Why Is China Ahead in Robotics? Atoms, Not Algorithms

Let me hand the sceptics their strongest argument first, because it's a good one. The entire global humanoid market last year came to somewhere in the low tens of thousands of units. That isn't an industry. It's a pilot scheme with a marketing budget, and a fair share of what you see in the demo videos is choreography that took several takes to land. So who cares which country sells nine in ten of a product almost nobody has bought yet? The clever part, the argument goes, still lives in the models, and the models are a fair fight.

That's right about the humanoids. It's wrong about everything underneath them.

China is ahead in robotics because robotics is an atoms problem, not a software one, and China owns the atoms. Roughly 90% of the humanoid robots sold worldwide in 2025 were Chinese, with six of the top sellers based in China, by Rest of World's count. And in Stanford HAI's 2026 AI Index, China installed 295,000 industrial robots in 2024 against 34,200 in the US, about nine times more. The humanoids are the shop window; the industrial installs are the shop. Neither lead rests on a clever model anyone can copy. They rest on a manufacturing base, a shared component supply chain, low unit costs, a state strategy, and a domestic willingness to put machines on real floors doing real work. This is the front of the AI race the West prices most cheaply, and the one the US will find hardest to close.

Key Takeaways- China's robotics lead is physical, not software-based. Around 90% of humanoids sold globally in 2025 were Chinese, by Rest of World's count.- On industrial robots, China installed roughly 9× more than the US: 295,000 against 34,200 in 2024, in Stanford HAI's AI Index.- The lead is hard to copy because it's built on a manufacturing base, cheap components and deployment appetite, not model IP. The US still holds the model frontier, and factories have never needed it.- China's EV giants, Xpeng and GAC, are pivoting into humanoids on the same batteries, motors and sensors that built their cars, and Xiaomi has begun factory trials of its own.- A national embodied-AI strategy backs the whole thing, with deployment rather than research as the target, as MERICS lays out.

Why is China ahead in robotics? The numbers

The gap isn't marginal, it's lopsided. About 90% of the humanoid robots sold globally in 2025 were Chinese, with six of the top-selling makers based in China and total volumes reported in the rough range of 13,000 to 18,000 units, by Rest of World's count. Two of those makers, Unitree and Deep Robotics, are here in Hangzhou, a short ride from my apartment. I don't need a market report to tell me the sector is real. I can see who they're hiring, and how fast.

The industrial picture is starker, and it matters more to the economy. China installed 295,000 industrial robots in 2024, against 34,200 in the US, roughly a 9× gap in Stanford HAI's AI Index. Humanoids get the magazine covers. The industrial install base is where the compounding actually happens, on factory floors nobody films.

Take the humanoid unit counts as reported ranges, not precise figures. The market is young, the accounting is loose, and the numbers will move. The direction won't. Whether the true figure is 13,000 units or 18,000, China's share sits around nine in ten, and no other country is anywhere near it.

Here's the part most coverage skips. The humanoid number is the flashy one, but the 9× industrial gap is the one that compounds. Every installed industrial robot is a data-generating, process-refining asset that lowers the cost of installing the next one. A 9× annual install rate does not stay at 9×. It widens, because the side deploying faster is also the side learning faster about what is worth deploying at all. That learning never shows up as a ranking, which is why it gets missed.

Everything here sits inside a bigger argument I make in the pillar piece on whether China is winning the AI race, which is really a question about layers rather than countries.

Sources: Stanford HAI, AI Index 2026 (industrial robot installs, 2024); Rest of World (humanoid share, 2025). Humanoid unit volumes are reported in the rough range of 13,000 to 18,000 units.

Around 90% of humanoid robots sold globally in 2025 were Chinese, with six of the top sellers based in China and roughly 13,000 to 18,000 units shipped, by Rest of World's count. In 2024 China installed 295,000 industrial robots against 34,200 in the US, about 9× more, in Stanford HAI's AI Index. China's robotics lead is volume and deployment, not a single product.

How does the EV playbook explain the robot lead?

The robot boom is China's electric-vehicle playbook, run again on a new product. The EV giants, including Xpeng and GAC, are pivoting directly into humanoid robots, as MIT Technology Review reported, and Xiaomi has begun factory trials of its own self-developed humanoids. These are companies that already know how to mass-produce complicated, sensor-laden machines cheaply. In supply-chain terms, a humanoid is a strange-shaped car.

The overlap sits in the parts list, and it runs deeper than it looks from outside. EVs and humanoids draw on the same core components: batteries, electric motors, actuators, sensors, power electronics. A country that built the world's dominant battery and motor supply chain for cars already owns most of the bill of materials for a robot. The hard, capital-heavy, unglamorous work was finished years ago, for a completely different reason.

Xpeng shows how far the convergence runs. The carmaker is targeting both robotaxis and humanoid robots built on its own in-house Turing AI chips, and in May 2026 it rolled out China's first mass-produced robotaxi, with pilot commercial operations slated for the second half of the year. The same sensing and silicon stack drives the car, the robotaxi and the robot, out of one R&D base.

When I walk UK boards through this, the EV framing is the thing that lands. People arrive assuming humanoids are a sci-fi moonshot, something you fund the way you fund fusion. In China they're an adjacency play: take a mature, cost-optimised car supply chain and point it at a new form factor. That's a far cheaper, far lower-risk path than building a robotics industry from nothing, and it explains the speed the incumbents are moving at.

China's EV giants, including Xpeng and GAC, are pivoting into humanoid robots, reusing the same batteries, motors, sensors and power electronics that built their cars, as MIT Technology Review reported, and Xiaomi has begun factory trials of its own. Xpeng is building robotaxis and humanoids on its own AI chips, with its first mass-produced robotaxi rolling off the line in May 2026. The car supply chain is the robot supply chain.

Why is China's robotics lead so hard to copy?

Because you can't copy a supply chain the way you copy a model. A US lab can match a Chinese model's benchmark within a quarter, given the compute and the people. It cannot conjure a domestic actuator industry or a million-unit assembly line in the same window, whatever it spends. China sold around 90% of 2025's humanoids by owning the components and the cheap assembly base, not by writing better code.

It helps to be clear about what the model lead is actually worth here. The frontier itself still sits in the US: on Epoch AI's capability index, US models have led Chinese ones by months rather than years since 2023, a mean gap of about seven months. But a factory has never needed the world's best model. It needs a cheap, reliable one running on a thousand machines by Thursday. AI pays for itself when it runs against real workloads a billion times, not when it tops a leaderboard once, and China has organised its robotics push around the first of those.

Component cost is the quiet decider. A humanoid is a bundle of expensive parts: harmonic reducers, actuators, force sensors, batteries. Make those parts domestically at volume and the unit cost of the finished robot collapses. Cheaper robots mean more get bought, which means more get made, which pushes component costs down again. Whoever started with the factories starts the loop already spinning.

Then there's deployment appetite, which is the piece Western strategists underrate most. A robot only gets better once it's deployed, run, and refined against real work with real tolerances. The side that puts more machines into more factories gathers more operational data, and it gathers it sooner. That 9× install lead is a learning lead too.

This is the atoms-versus-bits distinction the AI-race conversation keeps fumbling. Software advantages decay quickly, because code and model weights diffuse, leak, and get reverse-engineered by people who are very good at reverse-engineering. Manufacturing and deployment advantages compound instead, because they live in physical capacity that takes years and tens of billions to build. China's robotics lead sits almost entirely in the layer that doesn't decay. My honest read is that it's the least reversible lead in the whole race, and I say that as someone who spends most of his working life on the software side of it.

Which makes the reading an uncomfortable one for the US. The model gap is closeable in months. The factory gap is a decade-scale project that has to be funded before it can even be started, and the clock has been running for a while.

China's robotics lead is hard to copy because it lives in atoms, not bits. It rests on a domestic component supply chain that drives unit costs down, a 9× advantage in annual industrial-robot installs over the US in Stanford HAI's AI Index, and an appetite to actually deploy at scale. The US keeps the model frontier, by Epoch AI's measure, but deployment is where the value accrues. Software gaps close in months. Manufacturing and deployment gaps take years and tens of billions to close.

What is China's state strategy for robots?

The lead isn't only market-driven. It's state-backed, and nobody here is being coy about it. China has made embodied AI a national strategic priority, with policy frameworks often grouped under the "Robot+" banner that push robots into manufacturing, logistics, healthcare and elderly care, as MERICS details. Robotics is being treated the way solar and EVs were treated: as an industry to win, not one to enter.

The playbook is familiar because it has already worked twice. Pick a strategic sector, coordinate the funding and the standards, guarantee domestic demand, then let scale crush the cost curve. China ran that on solar panels and on EV battery cells until it dominated both. Robotics is the same machine pointed at a new target, with embodied AI written on the label.

Demand-side pull is the part Western markets simply don't have. When the state and state-linked buyers commit to deploying robots across factories and public services, a manufacturer is scaling against a market it can count on. That certainty takes the risk out of building capacity, and building capacity is what lets unit costs fall fast enough to win the export fight later.

You can already see where the coordination leads. Xiaomi runs a smartphone plant near Beijing with 100% of its key processes automated, and Gree operates a lights-out facility in Zhuhai that runs with barely a human inside. Whether these "dark factories" become the standard or stay the showcase is still an open question, and I'd rather watch it than declare on it.

The pattern I'd put in front of any UK policymaker reading this: the West funds robotics research, China funds robotics deployment. A grant for a clever prototype is not the same lever as guaranteed demand at industrial scale. The first buys you papers and a good demo. The second buys you a supply chain, and a supply chain is what shows up in the price of a robot five years later.

China has elevated embodied AI to a national strategy, with "Robot+" policy frameworks driving robots into manufacturing, logistics, healthcare and elderly care, as MERICS details. It's the same state playbook that delivered dominance in solar and EV batteries: coordinated funding, standards, and guaranteed domestic demand that lets scale crush unit costs. The result is already visible in lights-out "dark factories". The strategy targets deployment, not just research.

What should Western operators take from this?

The lesson isn't to panic about geopolitics. It's to read the layer you actually operate on. If your business touches hardware, logistics or manufacturing, your robotics supply chain increasingly runs through China whether or not you ever buy a Chinese robot. That's a sourcing and resilience question worth putting on the table this year, rather than in five.

For most knowledge-economy firms, humanoids are not yet an operational concern, and anyone telling you otherwise is selling something. The near-term point is about cost curves and timing. Industrial automation that was uneconomic at Western component prices becomes viable as Chinese-driven costs fall. Watch the price of automating a given task, because the floor is dropping faster than most plans assume.

What's worth copying from China isn't the politics, it's the reflex. The operators pulling ahead put a good-enough system into a real process this quarter, measure it honestly, and fix what breaks. Every cycle banks operational knowledge that a competitor still circling in pilot purgatory never gets to bank at all. Speed of deployment is itself a capability, and it's one a single plant can start building on Monday.

The strategic error I run into most is treating "China leads robotics" as a headline rather than a procurement signal. Admiration and alarm are equally useless. Two questions are not: where in my operation does a falling robot price change the build-versus-buy maths, and where does my supply chain already depend on Chinese hardware without anyone having decided that it should? Answer those and the abstract race turns into a concrete decision with a date on it.

If you want the operating-model context, the pillar on whether China is winning the AI race sets the frame.

For Western operators, China's robotics lead is a sourcing signal, not a spectator sport. With around 90% of humanoids and a 9× industrial-robot install lead now Chinese, the robotics supply chain runs through China regardless of where you buy. Treat falling robot costs as a build-versus-buy variable to monitor now, and borrow the deploy-first reflex while you're at it.

FAQ

Is China ahead of the US in robotics? Yes, clearly, on both humanoids and industrial robots. Around 90% of humanoids sold globally in 2025 were Chinese, by Rest of World's count, and China installed 295,000 industrial robots against the US's 34,200 in 2024, roughly 9× more, in Stanford HAI's AI Index. The lead is in manufacturing and deployment, not just research.

Which country makes the most humanoid robots? China, by a wide margin. About 90% of the humanoid robots sold worldwide in 2025 were Chinese, with six of the top-selling makers based in China and total volumes reported in the rough range of 13,000 to 18,000 units, by Rest of World's count. No other single country comes close on volume.

Why is China good at robotics? Because robotics is an atoms problem and China owns the relevant supply chain. Its EV giants, Xpeng, GAC and Xiaomi, repurpose the same batteries, motors and sensors for humanoids, domestic component-making keeps unit costs low, and a national embodied-AI strategy guarantees demand, as MERICS details.

Who are the main Chinese humanoid robot companies? The market leaders include Unitree and AgiBot, the two top global sellers in Rest of World's reporting, alongside EV makers moving in: Xpeng, GAC and Xiaomi. Six of the top humanoid sellers in 2025 were Chinese, which is why the country holds roughly 90% of global sales.

The bottom line

China leads humanoid and industrial robotics for a reason that's hard to argue with and harder to copy. It sold around 90% of 2025's humanoids and installed about 9× more industrial robots than the US, and that lead sits in atoms rather than bits. The car supply chain became the robot supply chain. Cheap domestic components drive unit costs down. A national embodied-AI strategy guarantees the demand. None of that is a model you can clone in a quarter, however good your lab is.

Which means the sceptics were right about the humanoids and wrong about what follows from them. The unit numbers are small, the videos are edited, and none of that touches the 295,000 industrial machines that went in the year before last, or the component industry that made them cheap enough to buy.

For Western leaders, the ask isn't alarm. It's attention. The cost of automating physical work is falling, and the supply chain behind it increasingly runs through China. Both are worth deciding on deliberately, rather than discovering late.

If you want the full picture, start with the pillar on whether China is winning the AI race.