China’s Embodied AI Supply Chain: Who Makes the Brains, the Brawn, and the Bones — and Why It Matters for Your Business
China shipped 97% of the world’s humanoid robots in the first half of 2026. But behind that staggering number lies a complex, multi‑layered supply chain that global buyers, investors, and compliance officers urgently need to understand.
Who actually makes the brains (the AI models that enable perception and decision‑making), the brawn (the motors and actuators that produce movement), and the bones (the precision components that hold it all together)? More importantly, how reliable, cost‑effective, and legally sound are these suppliers?
This article breaks down China’s embodied AI supply chain into three critical layers — algorithms (the brain), core components (the brawn and bones), and data infrastructure (the fuel) — and explains what every global business needs to know before signing a contract.
1. The Supply Chain at a Glance
China’s embodied AI supply chain is massive, vertically integrated, and increasingly self‑sufficient — but not uniformly so.
According to data from Qixinbao and the China Academy of Information and Communications Technology (CAICT), China’s embodied AI industrial chain comprises more than 8,000 upstream component suppliers, roughly 1,900 midstream robot manufacturers and system integrators, and about 2,370 downstream application companies. The upstream segment alone accounts for over 70% of the total industrial chain by enterprise count. Geographically, the sector is concentrated in Guangdong, Zhejiang, Jiangsu, Shanghai, and Beijing, which together account for more than two‑thirds of all embodied AI companies in China.
📈 The Big Picture: China has built the world’s most comprehensive embodied AI supply chain — from raw materials to finished robots. But self‑sufficiency varies dramatically by component. Some parts are world‑class and cost‑competitive; others remain critically dependent on foreign suppliers. Understanding this nuance is the difference between a successful partnership and a costly mistake.
2. The Brain: Algorithms and AI Models
The “brain” of an embodied AI system is the software stack that enables perception, reasoning, planning, and control. This is where the U.S. and China have taken fundamentally different paths — and where the competitive dynamics are shifting fastest.
2.1 The U.S.-China Divide
The global embodied AI algorithm landscape is often described as “U.S. leads foundational research, China leads applied deployment.” American companies and research institutions — NVIDIA, Google DeepMind, Physical Intelligence, Figure AI, and Tesla — continue to dominate in foundation model originality, simulation ecosystems, and general‑purpose reasoning.
Chinese companies, by contrast, have focused on vertical application, real‑world data scale, and hardware‑software co‑optimization. The result: Chinese embodied AI models may not always match U.S. counterparts in benchmark scores, but they work better in actual Chinese factory and service environments — precisely because they’re trained on real Chinese operational data.
As one industry analysis put it: the U.S. follows a “brain‑first” path — perfecting general intelligence first, then finding hardware to run it on. China follows a “body‑first” path — deploying robots at scale in real environments, accumulating operational data, and upgrading the brain through continuous real‑world feedback.
2.2 Leading Chinese “Brain” Developers
Several Chinese companies have emerged as leaders in embodied AI algorithms, each with a distinct technical approach:
智平方 (Zhipingfang / AI² Robotics) Brain‑Like VLA
Core Technology: Zhipingfang’s AlphaBrain series is the world’s only embodied AI model family that has completed the full evolution from “end‑to‑end VLA → enhanced VLA → brain‑like VLA.” In 2026, the company released NeuroVLA, which introduces a “cortex‑cerebellum‑spinal cord” three‑layer bio‑inspired architecture into robot control.
Key Metrics: NeuroVLA achieves 20‑millisecond safety reflexes, an 80.2% reduction in jerk peaks, and just 0.4 watts of spinal‑layer power consumption. The architecture enables robots to learn and make decisions with relatively little data, breaking the traditional reliance on massive datasets and compute.
Deployment: Zhipingfang has deployed its AlphaBot series in automotive manufacturing, semiconductor fabrication, biopharmaceuticals, and other industrial settings.
银河通用 (Galaxy General / Galbot) World‑Action Model
Core Technology: Galaxy General’s AstraBrain is built on the World‑Action Model (WAM) architecture, which the company’s team first proposed at ICCV 2025. WAM unifies VLA (Vision‑Language‑Action) generation with world‑model environmental reasoning — allowing robots to predict the physical consequences of their actions before executing them.
Key Differentiator: AstraBrain WAM 0.5 is the first model to effectively unify multiple data sources — including simulation, real‑world, and human‑machine mixed data — into a single training framework.
Deployment: Galaxy General has deployed its robots in pharmacies (100+ locations), retail stores, and industrial settings, with a 99.95% order accuracy rate in pharmacy operations.
智元机器人 (ZhiYuan / AGIBOT) Full‑Stack Ecosystem
Core Technology: ZhiYuan has built a full‑stack embodied AI platform encompassing hardware, software, and cloud services. The company’s approach emphasizes ecosystem building and RaaS (Robot‑as‑a‑Service) models. In June 2026, ZhiYuan launched the “YuanSheng Ecosystem Plan,” committing ¥2 billion over five years to foster innovation and partnerships.
Key Differentiator: ZhiYuan’s advantage lies in its scale and ecosystem — not just in any single algorithm. The company has the largest real‑world deployment footprint, which generates the most operational data for continuous model improvement.
Other notable players in China’s algorithm space include 自变量机器人 (ZBL Robotics), which focuses on end‑to‑end embodied AI models, and 星海图 (Xinghaitu), which has seen its valuation double to over ¥20 billion in the first half of 2026.
⚠ Due Diligence Note for Buyers: When evaluating a Chinese embodied AI company’s “brain” capabilities, ask: What real‑world data was this model trained on? Models trained on proprietary, real operational data from Chinese factories or service environments are far more likely to perform reliably than models trained primarily on synthetic or open‑source data. Verify the company’s data collection infrastructure and deployment track record — not just its academic publications.
3. The Brawn and Bones: Core Hardware Components
This is where China’s manufacturing superpower truly shines — and where the greatest supply chain risks also lie.
The core components of an embodied AI robot can be broken down into several key categories. According to industry data, the joint module (motor + reducer + driver + sensor) accounts for approximately 50% of total hardware cost. A more detailed breakdown of the Bill of Materials (BOM) for a typical humanoid robot shows: planetary roller screws (~22%), frameless torque motors (~18%), harmonic reducers (~12%), six‑axis force/torque sensors (~10%), and AI compute chips and mainboards (~15%).
Here’s how each critical component category is shaping up in 2026:
3.1 Harmonic Reducers — The Joints
Function: Harmonic reducers are precision gears that enable compact, high‑torque, high‑precision rotational motion — essentially the “joints” of a robot. They are the single most critical mechanical component in humanoid robots.
Global Landscape: The global harmonic reducer market is a duopoly dominated by Japan’s HDSI (Harmonic Drive Systems) and China’s Leaderdrive (绿的谐波), which together hold over 63% of global revenue market share. HDSI still leads in the ultra‑precision, long‑life segment, but Leaderdrive has successfully ascended to the global top tier.
China’s Position: Leaderdrive, founded in 2011, is now China’s largest harmonic reducer manufacturer. According to J.P. Morgan, Leaderdrive’s share of the Chinese harmonic reducer market is between 30–40%, while other sources report it holds over 60% of the domestic market. The company serves more than 1,800 global clients, including UBTECH and ZhiYuan in China, and has fulfilled small‑batch R&D orders from Tesla and Figure AI.
Localization Status: China’s harmonic reducer localization rate has reached 60–70%, with Leaderdrive accounting for approximately 25% of the global market. However, high‑end, long‑life harmonic reducers still rely heavily on Japanese suppliers — a critical vulnerability for high‑duty‑cycle industrial applications.
绿的谐波 (Leaderdrive) China’s Harmonic Reducer Leader
Market Position: Leaderdrive is China’s largest harmonic reducer manufacturer and the only Chinese company that has successfully broken into the global top tier alongside Japan’s HDSI.
Key Customers: UBTECH, ZhiYuan, and has fulfilled small‑batch R&D orders from Tesla and Figure AI.
Recent Developments: In February 2026, Leaderdrive formed a joint venture with Taiwanese auto‑parts supplier Minth Group to develop joint modules for humanoids in the U.S. market — positioning itself to serve Tesla, Figure AI, and Boston Dynamics. The company’s founders have become billionaires on the back of surging demand for humanoid robot components.
3.2 Frameless Torque Motors — The Muscles
Function: Frameless torque motors are the “muscles” that convert electrical signals into rotational force. They are direct‑drive motors that integrate directly into robot joints, offering high torque density, lightweight design, and excellent control precision.
China’s Position: China has made rapid progress in this category, driven by the spillover effect from the electric vehicle supply chain. The localization rate for frameless torque motors and hollow‑cup motors has exceeded 85%.
Key Players: Nearly 20 Chinese motor manufacturers are racing to capture market share. Leadshine Technology (雷赛智能) delivered over 120,000 frameless torque motor units in 2025, representing more than 20× year‑over‑year growth. Other notable suppliers include Step Electronics (步科), HaoZhi (昊志), and Zhongte Technology (中特科技), which is building the first fully automated frameless torque motor production line for humanoid robots.
Trend: The market is shifting toward higher torque density, higher power‑to‑weight ratio, and greater customization. The global frameless torque motor market is projected to exceed 12.8 million units in 2026, with the 1.5–12 N·m mid‑to‑high torque segment accounting for 54% of total volume.
3.3 Planetary Roller Screws — The Linear Actuators
Function: Planetary roller screws convert rotational motion into linear motion with high precision and load capacity. They are essential for linear joints — particularly in the legs and arms of humanoid robots.
China’s Position: This is an area of accelerating catch‑up. The localization rate for planetary roller screws was only under 12% in 2024, but is projected to reach 28.7% in 2026. High‑precision (C3 grade and above) products still rely heavily on imports, with ~63% of high‑end products imported.
Key Players: Traditional automotive component suppliers are leveraging their manufacturing expertise to enter this space. Companies like Hengli Hydraulic (恒立液压), Wuzhou Xinchun (五洲新春), and Shuanglin Group (双林集团) — which has mass‑produced a planetary roller screw joint component priced at just ¥499 (~$71) — are making significant inroads. State‑owned enterprises like China State Shipbuilding Corporation’s 704 Research Institute have also made breakthroughs, with their independently developed high‑reliability planetary roller screw selected as one of the top ten “central SOE robot innovation achievements” at WRC 2026.
Cost Trend: The average cost of a planetary roller screw has fallen to approximately ¥1,850, and is expected to decline further as volume increases.
3.4 Force/Torque Sensors — The Sense of Touch
Function: Six‑axis force/torque sensors measure forces and torques in all three dimensions, enabling robots to perform precise, force‑sensitive tasks like assembly, polishing, and human‑robot interaction.
China’s Position: This is the most challenging component category for domestic suppliers. The localization rate for six‑axis force sensors is still under 30%, making it the least self‑sufficient of all major component categories.
Key Players: Despite the overall low localization rate, some Chinese companies have achieved dominant positions in specific segments. Landpoint (蓝点触控) holds 72.6% of the Chinese humanoid robot six‑axis force sensor market, with ZhiYuan, Galaxy General, Xiaomi Robotics, UBTECH, Zhongqing Robotics, and Xinghaitu all using its products in volume. Kunwei Technology (坤维科技) leads the broader Chinese intelligent robotics force sensor market with a 53% market share.
Trend: The Chinese six‑axis force sensor market is growing rapidly, with domestic manufacturers’ market share reaching 58.8% in 2026 — the first time domestic suppliers have surpassed foreign competitors in this category. The overall market is projected to reach ¥1.25 billion in 2026, with over 40% year‑over‑year growth.
3.5 The Integrated Joint Module Trend
Perhaps the most significant trend in China’s component sector is the move toward integrated joint modules — pre‑assembled units that combine motor, reducer, driver, and sensor into a single, ready‑to‑install package.
This approach reduces system cost, improves reliability, and simplifies robot assembly. At CITE 2026, it was noted that while most robot manufacturers (like Unitree and Zhongqing) still use a “self‑designed + externally sourced components + in‑house assembly” model, companies like ZhiYuan and Songyan are directly purchasing integrated joint modules — signaling a clear trend toward “一体化” (integrated) solutions.
However, integrated joint modules still face engineering challenges in areas like thermal management, electromagnetic compatibility (EMC), and wear resistance.
⚠ Due Diligence Note for Buyers: When sourcing components from China, verify the supplier’s quality control systems, production capacity, and intellectual property position. The rapid pace of localization means that supplier capabilities can change dramatically in 12–18 months. A supplier that was “second‑tier” last year may be a market leader this year — and vice versa. Always verify current certifications, customer references, and financial health before committing to large orders.
4. The Fuel: Data Infrastructure
As we explored in a previous article, data is the critical bottleneck for embodied AI. China’s unique advantage in this area is not just its manufacturing scale — it’s its ability to generate real‑world operational data at an unprecedented rate.
Key Data Points:
- The global stock of high‑quality real‑world physical interaction data is only in the hundreds of thousands to low millions of hours. Training a truly general‑purpose embodied AI model requires at least 10 million hours.
- China is addressing this gap through government‑led initiatives like the “Real‑Scene Training” program, which aims to generate millions of hours of high‑quality real‑machine data.
- Private companies are also building data infrastructure. 51WORLD (五一视界), a Hong Kong‑listed company, has launched AperData, an “embodied data infrastructure” platform that provides end‑to‑end services from real‑world data collection to trainable datasets.
- Guanglun Intelligence (光轮智能) has announced a “5‑year, ¥10 billion embodied AI data co‑construction plan” and released a 100,000‑hour full‑modality human behavior open‑source dataset.
- Companies like ZhiYuan have open‑sourced real‑scene datasets like “agibot world 2026,” covering commercial and home environments with structured, finely annotated data.
For global buyers, this means that Chinese embodied AI companies have access to a volume and diversity of real‑world training data that no other country can match. This is a structural advantage that will likely widen over time as more robots are deployed and more data is collected.
5. What This Means for Your Business
If you’re a global manufacturing executive, supply chain leader, investor, or legal counsel, here are the key takeaways:
✅ Key Takeaways:
- China’s supply chain is mature — but not uniform. Some components (frameless motors, harmonic reducers) are world‑class and cost‑competitive. Others (high‑end force sensors, precision roller screws) still rely heavily on imports. Know the difference.
- Localization is accelerating fast. The 75%+ overall localization rate is rising, but the pace varies by component. What was true six months ago may already be outdated.
- Integrated joint modules are the future. Suppliers that can deliver ready‑to‑install, validated joint modules will have a significant competitive advantage.
- Data is the new differentiator. Chinese companies’ access to real‑world operational data is a structural advantage that will shape the competitive landscape for years to come.
- Due diligence is non‑negotiable. The rapid pace of change means that verified, up‑to‑date intelligence on Chinese suppliers is essential. Don’t rely on marketing claims — verify the facts.
6. How ChinaBizInsight Supports Your Due Diligence
At ChinaBizInsight, we understand that verifying the claims of potential Chinese supply chain partners is essential — and challenging. Language barriers, limited access to authoritative data sources, and the rapid pace of change make it difficult to separate real capability from marketing hype.
We provide verified, authoritative intelligence on Chinese embodied AI suppliers — exactly the kind of data you need to make informed sourcing, investment, and partnership decisions:
- Official Enterprise Credit Reports — directly from China’s National Enterprise Credit Information Publicity System, with official watermarks and seals. Verify a supplier’s legal status, business scope, and regulatory compliance.
- Custom Due Diligence Reports — combining multiple authoritative sources to provide a 360‑degree view of a supplier’s financial, legal, and operational status.
- Executive Background Checks — detailed reports on directors, supervisors, and senior management, including their investment history, external positions, and risk records.
- IP Searches — trademark, patent, and copyright verification to ensure your supplier actually owns the technology they claim.
- Document Legalization & Apostille — ensuring that Chinese corporate documents are recognized internationally for contracts, compliance, and legal proceedings.
All our data is traceable to official sources, presented in English, and delivered with the speed and professionalism that global businesses demand.
🔎 Know your Chinese supply chain partners before you commit.
Get the verified intelligence you need to make confident sourcing and investment decisions in China’s embodied AI sector.
Contact ChinaBizInsight →📍 Based in China · Serving the world · Know your Chinese partners
Sources:
China Academy of Information and Communications Technology (CAICT), Q1 2026 Monitoring Data
Qixinbao Industry Data, Embodied AI Industrial Chain Report (2026)
J.P. Morgan, Humanoid Robot Component Market Analysis (March 2026)
IIMIGTU, Global Humanoid Robot Core Components and Actuators Industry Report (2026)
Big‑Bit Research, CITE 2026 Humanoid Robot Joint Module Analysis (May 2026)
Forbes, Brothers Ride Humanoid Robot Wave To Become China’s New Billionaires (April 2026)
Various industry reports: Pandaily, CCTV, DoNews, 21jingji, ce.cn
EO Intelligence, 2026 Global Embodied AI Commercialization Report (2026)
ChinaBizInsight
Your strategic bridge to transparent business in China.