The Robotics Supply Chain in China: A Field Map

Here is the map in one paragraph. If you are building a robot, the components come from Shenzhen and the Pearl River Delta – motors, drives, sensors, boards, and the fastest iteration loop on Earth. The tooling and volume production sit an hour away in Dongguan. The AGV and AMR chassis come from Suzhou and the Yangtze River Delta, where industrial automation has decades of roots. The embodied-AI research and the highest-end autonomy work concentrate in Beijing, around Tsinghua, Peking University, and state-backed labs. And Shanghai is where international partnerships actually get signed. Everything else in this essay is detail on that map – what each cluster makes, why it wins, what it costs, how long it takes, and how a Western team navigates a market that does not work the way Western markets do.

A word on who is drawing the map, because honesty about vantage point matters more in this topic than most. We are not a robotics company with decades of manufacturing history – we are a technology studio entering robotics, and we are doing it from inside China. Our founder is getting his MBA at Peking University's Guanghua school, founded a student club with 4,000+ members across Peking University and Tsinghua, and has run recruiting for San Francisco companies hiring Chinese engineers. This essay is the field map we assembled as we move into the robotics supply chain ourselves: the research we compiled, cross-checked against what operating on the ground has taught us about how business in China actually works. Where a claim is market data, we present it as market data. Where it is our first-hand observation, we say so. We will not dress either up as the other.

Why does the robotics supply chain run through China at all?

Because the demand, the components, and the policy all concentrated there at once. According to IFR World Robotics, 2025, China installed 295,000 industrial robots in 2024 – 54 percent of global deployments, more than the rest of the world combined. That installed base is not just a market; it is a training ground. Every deployed robot is a customer for actuators, reducers, sensors, and service, which is why the component ecosystem deepened fastest where the robots actually went to work.

Policy then poured accelerant on it. According to The Robot Report, 2023, China's Ministry of Industry and Information Technology set national targets for humanoid robots: mass-production readiness by 2025 and humanoids as "an important new engine of economic growth" by 2027. Municipal governments followed with billions in robotics funds, procurement programs, and innovation centers. Whatever you think of industrial policy, the practical consequence for a hardware buyer is simple: the deepest, fastest, cheapest robotics component ecosystem in the world is Chinese, and it is compounding.

There is one more structural fact worth holding in mind: China's EV industry built much of this before robotics needed it. The battery, motor, and power-electronics supply chain that made BYD the world's largest EV maker overlaps heavily with what a humanoid or mobile robot needs – actuators, gears, and precision motors often repurpose directly. And at the raw-material layer, China processes roughly 85–90 percent of the world's neodymium, the rare-earth element inside virtually every permanent-magnet motor. You can source a robot without touching China. You cannot source one without touching something China processed.

Which city makes what? The regional map

The clusters are real and they specialize hard. Choosing a supplier is still primarily about the company, not the postcode – but knowing the map tells you where to look first and what a given region will be genuinely good at.

Two notes the table cannot carry. First, the Shenzhen row understates the point: the Pearl River Delta is not a place where parts are available, it is a place where the distance between parts has collapsed. Component markets in Huaqiangbei and the factory floors of Bao'an now operate as one synchronized system – inventory and pricing signals move between them in close to real time, which is why the ecosystem absorbs supply shocks that would stall hubs elsewhere for weeks. Second, the map has satellites: Hangzhou hosts Unitree, the most famous name in low-cost humanoids and quadrupeds; Nanjing hosts Estun's vertically integrated industrial arms; Shenyang hosts Siasun, the Chinese Academy of Sciences spin-off that builds entire smart-factory systems. The clusters anchor the map; the champions are scattered across it.

What does the component landscape actually look like?

Break a robot into its bill of materials and the sourcing story gets concrete. Five component families cover most of what a robotics team buys, and each has a distinct Chinese landscape.

• Actuators and motors. The deepest category, riding on EV-industry scale. Inovance – originally a servo-motor giant – shipped over five million robotic joint servo motors in 2025 and now builds force-control assembly robots with BYD. Unitree manufactures its own motors and reducers in-house, which is how its G1 humanoid starts at $13,500 and its R1 launched at $5,900. For anything that flies, T-Motor, SunnySky, and Hobbywing dominate brushless motors and ESCs globally. The ecosystem's rule of thumb: motors and servos run roughly 3x cheaper than Western equivalents.
• Reducers and harmonic drives. The precision bottleneck of every arm and humanoid joint – and a category where Chinese suppliers have closed most of the gap with the Japanese incumbents at a fraction of the price. Harmonic drives and strain wave gears are sourceable within the Shenzhen 100 km radius, with first samples often delivered within a week; the vertically integrated players (Unitree, Estun) increasingly make their own.
• Sensors. Depth cameras, IMUs, force-torque sensors, LiDAR, and tactile arrays are all Pearl River Delta strengths, reinforced by the drone industry's decade of demand. The frontier is dexterity: LinkerBot manufactures around 5,000 dexterous hands per month across five factories and supplies Siemens, Foxconn, and Samsung; Inspire Robots delivered over 10,000 dexterous hands in 2025 on the strength of its micro linear servo actuators. That an entire sub-industry exists for robotic hands tells you how deep the specialization runs.
• Controllers and compute. Custom PCBs fabricate in 24–72 hours in Shenzhen; assembled boards ship in under 10 days. Motion controllers and servo drives are commodity-deep. The honest caveat: the highest-end AI compute in Chinese robots is still largely imported or constrained, which is one reason the smart move for a Western team is to source the body in China and keep the brain – compute choices, models, and control software – at home.
• Chassis and structure. CNC-machined aluminum frames and injection-molded housings are Shenzhen-and-Dongguan bread and butter, with aluminum soft tooling in 10–14 days. For wheeled platforms, the Suzhou/YRD cluster is the world's AGV/AMR chassis workshop – standardized, quality-systemed, built for batch delivery rather than one-off wow.

How much cheaper and faster is it, really?

The numbers are the argument, so here they are plainly. A robotic arm prototype that costs $15,000–$25,000 and takes 10–14 weeks in the United States or Germany is produced in Shenzhen for $4,000–$8,000 in 10–14 days. A full iteration spin – physical prototype in hand, CAD and PCB revisions, new samples delivered – takes 10–14 days against roughly 6 weeks in the West. At the system level, industry analyses put the raw bill of materials for a humanoid robot at about $130,000 on a non-Chinese supply chain versus roughly $46,000 on a Chinese one. Chinese OEMs are targeting $10,000 for entry humanoids while Western competitors target $20,000–$30,000, and of the fourteen manufacturers worldwide selling robotic arms under $10,000, eight are Chinese.

Notice what kind of advantage this is. It is not a discount; it is a clock speed. If your competitor completes an iteration loop in two weeks and you complete one in six, they do not beat you by being 30 percent better – they beat you by learning three times as fast, compounding, for as long as you both stay in the game. This is the mechanism behind Chinese humanoid companies shipping new generations in months rather than years, and it is the single strongest reason a Western robotics team engages with this supply chain even if final assembly happens elsewhere.

Why is the middleman layer the actual interface?

Everything to this point is legible from a desk: clusters, components, prices, lead times. What the spreadsheets do not show is the part we have learned by operating here, and it is the part that decides whether a Western team gets the advantages above or bounces off the market entirely. China is simply a different market in terms of culture – business does not work the way it works in the US. In our experience on the ground, three observations matter most.

First: you cannot enter randomly. It is genuinely difficult to show up without connections and start doing business – you have to know people. Relationships gate access in ways that have no Western equivalent: which factory takes your call seriously, which quote reflects a real price, whose production slot you actually get in a busy month. Western founders hit a ceiling here without relationships, and no amount of budget substitutes for them. The upside, also from experience: people are more straightforward than the stereotype suggests. Once you are inside a relationship, communication tends to be direct. The gate is getting inside.

Second: China is a middleman culture – and the middlemen make everything easier. Much easier. The Western instinct is to disintermediate: go factory-direct, cut out the agent, save the margin. In China's hardware ecosystem that instinct is exactly backwards. The sourcing agent, the design-for-manufacturing partner, the bilingual broker who has spent ten years walking factory floors – these are not friction, they are the interface. They carry the relationships you do not have, they know which of five near-identical listings is a real factory and which is a trading company, and they translate not just language but expectations: tolerances, quality norms, what a deadline means. Our strong advice is to budget for the intermediary as infrastructure, the way you budget for cloud hosting – a percentage that buys you access, error-correction, and speed you cannot buy any other way.

Third: face-to-face still governs. People here prefer in-person meetings, and the preference has teeth. Deals move after dinners, factory visits change how you are quoted, and a partner who has met you allocates you real capacity when things get tight. The practical logistics are more manageable than most Westerners expect – even payments have smoothed out; Alipay now works reasonably well for foreigners, WeChat Pay somewhat less so – but the principle stands: presence is a signal of seriousness, and the market prices it. We cover the broader operating layer – culture, market entry, and the relationship ceiling – in our companion essay on doing business in China as a Western tech founder; here the point is narrower: in the robotics supply chain, the relationship layer is the sourcing stack.

How do you find suppliers without getting burned?

The failure pattern is consistent enough to name: transactional sourcing. A team picks the cheapest search listing, wires a deposit, skips the visit, and discovers months later that the "factory" was a trading company, the sample line was not the production line, or the quote quietly excluded tooling. Every element of the playbook below exists to break that pattern.

1. Start remote, start small. Run your first boards and machined parts through the online prototyping services. You learn the quality baseline and the communication rhythm with zero relationship risk, and you arrive at later conversations with real artifacts instead of slideware.
2. Source suppliers through referrals, not rankings. The good factories are busy and do not optimize their listings; the aggressive listings are often resellers. Referrals through an intermediary, an industry contact, or an existing customer are worth more than any directory – this is the middleman layer doing its job.
3. Visit before tooling. The point where money becomes hard to recover is the tooling commitment. Before it, someone on your side – you or your agent – walks the floor: does the production line match the samples, is the QC bench real, do the people who answered your emails actually work there.
4. Test with paid pilot orders. A small paid order reveals more than any audit: actual lead time, actual tolerances, how the supplier handles a defect. Escalate volume only as trust is demonstrated, not promised.
5. Verify locally before you certify globally. Pre-compliance EMC and safety scans at Shenzhen-area labs cost a fraction of a failed certification round later, and a supplier's willingness to support them is itself a signal.

What about IP risk when you prototype in Shenzhen?

The risk is real, the paranoia is usually miscalibrated, and the practical answer is structure rather than avoidance. Four practices cover most of the exposure.

• Keep the brain at home. Source the body – motors, frames, boards, sensors – where the body is cheapest and fastest. Keep the control stack that makes your robot your robot – firmware, models, calibration, integration software – developed and held by your own team. A supplier who builds your chassis has your chassis; without the stack it is furniture.
• Split the design across vendors. Mechanical here, electronics there, final assembly somewhere else – so no single supplier holds the complete picture. This is standard practice in the ecosystem, adds modest coordination cost, and is the cheapest structural protection available.
• Use agreements that work in Chinese courts. A US-style NDA is close to worthless here. The instrument that has teeth is the NNN agreement – non-disclosure, non-use, non-circumvention – drafted under Chinese law, in Chinese, enforceable where your counterparty's assets are. Put tooling ownership in writing in the same pass, so the molds you paid for cannot quietly anchor someone else's product line.
• Treat speed as the deepest moat. In an ecosystem where a design iteration takes two weeks, anything you ship is copyable eventually – by someone, from somewhere. The teams that stay ahead are the ones that use the same iteration speed advantage the copiers would: by the time a clone lands, you are two generations on. Legal structure buys you time; velocity is what you spend it on.

How would we approach it? The sequence

Strip this essay to the operating sequence we are following ourselves as we move into robotics – the same sequence we would hand any Western team entering the Chinese supply chain.

1. Prototype remotely first. Boards in 72 hours, assembled PCBA in under 10 days, quick-turn CNC – all without leaving your desk. Iterate until the design is worth a relationship.
2. Engage the intermediary before the factory. Find the bilingual sourcing partner through referrals and let them run the supplier funnel. In a middleman culture, hiring the middleman early is not overhead – it is the single highest-leverage decision on the list.
3. Show up before tooling. Walk Huaqiangbei to calibrate on components; walk any factory that will cut steel for you. Face-to-face is how the market prices seriousness, and the tooling commitment is where seriousness starts costing real money.
4. Run IP hygiene from day one. Brain at home, design split across vendors, NNN agreements under Chinese law, tooling ownership in writing. Cheap when done early, nearly impossible to retrofit.
5. Model landed cost before volume. Apply the current duty stack and freight to the ex-factory price and decide – with a broker, against current rules – where final assembly should live. The answer may be China, and it may not be; the prototyping answer is China either way.
6. Match region to component. Shenzhen for electronics, sensors, and speed; Dongguan for tooling and volume; Suzhou/YRD for wheeled platforms and batch discipline; Beijing and Shanghai for research partnerships and internationally-facing deals.

The robotics supply chain in China is not a secret, but it is also not a website – it is a geography, a component landscape, and above all a relationship layer, and the teams that win engage all three. The cost and speed numbers are structural and verifiable; the navigation layer is cultural and earned; and the honest position for any Western team – ours included – is that of a well-prepared entrant, not a veteran. That preparation is what we do for a living: mapping a hard technical landscape, building the system, and shipping – the same implementation-studio discipline we apply across our services, now pointed at hardware. If you are a robotics founder or a hardware buyer weighing your first move into this supply chain, start with the map in this essay – and if you want a partner who is already on the ground, we are easy to find.