, Stanford, MIT, CMU, ETH Zurich; ~$12K-15K ASP; V5 adds improved torque control and ROS2 compatibility; production in South Korea
Clone Robotics未上市新兴27-DOF musculoskeletal biomimetic hand using hydraulic artificial muscles; 279 Clone Alpha units in 2025 pre-order; dual HQ Mountain View CA + Poland; targets 'Robo Butler' enterprise market 2028; under $20K full-robot build cost; pre-commercial
Agile Robots SE未上市新兴Agile ONE humanoid (Nov 2025) integrates proprietary dexterous hand with per-joint force-torque + tactile fingertip sensing; SoftBank-led €745M round; Munich-based; acquired Franka Emika and Thyssenkrupp Automation Engineering assets; powers Europe's largest AI factory pilot
催化剂
Tesla Optimus Gen 3 mass production ramp (2026 target): first >50-actuator-per-hand platform at scale, anchoring high-DOF supplier qualification pipelines globally
China humanoid factory deployments (UBTECH, AgiBot, Unitree) driving domestic volume jump from ~19,200 units (2025) to ~70,200 units (2026)
Dexterous hand ASP inflection: unit prices falling from ~130K CNY (current high-DOF) toward sub-10K CNY threshold with scale, unlocking commercial humanoid economics
Open-ecosystem tactile+visuotactile API platforms (BrainCo Revo3, Linkerbot) reducing AI policy development cycle, accelerating design-win velocity for OEM integrators
Humanoid-specific crossed roller bearing sub-market: $16M (2025) → $213M (2032), CAGR ~45%. Total precision cross-roller bearing market (including industrial robots, semiconductor, CNC): ~$1.44B (2025) at ~2.6–3.5% CAGR. Humanoid sub-segment is the fast-growth driver; total market figure is more representative of current supplier revenue base.
市场规模 $0.016B
CAGR 45%
BOM med
瓶颈：High-precision (P4/P5-grade) crossed roller and angular contact bearings remain dominated by Japanese makers (THK, IKO) and Schaeffler INA; China's high-end bearing localization rate is below 20%. Scaling to humanoid mass production requires orders-of-magnitude volume increase vs current robotics supply; peak lead times exceed 26 weeks. Domestic Chinese alternatives (HONB, 国机精工) cover mid-tier precision but consistent P4-grade tolerance for humanoid joint dynamic loading remains a qualification gap, creating a dual-track supply risk of both volume shortage and precision shortfall.
中国头部 (7)
HONB (洛阳鸿元轴承 Luoyang Hongyuan)未上市龙头 · 80%Claimed ~80% domestic China share in crossed roller bearings and ~90% in harmonic reducer bearings (company claim corroborated by CN financial media; domestic market only); key import-substitution reference for domestic humanoid OEMs; unlisted
洛轴LYC (洛阳轴承集团 Luoyang Bearing Group)未上市挑战者China's largest bearing manufacturer by output volume; full range including P4/P5 precision bearings; IPO review passed May 2026 (not yet listed); serves machine tools, wind, rail, and emerging robotics verticals
国机精工 (Guoji Jingong / Sino-Precision)002046.SZ挑战者Aerospace-grade P4-level crossed roller bearings applying to humanoid joints; part of China National Machinery (CNMC); humanoid bearing formally in 15th Five-Year Plan strategic roadmap; formerly listed as 轴研科技
五洲新春 (XCC Group)603667.SS挑战者Precision bearing rings for harmonic and RV reducers; robot bearing delivery CAGR projected >50% over 5 years; ceramic bearing line extends lifespan 3× vs steel; major domestic reducer manufacturer supply partner
长盛轴承 (Changsheng Bearing)300718.SZ挑战者Self-lubricating bearings for robot joints; reported design win with Unitree Robotics (宇树科技), leading domestic humanoid OEM; 98.7% yield in high-frequency robot scenarios vs 92% industry average
龙溪股份 (Longxi)600592.SS挑战者World's #1 in joint/spherical plain bearings; supplies C919 aircraft and Shenzhou spacecraft; military-grade precision heritage; diversifying into humanoid robot joint bearings leveraging high-tolerance manufacturing
苏轴股份 (Suzhou Bearing)430418.BJ新兴Ultra-thin precision needle roller bearings; robot bearing revenue grew from 5% to 12% of total in 2024; Beijing Stock Exchange listed small-cap; targeting robot bearing as second growth curve
国际头部 (6)
THK Co., Ltd.6481.T龙头Global pioneer of LM Guide and crossed roller bearings; standard robotics motion supplier; strategic partnership for sensor-integrated smart CRB systems (Sept 2024); one of top-2 global CRB suppliers alongside IKO
IKO / Nippon Thompson6480.T龙头CRBH/CRB/CRBS ultra-thin series are industry standard for robot wrist and waist joints; XR series offers 30% higher load capacity vs standard; with THK commands majority of global precision crossed roller bearing supply
Schaeffler GroupSHA.DE龙头Supplies precision bearings and smart actuators to ~45 humanoid OEM customers; signed 1,000+ robot purchase deal (May 2026); CEO targets hundreds-of-millions-EUR humanoid order book by 2030; SKF+Schaeffler ~36% total bearing market revenue
NSK Ltd.6471.T挑战者Developing dedicated humanoid robot rotary/linear actuators targeting 2028 launch; signed MOU with Delta Electronics (May 2026) for robot joint module co-development; NSK-NTN merger announced May 2026 (close Oct 2027)
Timken CompanyTKR挑战者Robotics/automation ~8% of sales (>$350M), 10%+ CAGR; acquired CGI Motion (Aug 2024) to expand into medical and industrial robotics; explicitly targeting humanoid bearing supply chain
NTN Corporation6472.T挑战者Precision bearings for industrial robot joints; merging with NSK (close Oct 2027) to create world's largest bearing maker with explicit humanoid robot and robotics focus areas
催化剂
Tesla Optimus Gen3 mass production ramp H2 2026 (50K unit initial target, line capacity 1M/yr) creates first large-scale demand inflection for crossed roller and angular contact bearings in humanoid joints
NSK-NTN merger (signed May 2026, closing Oct 2027) creates world's largest bearing maker with explicit humanoid robot bearing development as a named growth mandate
China domestic humanoid OEM production ramp (Unitree, XPeng IRON, BYD) accelerating import substitution race; A-share bearing stocks saw 2–3× price appreciation in 2025 on supply-chain re-rating
LiDAR for Robotics segment (2025, total not humanoid-specific), projected $14.46B by 2034; broader robotics perception incl. machine vision systems adds another ~$4.5B; humanoid robot sensors TAM ~$1.84B (all sensor types, 25.8% CAGR to 2032), vision/imaging is ~34% of humanoid sensor spend
市场规模 $4.79B
CAGR 13.1%
BOM med (perception sensors estimated at 15–25% of humanoid robot BOM; second-largest cost center after actuators; LiDAR units alone at current ASPs represent the single largest per-unit line item in the perception stack)
瓶颈：SPAD detector ASICs and VCSEL chips: high-performance dToF LiDAR receivers remain dominated by Sony and ON Semiconductor; VCSEL manufacturing scale-up is the pacing constraint for solid-state LiDAR cost reduction below $100–200/unit needed for mass humanoid deployment; US export controls on advanced semiconductor equipment limit Chinese LiDAR makers' access to leading-edge detector fabs; solid-state LiDAR yield and reliability at scale remains unproven in high-volume humanoid production
中国头部 (5)
Hesai 禾赛HSAI龙头Global #1 LiDAR by total volume (1.62M units in 2025, +223% YoY); #1 in humanoid+quadruped segment per GGII/Yole; JT128 equips every Unitree robot at 2026 Spring Festival Gala; 2026 target 3–3.5M units; revenue 3.03B RMB in 2025
RoboSense 速腾聚创2498.HK龙头#1 global robotics LiDAR shipments Q1 2026 per GGII; 185,500 robotics units in Q1 2026 alone (+1,459% YoY); ~50 humanoid robot OEM partnerships; proprietary SPAD-SoC+VCSEL chip architecture (EM Digital Platform); achieved first-ever quarterly profit
Orbbec 奥比中光688322.SS挑战者Leading RGB-D/stereo depth camera for Chinese humanoid robots; Gemini 305 wrist-mount camera for humanoid hand manipulation launched CES 2026; NVIDIA Jetson Thor certified; Basler OEM partnership for 3D robot vision; dual China+Vietnam manufacturing
Livox / DJI未上市挑战者DJI-incubated LiDAR subsidiary; Mid-360S hybrid solid-state 360°H×59°V LiDAR for mobile robot SLAM and obstacle avoidance; cost-competitive vs mechanical spinning LiDAR; widely deployed in research and commercial robots globally
Seyond (前Innovusion)未上市挑战者Hummingbird D1-R fully solid-state LiDAR for robotics (CES 2026); no moving parts; 140°×100° FOV; world-first automotive solid-state OEM design win confirming production readiness; end-to-end portfolio from Falcon K to Hummingbird
国际头部 (8)
RealSense (ex-Intel)未上市龙头Spun out from Intel Oct 2025 with $50M (Intel Capital + MediaTek Innovation Fund); NVIDIA Physical AI collaboration; D585 Pro AI-native depth camera unveiled Automate 2026 (ships Q1 2027); company self-reports dominant share in AMRs and humanoid robots (unverified by independent source)
OusterOUST龙头Last credible Western pure-play LiDAR after Luminar's 2025 bankruptcy; absorbed Velodyne 2023; acquired Stereolabs (ZED stereo cameras) Feb 2026 for $35M adding 90K+ deployed depth cameras; Rev8 native color LiDAR; key adopters: Google, Skydio, Volvo Autonomous Solutions
Keyence6861.T龙头 · 12.1%>12.1% share in robotic vision systems (2025, market reports); AI-powered In-Sight vision sensors dominating industrial factory robot inspection; expanding into embodied AI vision guidance and AMR perception; strong Japanese and global install base
Sony Semiconductor6758.T挑战者IMX560 SPAD dToF and IMX556 iToF depth sensor chips; IEC 61508 SIL 2 functional safety certification obtained Apr 2026; foundational chip supplier for third-party depth cameras (Orbbec, Basler, Seeed, Lucid); global CMOS image sensor market leader
Teledyne TechnologiesTDY挑战者FLIR thermal imaging + DALSA area-scan cameras; Tura ASIL-B qualified thermal camera for AV/ADAS (CES 2026); Lepton XDS compact thermal; machine vision for industrial robot inspection; strong in defense and precision industrial imaging
CognexCGNX挑战者Leading 2D/3D machine vision system for industrial robots; In-Sight 3900 runs inspections 4× faster; part of top-5 vision players holding ~42.7% collective market share; expanding into embodied AI platforms and humanoid robot guidance as market evolves
Innoviz TechnologiesINVZ挑战者InnovizThree: first sensor-fusion colored 3D LiDAR+camera in single unit (CES 2026); 600g, 35% cost reduction vs prior gen; compact form explicitly targeting humanoids, drones, micro-robotics; BMW OEM automotive design win validates production pedigree
AevaAEVA新兴FMCW 4D LiDAR-on-Chip providing per-point velocity data; new Physical AI sensor introduced CES 2026; NVIDIA autonomous driving platform integration; Cadence Tensilica Vision DSP licensed for industrial robotics signal processing; differentiates via instantaneous velocity sensing
催化剂
Humanoid mass-production volume inflection: Tesla Optimus Gen 3 Fremont ramp (Jan 2026) and multiple Chinese humanoid OEMs targeting 10K+/month are driving perception sensor demand to a new scale requiring dedicated supply chains
Ouster + Stereolabs integration creating unified digital LiDAR + stereo depth stack for Western humanoid OEMs seeking a single-source perception supplier after Luminar's bankruptcy consolidates the field
AI-native perception chips: RealSense + NVIDIA collaboration and Orbbec + Jetson Thor certification enabling on-device neural inference in the camera module itself, reducing compute BOM and latency in robot manipulation tasks
Edge AI hardware market (broad, includes robotics/ADAS/industrial) at $5.2B (2025), growing to ~$20.4B by 2034 at 16.3% CAGR (GMInsights). Humanoid-robot-specific SoC sub-segment much smaller: ~$0.04B (2025) at ~50% CAGR reaching ~$0.68B by 2032 (DataInsightsMarket); total figure cited here is not humanoid-specific.
市场规模 $5.2B
CAGR 16.3%
BOM ~8% of humanoid robot BOM (2025) as silicon/compute line item; part of broader 10-20% perception+compute category; expected to decline to ~5% by 2035 as scale drives unit cost down (BofA/RoboZaps estimates)
瓶颈：TSMC advanced-node (3nm/2nm) wafer capacity and CoWoS advanced packaging sold out through 2026-2027; humanoid robot SoC demand stacks on top of existing AI-datacenter backlog, creating hard supply ceiling. Chinese players face an additional structural constraint: US export controls limit SMIC to ~7nm N+2 equivalent, capping peak on-chip performance versus NVIDIA/Qualcomm built on TSMC N3/N4. Software ecosystem fragmentation (CUDA vs. non-CUDA stacks) creates a secondary lock-in bottleneck for challengers.
中国头部 (6)
Horizon Robotics (地平线)9660.HK龙头Journey 6 (560 TOPS) leads ADAS; D-Robotics spinoff RDK S100 (80-128 TOPS, ARM A78AE) purpose-built for humanoid/quadruped robots; 5M+ cumulative edge AI chips shipped; 10 automaker + 12 Tier-1 design wins for J6; BYD and VW/CARIAD signed; integrates brain+cerebellum for embodied AI.
Black Sesame Intelligence (黑芝麻智能)2533.HK挑战者Huashan A2000 (500+ TOPS, 7nm, 8TB/s near-memory cache) + Wudang C1236 cerebellum chip form a dual-SoC humanoid brain-cerebellum platform; confirmed design win: Wuhan University Tianwen humanoid robot; A2000 mass production 2026; 10M+ chip shipment target for 2026.
Rockchip (瑞芯微)603893.SS挑战者RK3588 (6 TOPS NPU, 8-core A76+A55, TSMC 8nm) is cost-effective workhorse for mid-tier Chinese humanoid robots: ZHIYUAN Lingxi X2, LimX Oli, Gaoqing Pi; new RK3688 (12-core, 4nm) in roadmap; AIoT/robot segment drove strong H1-2025 growth; weak in top-tier >100 TOPS humanoid compute.
Huawei Ascend (华为昇腾)未上市挑战者Ascend 910C (320 TFLOPS FP16, SMIC 7nm N+2, ~53B transistors); Atlas series used in industrial edge robots; 37% CN AI accelerator market share (Q1 2026); ~600K unit production target 2026; 910D (FP8) roadmap for later 2026; US export controls block overseas sales and restrict node advancement.
D-Robotics (地瓜机器人)未上市新兴Horizon Robotics spinoff focused exclusively on robot edge compute; RDK S100 (80/128 TOPS, ARM A78AE) debuted Embedded World 2026; RDK X3/X5 series with 5M+ units shipped; M-Robots OS integration planned; $100M Series A raised; humanoid, quadruped, and logistics robot focus.
国际头部 (6)
NVIDIANVDA龙头 · 11%Jetson Thor (2070 FP4 TFLOPS, 130W) is the de-facto humanoid compute standard; design wins include Agility Robotics Digit Gen 6, Boston Dynamics Atlas, Galaxy Galbot, 1X NEO; GR00T open reference design debuted COMPUTEX 2026; CUDA/Isaac GR00T software lock-in reinforces dominance.
QualcommQCOM挑战者Dragonwing IQ10 robotics reference design announced CES 2026, commercial availability Sep 2026; Hexagon NPU + 5G + 12 GMSL2 camera inputs targets AMR and humanoid; power-efficient automotive-grade alternative to Jetson; executes same platform playbook as Snapdragon Digital Chassis.
IntelINTC挑战者Core Ultra (Meteor Lake) with Arc iGPU + NPU (up to 11 TOPS); powers Ubtech Walker X (i7-8665U + Core i7); Intel AI Suites provide robotics-specific SDKs; focus on industrial and collaborative robots rather than pure humanoid tier.
Texas InstrumentsTXN挑战者New high-performance computing SoC family (announced Feb 2026) delivers up to 1200 TOPS via proprietary NPU with chiplet-ready design; explicitly targets humanoid robot edge AI; leverages large embedded/microcontroller installed base in industrial robotics.
Samsung005930.KS挑战者Exynos 2600 (first 2nm mobile SoC, in-house GPU); partnering with NVIDIA to build robotics AI factories using Jetson Thor; Korea government $649B AI push (Jun 2026) prioritises chips and robotics; no direct humanoid robot chip design win confirmed as of Jun 2026.
AmbarellaAMBA新兴CV3/CV7 edge vision SoC strong in perception and multi-sensor fusion; ~80% of FY2026 revenue from edge AI; ~45M edge AI SoCs shipped cumulatively; robotic aerial drone production win; CV7 offers 2.5x AI uplift over predecessor at 20% lower power.
催化剂
NVIDIA Jetson Thor / IGX Thor commercial ramp and GR00T open reference design (COMPUTEX 2026) shortens dev cycles; Jetson T4000 ($1,999 at 1K units) extends reach to budget AMR fleets
Qualcomm Dragonwing IQ10 global commercial availability (Sep 2026) provides power-efficient alternative to Jetson and breaks near-monopoly in mobile/5G-connected humanoid deployments
Tesla AI5 chip tape-out (Apr 2026) targeting Optimus mass production mid-2027 validates vertically-integrated silicon path; expected to trigger similar captive-SoC strategies at Figure, Apptronik, and Chinese OEMs
China national humanoid robot scale-up (2025-2026): Unitree, ZHIYUAN, Agibot production mandates force rapid domestic SoC qualification cycles for RK3588, A2000, and RDK S100, redirecting volume away from imported Jetson Orin
Humanoid-specific robot battery market: USD 14M (2025) → USD 24M (2026) → USD 851M (2034) at ~74% CAGR (24MarketReports). TrendForce projects solid-state demand for humanoid robots at 74 GWh by 2035 (~1,500× growth). Broader all-robot battery market estimated ~USD 2.5B+ with ~15.5% CAGR to 2030 (MarketsandMarkets). Figures are humanoid-specific unless noted.
市场规模 $0.024B
CAGR 74%
BOM med; ~3–8% of humanoid unit BOM (pack $300–1,100 on a $30–50k robot); actuators/joints dominate at >30%; battery is a mid-tier cost contributor today but share rises if solid-state cells command premium pricing
瓶颈：Energy density constrained by weight budget: humanoids can allocate only ~1/8 of total mass to batteries (vs ~1/3 for EVs), yielding 2–4 h runtime on current 2–3 kWh Li-ion packs. Solid-state cell mass production has slipped to 2H 2027+ for leading Korean/Chinese players, leaving a multi-year gap. Thermal management within tight bipedal form factors is unsolved at scale. Joint/actuator design uncertainty makes it hard to finalize custom battery integration spaces, further delaying optimized cell qualification.
中国头部 (5)
CATL 宁德时代300750.SZ龙头World's largest battery maker (~37% global EV share); first to deploy humanoid robots (Spirit AI Galbot) at scale in its own Luoyang battery-pack production line (Dec 2025); solid-state roadmap via CASIP consortium; strategic investor in robot startups.
EVE Energy 亿纬锂能300014.SZ龙头Launched Longquan II all-solid-state cell (300 Wh/kg, 700 Wh/L) explicitly targeting humanoids; Chengdu solid-state production base Phase 1 complete Dec 2025 (~500k cells/yr); targets 360 Wh/kg by 2026; Tesla EV energy-storage supplier; exhibited robot battery solutions at World Robot Expo 2025.
Sunwoda 欣旺达300207.SZ挑战者Tesla's 5th global EV battery supplier (3rd-gen LFP, Yiwu factory); developing 350 Wh/kg high-nickel cells for humanoid robots; actively building robot battery layout for 2026; raised scrutiny over legal disputes but operationally expanding.
BYD 比亚迪002594.SZ新兴Confirmed humanoid robot development program (Jun 2026); vertically integrated from battery chemistry to final assembly; massive Blade LFP and ternary scale if redirected to robot supply; plans dealer-network robot distribution.
SVOLT Energy Technology 蜂巢能源未上市新兴Great Wall Motor spinoff; cylindrical and pouch cells; actively pursuing humanoid robot battery customers; limited confirmed robot design wins to date.
国际头部 (5)
Samsung SDI006400.KS龙头Unveiled first pouch-type all-solid-state battery sample for physical AI / humanoid robots at InterBattery Seoul (Mar 2026); mass production targeted 2H 2027; collaboration with Hyundai Motor and Kia on robot-optimized cells; expanding from EV prismatic SSB into humanoid form factors.
LG Energy Solution373220.KS龙头Confirmed supplier of NCM 2170 cylindrical cells to Tesla Optimus Gen 2; exclusive cylindrical battery supplier to Bear Robotics (Servi Plus / AMR Cart) from 2025; supplying 6 global robotics firms per company disclosure; 4680 cells and anode-less SSB in next-gen pipeline for 2030 target.
Panasonic Energy (Panasonic Holdings)6752.T挑战者Solid-state battery samples for robot/automotive applications targeted FY ending Mar 2027; 2170 cylindrical cell manufacturing in Japan; multi-year 2170 supply deal for Zoox robotaxi (starts 2026); longtime Tesla cylindrical cell partner—natural humanoid candidate as Optimus scales.
Murata Manufacturing (ex-Sony Energy Devices)6981.T挑战者Acquired Sony's battery business (2017); manufactures 18650/21700 cylindrical Li-ion targeting high-drain applications (power tools, ESS); expanding robotics segment; conservative on design-win announcements but mature cell quality recognized by robot developers.
Solid PowerSLDP新兴US solid-state battery developer; BMW and SK On strategic partnerships; pilot line production underway; targeting humanoid, aviation, and EV applications; small scale vs. Korean / Chinese rivals limits near-term humanoid robot relevance.
催化剂
Tesla Optimus targeting 50,000+ unit production in 2026, concentrating direct volume demand on NCM 2170 cylindrical cells and forcing supply-chain scale-up across LG Energy Solution and other confirmed suppliers.
China's 2025 Humanoid Robot Action Plan targeting 100,000 deployed humanoids by 2027; domestic shipments projected >62,500 units in 2026 (700%+ YoY), pulling Chinese battery makers into robot-specific qualification.
China national all-solid-state battery standards (2026) plus EVE Energy Chengdu line and Samsung SDI InterBattery sample debut accelerating commercialization race—multiple vendors targeting 2027 small-scale production.
LG Energy Solution 4680 cylindrical cells entering robotics pipeline alongside BYD and CATL moving into humanoid OEM roles, blurring supplier/customer boundaries and expanding design-win surface area for cell partners.
Global embodied AI total market 2025 base → $23.06B by 2030 at 39% CAGR (MarketsandMarkets); software/VLA platform sub-segment alone estimated $2.1B in 2025 at 42% CAGR → $38.6B by 2034 (marketintelo.com); covers all geographies; includes hardware + software; not humanoid-only
市场规模 $4.44B
CAGR 39%
BOM -
瓶颈：Action-labeled training data is the primary binding constraint: no internet-scale source exists and every robot demonstration requires physical human teleoperation, driving high per-sample cost; cross-embodiment negative transfer is documented when pooling heterogeneous robot datasets. Secondary constraint: real-time inference overhead of full-Transformer VLA architectures at robot control frequencies (10-50 Hz) strains edge compute budgets, particularly on lower-cost humanoid platforms.
中国头部 (6)
AgiBot (智元机器人)未上市龙头Led global humanoid shipments at ~5,100 units; 'One Body, Three Intelligences' full-stack; GO-2 (ViLLA) VLA with Action Chain-of-Thought for long-horizon tasks; AGIBOT WORLD 2026 open heterogeneous dataset for embodied AI research
Spirit AI (千寻智能)未上市龙头Spirit v1.5 VLA ranked No.1 on RoboChallenge global benchmark (score 67.19), surpassing Physical Intelligence π0.5; ~¥3B (~$417M) raised in 30 days; strategic 'Universal Brain' partnership with Bosch China for OEM industrialization
Galbot (银河通用)未上市龙头GraspVLA: world's first end-to-end grasping model pretrained purely on synthetic data (billions of NVIDIA Isaac sim samples); ¥20B+ valuation; CATL-led Series B; designated embodied AI robot for 2026 CCTV Spring Festival Gala; inspected by President Xi Feb 2026
Galaxea AI (星海图)未上市挑战者$654M raised; G0.5 autoregressive VLA released Jun 2026 (SOTA on LIBERO benchmark); GalaxeaVLA open-sourced; R1 Pro/Lite robots secured thousand-unit-scale orders from automotive and logistics OEMs; $2.9B valuation after Apr 2026 Series B+
Unitree Robotics (宇树机器人)未上市挑战者UnifoLM-VLA-0 open-sourced Jan 2026 (Qwen2.5-VL-7B backbone, 12 manipulation task categories); world's top humanoid seller by volume; G1 at $16K; STAR Market IPO cleared Jun 2026 with ~$6.2B target valuation
Huawei CloudRobo未上市挑战者CloudRobo embodied intelligence platform supports cross-robot and cross-scenario VLA deployment; leverages Ascend AI chips and Huawei Cloud infra; positioned as sovereign alternative to NVIDIA Isaac for Chinese robot OEMs
国际头部 (8)
NVIDIA (Isaac / GR00T)NVDA龙头GR00T N1.7 open-reasoning VLA (commercial license, early access Jun 2026); Isaac Sim / Omniverse data engine generates 780K synthetic trajectories in 11 hours; runs on Agility Digit, Neura 4NE1, 1X NEO, and 10+ other robot OEMs
Physical Intelligence (π)未上市龙头$11B valuation; π0/π0.5 VLA with flow-matching action generation; $600M Series B; open-sourced π0 weights; de-facto international benchmark model against which Chinese VLA challengers measure themselves
Google DeepMind / IntrinsicGOOGL龙头Gemini Robotics-ER 1.6 reasoning-first VLA (Apr 2026); Intrinsic (acquired ROS/Open Robotics) merged into Google Feb 2026; design wins with Apptronik Apollo and Boston Dynamics Atlas announced Jan 2026
Skild AI未上市龙头$14B valuation, $1.83B raised; Skild Brain 'omni-bodied' model controls any robot form without prior embodiment knowledge; SoftBank-led Series C with NVIDIA, Samsung, Bosch as strategic investors; $30M+ ARR
Figure AI (Helix)未上市挑战者Helix-02 full-body VLA (Jan 2026) delivers 8-hour continuous autonomous warehouse shifts; first VLA to simultaneously control two robots on a shared long-horizon task; OpenAI investor relationship
BlackBerry QNXBB挑战者Safety RTOS powering NVIDIA IGX Thor safety layer in Agility Digit; ISO 26262-grade deterministic OS expanding from automotive (275M vehicles, $950M royalty backlog) to humanoid certifications; April 2026 partnership expansion to robotics
Neura Robotics未上市新兴Neuraverse OS enables fleet-wide skill sharing across robots in real time; 4NE1 humanoid built on full NVIDIA Isaac stack; CES 2026 debut of Neuraverse as shared learning platform
1X Technologies未上市新兴Redwood VLA trained on 10K+ hours real-world operational data powers NEO home robot ($20K); EQT deal for up to 10,000 NEO units in portfolio companies 2026-2030
催化剂
NVIDIA GR00T N1.7 commercial licensing (early access Jun 2026) enables first-generation production VLA deployments across the NVIDIA-partnered robot OEM ecosystem
China HEIS 2026 national humanoid standard mandates plug-and-play software/hardware interfaces and unified VLA definitions, accelerating cross-vendor OEM adoption and government-mandated 10,000-unit factory deployment by end-2026
World models entering the VLA stack as a standard inference layer (2026 trend) reduces dependence on physical demonstrations via scalable sim-to-real pipelines
Spirit AI–Bosch and Skild Brain OEM licensing deals establish platform software business models — analogous to mobile OS licensing — validating recurring-revenue paths for foundation-model providers
关联标的: NVDA · BB · GOOGL · MSFT · KOID
稀土永磁 conf: med
Total rare earth magnets market (NdFeB-dominant, ~71% share); $21.98B in 2025 to ~$30B by 2030 at 6.4% CAGR (MarketsandMarkets). Humanoid robot segment nascent but IDTechEx forecasts 7x robotics magnet volume demand by 2036 — not humanoid-specific. Global NdFeB volume ~385,000 MT/yr in 2025.
市场规模 $22B
CAGR 6.4%
BOM med (~5-8% est. of humanoid robot total BOM; NdFeB magnets are core to servo motors which account for ~20% of robot BOM, and magnets represent a meaningful fraction of each motor's cost)
瓶颈：China controls ~90% of global sintered NdFeB production. April 2025 MOFCOM export controls on heavy rare earths (Dy, Tb) and finished NdFeB magnets containing them require case-by-case export licenses with 45-day to multi-month lead times; EU spot prices for Dy/Tb peaked at 6x Chinese domestic levels in Apr-May 2025; Chinese NdFeB exports fell ~75% in the two months post-controls. Upstream concentration (Mountain Pass US, Bayan Obo CN) and lack of ex-China separation/magnet capacity limits diversification on any 1-3 year horizon.
中国头部 (6)
中科三环 (Zhongke Sanhuan)000970.SZ龙头 · 6.7%World's largest sintered NdFeB producer by volume (~30,000-35,000 MT/yr capacity); key EV + industrial robot supplier; Q1 2025 net profit +113% YoY; Tesla vehicle supply chain; share_pct is production volume share per Chinese analyst estimate, not independently verified.
金力永磁 (JL MAG Rare-Earth)300748.SZ龙头 · 5.6%Confirmed Tesla Optimus Tier-1 R&D supplier with small-batch deliveries since late 2024; robot-sector orders +120% in 9M 2025; net profit +157-179% YoY; Mexico magnetic-component factory planned (1M sets/yr capacity); also listed HK as 6680.HK.
正海磁材 (Zhenghai Magnetic Materials)300224.SZ龙头 · 4.1%Major EV traction-motor NdFeB supplier; targeting 30,000 MT/yr capacity by end-2026; strong Tier-1 auto OEM customer base; share_pct is production volume share per analyst estimate.
宁波韵升 (Ningbo Yunsheng)600366.SS龙头In mass delivery to domestic humanoid robot OEMs incl. Zhiyuan Robots since 2025 (~600 MT in 2025, targeting 3,000 MT in 2026); Baotou 15,000 MT expansion (first 5,000 MT in production, remainder by Jun 2026); low-altitude eVTOL magnets also qualifying.
大地熊 (Anhui Daidi Bear New Materials)688077.SS挑战者High-performance NdFeB focus (>70% high-spec grades); wind energy accounts for ~40% orders; 5,000 MT new capacity commissioned 2025; humanoid robot joint-motor application under qualification — Tesla Optimus supply unconfirmed by customer.
英洛华 (INLOA / Yingluo Huaye)000795.SZ挑战者Industrial robot magnet orders +50% in 2025; humanoid robot joint-motor magnetic components in sampling/qualification stage; national 'Little Giant' specialized enterprise; dual traction from EV and robotics verticals.
国际头部 (6)
TDK Corporation6762.T龙头Largest non-Chinese sintered NdFeB producer (~8,000-12,000 MT/yr); manufactures both NdFeB and ferrite grades; major automotive and consumer-electronics supply chain globally; FY2025 group revenue ~¥2.20T (~$14.6B).
信越化学 (Shin-Etsu Chemical)4063.T龙头NdFeB pioneer with 66 international facilities; Vietnam magnet plant (~2,000 MT/yr) + Fujian China alloy plant (3,000 MT/yr); building Japan domestic rare-earth smelter to cut China dependency; high-performance specialty grades for industrial applications.
Proterial Ltd. (formerly Hitachi Metals / 日立金属)未上市龙头Holds foundational sintered NdFeB patents; taken private by Bain Capital 2022; FY2025 revenue ~¥769B (~$5.2B); developed heavy-RE-free EV motor magnets (NMX-F1SH-HF) in 2025; exploring India manufacturing expansion to reduce China exposure.
MP MaterialsMP挑战者Only integrated US mine-to-magnet producer; Independence facility (Fort Worth TX, ~3,000 MT/yr NdFeB) in commercial ramp; 10X Northlake TX campus groundbroken Q1 2026 (targeting additional 7,000 MT/yr); GM and Apple confirmed anchor customers.
Vacuumschmelze (VAC)未上市挑战者Europe's sole significant NdFeB producer; facilities in Hanau (DE), Slovakia, Finland + new US plant opening ~2025; supplies Porsche, Audi, and EU defense/aerospace; Ucore rare-earth oxide supply MOU signed Nov 2025 to reduce China raw-material risk.
Daido Steel5471.T新兴Honda-partnered pioneer of hot-deformation NdFeB magnets requiring no heavy rare earths (Dy/Tb-free); strategically important as Dy/Tb supply choke point intensifies; specialty grades for EV and robotics motors.
催化剂
China export-license regime (April 2025) forcing accelerated ex-China supply chain build-out: MP Materials 10X Texas campus, VAC US plant opening, Proterial India magnet plant exploration, Energy Fuels HRE oxide qualification.
Tesla Optimus and humanoid robot production ramp targeting 10,000 units/month by mid-2026 — each unit requires ~3.5 kg high-spec NdFeB, creating step-change incremental demand of 3,400-34,000 MT at 1M-unit scale.
Heavy-rare-earth-free magnet commercialization (Daido Steel hot-deformation, Proterial NMX-F1SH-HF) could unlock supply constraints and expand addressable market if scaled to EV and robotics volume.
IDTechEx (2026-2036 report) projects robotics to overtake automotive as the single largest NdFeB demand segment by 2040, attracting institutional capital rerating for magnet producers with confirmed humanoid robot design wins.