Signals Inbox·August 27, 2026·Robotics

Who are the top Chinese robotics startups?

AgiBot is China’s top private robotics startup right now, followed by Pudu Robotics and Galbot, with Hai Robotics and RobotEra completing a top five that mixes humanoid upside with businesses already operating at serious commercial scale.

Q1What should “top Chinese robotics startup” mean here?

For this ranking, we consider the top Chinese robotics startups to be the private companies that combine real customer adoption, technical strength and a credible chance of becoming much bigger from here.

That definition changes the list quite a lot. Pudu Robotics has already shipped well above 100,000 commercial robots. Hai Robotics has tens of thousands of warehouse robots operating or being implemented. AgiBot, meanwhile, is much younger but currently leads global humanoid shipments. Galaxea AI has attracted huge funding because investors are betting on its embodied-AI technology even though its commercial footprint is still much smaller.

We therefore give the most weight to deployments, orders, repeat customers and revenue. Technology matters too, especially for companies trying to build general-purpose robots, but a great demo carries much less weight than a customer coming back for another hundred machines.

We also keep the definition of “startup” fairly strict. Unitree, UBTech and Geek+ are public companies now, while XPeng Robotics sits inside listed automaker XPeng. They are important benchmarks, but mixing them into a private-startup ranking would make the answer less useful.

What we look at when ranking Chinese robotics startups

Q2Is China’s robotics startup boom producing real scale or mostly demos?

China’s robotics startup boom is producing real scale now, although humanoid robots are still much better at being manufactured than at doing general-purpose work.

The industrial base behind these startups is already huge. The International Federation of Robotics counted about 295,000 industrial robot installations in China in 2024, more than half of the global total. Chinese suppliers also captured 57% of their domestic industrial robot market, roughly twice their share a decade earlier. Startups are growing inside an ecosystem that already knows how to make motors, actuators, batteries, sensors and complete robots at scale.

Humanoids are now starting to show the same manufacturing effect. Counterpoint Research's latest count puts global humanoid shipments above 22,000 units in the first half of 2026, almost four times the level a year earlier. AgiBot shipped around 9,700 and Unitree more than 7,000. Chinese companies occupied all five places at the top of Counterpoint's shipment ranking.

The latest robot competitions in Beijing also show why manufacturing progress and intelligence need to be separated. More than 2,000 robots competed at the World Humanoid Robot Games, with machines now running extremely fast and handling increasingly delicate objects. Yet the Financial Times reported at the same time that truly general autonomous behavior is still years away, and some impressive tasks still rely on teleoperation.

So we are already past the prototype phase in hardware. We are much earlier in the race to build robots that can reliably walk into a new workplace, understand what needs doing and perform it without constant human help.

Q3Is AgiBot China’s top private robotics startup right now?

AgiBot is currently our number-one Chinese robotics startup because no other private company combines humanoid shipment leadership, fast-growing revenue and such an aggressive manufacturing ramp.

Counterpoint's latest market data puts AgiBot at roughly 9,700 humanoid shipments in the first half of 2026, giving it more than 43% of the global market. It is a striking number for a company founded in 2023. Counterpoint had already estimated more than $140 million in AgiBot revenue in 2025, the highest humanoid-robot revenue in the world.

The growth behind that figure is even more revealing. According to Counterpoint, AgiBot went from roughly RMB3 million of revenue in 2023 to more than RMB1 billion in 2025. The company also passed 10,000 cumulative robot shipments and has built several product families rather than betting everything on one expensive biped.

That mix helps explain why we put AgiBot ahead of newer contenders. Its wheeled G-series robots can work in manufacturing and logistics while its humanoids, smaller research robots and data-production machines attack different parts of the market. AgiBot gets several chances to find repeatable commercial demand while the hardest humanoid problems are still being solved.

The company is currently preparing a Hong Kong IPO. Reporting around the process has produced very different valuation figures, so we would be careful with the highest numbers being circulated. The operating evidence is much harder to dismiss: AgiBot is already shipping humanoids at a scale that most global competitors have not reached.

Q4Can Galbot actually catch AgiBot?

Galbot is the private Chinese humanoid startup with the clearest chance of catching AgiBot because its robots are now moving from big funding rounds into serious industrial deployments.

Counterpoint estimates Galbot shipped more than 1,100 humanoids in the first half of 2026, enough for roughly 5% of the global market and third place behind AgiBot and the now-public Unitree. That still leaves a huge gap with the two leaders, so calling Galbot a co-leader today would be premature.

Its customer evidence is stronger than the shipment ranking alone suggests. Galbot says it has accumulated orders for several thousand robots from companies including CATL, Bosch, Toyota, BAIC and SAIC. CATL has gone further than simply signing a partnership: the battery maker says Galbot's S1 is now operating on its production lines in battery-module and pack manufacturing, handling heavy materials during extended autonomous operation.

Galbot is also trying to find demand outside factories. Its work with Meituan on robotic pharmacies and its “Galaxy Space Capsule” retail format give the company another environment in which robots can repeatedly perform a limited set of tasks. That is a much more realistic route to commercialization right now than asking one humanoid to do anything a human can do.

Investors are paying accordingly. Galbot raised RMB2.5 billion early this year after another round of more than $300 million only a few months earlier, with backers including China's national AI fund, Sinopec and automotive investors. Its valuation had already passed RMB20 billion.

AgiBot remains well ahead on shipments and disclosed revenue. Galbot, however, has probably done more than any other private challenger lately to turn industrial partnerships into robots that actually work inside customer facilities.

Q5Is Pudu Robotics stronger than the humanoid startups everyone talks about?

Pudu Robotics is currently one of China's strongest robotics startups by almost any commercial measure, even though its delivery and cleaning robots attract much less hype than humanoids.

Pudu says it has shipped more than 130,000 robots worldwide and sells across more than 80 countries and regions. Overseas markets have generated more than 80% of revenue for three consecutive years. Frost & Sullivan's 2025 market study ranked Pudu first globally in commercial service robotics by both revenue and shipments, with roughly one quarter of the market.

Recent activity suggests the business is still expanding rather than coasting on an old installed base. Pudu opened a new US headquarters, announced a large cleaning deployment with Swiss retailer Denner and recently launched another AI-native cleaning robot aimed at smaller commercial spaces. These are fairly mundane applications compared with humanoid demos, which is precisely why we take them seriously: restaurants, supermarkets, hotels and warehouses already know what job they are buying the robot to perform.

Investors have also come back. Pudu raised nearly $150 million this year at a valuation above $1.5 billion, taking disclosed cumulative funding above $300 million.

We rank Pudu just behind AgiBot because AgiBot has a stronger position in the potentially much larger general-purpose robotics market. If we ranked Chinese robotics startups only on what customers are already paying for today, Pudu would have a very good case for first place.

Q6Is Hai Robotics already bigger than most humanoid startups?

Hai Robotics is already a much more proven robotics business than most private humanoid companies, with thousands of enterprise deployments rather than a collection of pilots.

Hai says more than 29,000 of its warehouse robots are operating or being implemented for over 1,200 customers. It delivered more than 7,000 robots during 2025 alone. Those numbers put the company in a completely different commercial stage from startups still celebrating their first few hundred humanoid deliveries.

The quality of those deployments strengthens the case. Hai has worked with Maersk on a high-density automated fulfilment center in Singapore and has partnerships with major automation and logistics groups such as TGW Logistics and LX Pantos. It recently expanded its Americas leadership team and opened an EMEA innovation center in the Netherlands, so international growth is still a priority rather than a finished chapter.

Warehouse robotics is narrower than the humanoid dream. Hai's robots operate in environments that can be designed around them, while a truly general robot would eventually tackle thousands of less structured jobs.

For now, that narrower problem has produced a real business. We rank Hai near the top because large enterprise customers have already shown that they will deploy these systems repeatedly and at scale.

Q7Is RobotEra quietly becoming a serious humanoid contender?

RobotEra is currently one of the easiest Chinese humanoid startups to underestimate because its commercial evidence is stronger than its international profile.

The company disclosed more than RMB500 million in cumulative orders, with industrial customers across automotive, logistics, consumer electronics and manufacturing. Named customers and partners include Samsung, Geely, Renault, Lenovo, Haier and TCL. RobotEra also says some customers have bought from it as many as six times.

Repeat orders are particularly useful here. A humanoid company can manufacture a few dozen machines for an experiment and still generate impressive photos. A sixth order means somebody has repeatedly decided that the product is useful enough to spend more money on it.

RobotEra has also found traction in logistics. Its M7 has been used for parcel induction and sorting, and SF Group led the company's latest funding round. Before that, RobotEra raised RMB1 billion in a strategic round that valued the business above RMB10 billion. It then added more than $200 million in fresh capital shortly afterward.

Its dexterous-hand business gives RobotEra another interesting route into the market. The company says it shipped more than 1,000 hands in 2025, roughly half of them overseas. RobotEra is therefore more than a bet on one complete humanoid body: it can sell components, development platforms and complete robots while the market takes shape.

Q8Is Mech-Mind building a better robotics business than the humanoid crowd?

Mech-Mind has built one of China's most convincing private robotics businesses by selling the perception and manipulation technology that many different kinds of robots need.

The company makes industrial 3D cameras, AI software and manipulation systems that help robots see an object, understand where it is and work out how to handle it. Mech-Mind says more than 29,000 of its cameras are installed worldwide, across roughly 50 countries and regions, with more than 100 Fortune Global 500 customers.

That gives Mech-Mind a useful position in the robotics stack. Factories do not have to decide that humanoids will replace robotic arms before Mech-Mind can win. Its technology can be used with existing industrial arms, mobile systems and newer embodied robots.

We also like the variety of environments in which the technology is already being used. Bin picking, depalletizing, machine tending and assembly all require the same underlying perception problem to be solved repeatedly. That creates a much clearer path to revenue than waiting for one general-purpose robot to master every factory task.

Mech-Mind is now moving toward the public markets. Its lower profile compared with humanoid names can make it look less important than it is, but the installed base suggests otherwise. Among Chinese robotics startups, it is one of the strongest examples of a company selling an enabling layer rather than chasing the most fashionable robot shape.

Q9Are Galaxea AI and LimX already worth the hype?

Galaxea AI and LimX are two of China's strongest embodied-AI technology bets, but their funding has moved faster than the public evidence of large commercial deployments.

Galaxea shows the gap most clearly. Caixin reported that the company raised RMB1 billion at a RMB10 billion valuation and then, only weeks later, raised another RMB2 billion at a valuation above RMB20 billion. Investors effectively doubled the price of the company in a very short period.

The attraction is understandable. Galaxea is working on generalist embodied models that can control different robotic forms, which could become an important layer if robot intelligence starts transferring cleanly across hardware. The upside is much larger than selling a single specialized machine. What we still lack is a Pudu-sized installed base, Hai-sized customer count or AgiBot-sized revenue number to validate that valuation commercially.

LimX sits in a similar category but has recently moved closer to production. The company raised about $400 million over six months, including a nearly $200 million pre-IPO round, and was valued at roughly $2.2 billion in that financing according to CB Insights. LimX says the money will help it deploy thousands of autonomous humanoids and expand internationally.

Its technical focus on combining whole-body control with higher-level cognition is genuinely important. LimX also released new software infrastructure, launched its full-size Luna humanoid and is preparing for an IPO. Still, “deploy thousands” remains a plan until those robots reach customers.

We rank both companies highly because the technology could become central to general-purpose robotics. Right now, though, AgiBot, Galbot and RobotEra have more commercial proof.

Q10Is Deep Robotics proof that specialized robots can beat humanoids commercially?

Deep Robotics shows that a Chinese robotics startup can build a fast-growing, profitable business without waiting for general-purpose humanoids to work.

The company's IPO filing gives us unusually clean financial evidence. According to Caixin's review of the filing, Deep Robotics increased revenue from RMB103 million in 2024 to RMB337 million in 2025. It also reported RMB28.7 million in net profit.

The important part is where the money came from. Deep Robotics built its business around quadruped robots used for inspection and other difficult industrial environments. The same filing showed only four humanoid sales across 2024 and 2025.

Government subsidies contributed RMB11.9 million to 2025 profit, so we should not read the net-income figure as pure operating strength. Even after making that adjustment, however, tripling revenue while selling robots into real industrial jobs is much stronger evidence than another funding announcement.

Deep Robotics is now seeking roughly RMB2.5 billion through a Shanghai listing. Its trajectory is a useful reminder for the rest of this ranking: highly specialized robots can become valuable businesses long before a general-purpose humanoid can reliably do the same job.

Q11Which smaller Chinese humanoid startups could still break into the top tier?

Booster Robotics, EngineAI and Noetix are the smaller Chinese humanoid startups we would watch most closely now because each is finding a different way around the industry's cost and commercialization problems.

Booster is currently the most convincing of the three. The company says its K1 platform has passed 1,000 cumulative deliveries, with customers across more than 20 countries, and it reached positive monthly operating cash flow late last year. Its latest funding round brought in nearly RMB1 billion. Booster has focused heavily on universities, developers and robot competitions, where customers are willing to buy useful hardware before general-purpose autonomy exists.

EngineAI is attacking the cost curve. CEO Zhao Tongyang said this week that the manufacturing cost of a practical general-purpose humanoid has fallen below RMB100,000, compared with more than RMB1 million roughly three years ago. We would still want to see large paid deployments before putting EngineAI into the top group, but a tenfold cost reduction changes what customers can realistically test and buy.

Noetix is pushing even further downmarket. Its small Bumi humanoid launched below RMB10,000 and secured a disclosed 1,000-unit order, while the company raised nearly RMB1 billion in its Series B. Noetix is effectively testing whether humanoids can become education and consumer products before they become universal workers.

Fourier Intelligence remains relevant too, especially in rehabilitation and healthcare, where it has years of robotics experience. We currently give the three companies above more weight because their latest commercialization moves tell us more about where they could break through next.

Smaller Chinese humanoid startups to watch

Q12Is China’s humanoid demand real or still heavily policy-driven?

China's humanoid market is growing very fast today, but a large part of current demand still comes from research, data collection, entertainment and policy-backed programs rather than companies replacing workers because the economics are already obvious.

Counterpoint's latest application breakdown is revealing. Entertainment and performance together with data production and research still account for more than 60% of humanoid shipments. Intelligent manufacturing has reached about 13%, while warehousing and logistics account for around 5%.

Industrial use is clearly growing, and examples such as Galbot at CATL show that factory deployment is becoming real. The proportions tell us that the market has not yet reached the point where ordinary commercial labor replacement drives most purchases.

China's huge network of robot training centers also affects the numbers. Interact Analysis counted at least 90 humanoid data-collection and training centers operating, planned or under construction by the spring, including at least 64 already operating. The Financial Times recently examined how government-backed centers buy robots, generate teleoperation data and sometimes sell that data back to robot developers. That structure helps the industry learn faster, while also making raw shipment numbers harder to interpret as pure end-customer demand.

The government is openly trying to push the market toward real use. China's current MIIT program calls for more than 100 high-value real-world application scenarios and capacity for deployments at the ten-thousand-unit level, covering factories, logistics, retail, healthcare, inspection and emergency response.

So deployment numbers become much more convincing when we can connect them to a named customer, a repeated task and follow-on purchases. Those three pieces are still much rarer than the overall shipment growth would suggest.

Q13Which Chinese robotics startups are actually winning outside China?

Pudu Robotics, Hai Robotics, Mech-Mind and RobotEra currently give us the strongest evidence that private Chinese robotics companies can sell globally rather than relying mainly on China's domestic robotics boom.

Pudu is the clearest case. More than 80% of its revenue has come from overseas markets for three straight years. The company is now building regional infrastructure around that demand, including a US headquarters and local commercial partnerships.

Hai has also built a genuinely international enterprise business. Its warehouse systems are deployed through large logistics customers and integrators, and the company now operates regional facilities and teams across Europe, Asia and the Americas. Maersk's Singapore fulfilment project is a good example of Hai technology being integrated into a global operator's normal logistics workflow.

Mech-Mind says its technology has reached roughly 50 countries and more than 100 Fortune Global 500 clients. The company can internationalize without exporting an entire new robot category because its cameras and software plug into industrial automation that customers already understand.

RobotEra is the younger name we would add to this group. The company says roughly half of its RMB500 million order book comes from international business, while about half of the dexterous hands it shipped in 2025 also went overseas. As seen above, repeat orders from industrial customers make that more convincing than a collection of overseas demonstrations.

AgiBot and Galbot are both pushing abroad too, but today their strongest evidence still comes from manufacturing scale and Chinese industrial deployments. Pudu, Hai and Mech-Mind have already shown that a Chinese robotics company can build a substantial customer base across very different countries.

Q14Should Unitree, UBTech, Geek+ and XPeng Robotics count as startups?

Unitree, UBTech and Geek+ have already graduated from the startup category, while XPeng Robotics is better treated as a heavily funded corporate robotics business.

Unitree is the most recent change. The company has just listed on Shanghai's STAR Market after raising about RMB6.1 billion in its IPO. Its shares closed their first trading day roughly 460% above the offer price, giving Unitree a market value of around $50 billion. Unitree is now one of the most important public benchmarks for every private humanoid company in China, but it no longer belongs in a startup ranking.

The IPO also gave us unusually useful financial data. Unitree reported RMB1.69 billion in 2025 revenue and has been profitable, making it one of the clearest examples that Chinese humanoid and quadruped hardware can already support a substantial business.

UBTech has been listed in Hong Kong since 2023. Geek+ went public in Hong Kong in 2025 and then reported RMB3.17 billion of 2025 revenue, 31.6% year-on-year growth and positive adjusted profit. Both remain relevant competitors, but their public-company status puts them in a different group.

XPeng Robotics is the opposite case. The unit has just raised more than $900 million at a post-money valuation above $6.3 billion, the largest single private funding round disclosed in China's embodied-AI sector. Tencent and Alibaba joined the financing, and XPeng plans to mass-produce its Iron humanoid.

Calling XPeng Robotics a normal startup would still be misleading. It was built inside XPeng and can use the listed automaker's factories, engineers, supply chain and autonomous-driving expertise. We would include it in any analysis of who could win Chinese humanoid robotics, but not in the private-startup ranking below.

Q15Who are the top Chinese robotics startups right now?

AgiBot is currently the top Chinese robotics startup overall, followed by Pudu Robotics and Galbot, while Hai Robotics and RobotEra round out our top five.

AgiBot wins because it now has both scale and upside. It leads private humanoid companies in shipments and has already generated meaningful revenue, while still attacking a market that could become far larger if general-purpose robots improve.

Pudu comes next because its business is more proven than almost every humanoid startup on this list. Galbot takes third because its gap with AgiBot is still large, but its CATL deployment, industrial order book and rapid shipment growth give us more confidence than funding alone would.

Hai deserves fourth place on commercial maturity. RobotEra gets fifth because its repeat orders, logistics deployments and component business make it one of the most interesting private humanoid challengers these days. Mech-Mind follows closely with arguably the safest business model in the group.

The speculative part of the ranking starts around LimX and Galaxea. Both could eventually become much more important if generalist embodied AI works at scale, but we currently have less customer evidence to work with. Deep Robotics, Booster, EngineAI and Noetix then represent four very different bets: specialized industrial robots, developer platforms, cheaper full-size humanoids and consumer-scale machines.

The bigger conclusion is clear. China's strongest robotics startups today are spread across several very different markets, and the companies with the best businesses are not always the ones building the most impressive humanoid. AgiBot currently gives us the best combination of both worlds, which is why we put it first.

Top Chinese robotics startups right now