Which Chinese humanoid company is ahead?

Which Chinese humanoid companies are actually serious?

China’s serious humanoid race includes about ten independent vendors, although only four have enough commercial evidence to contend for the overall lead.

China’s Ministry of Industry and Information Technology says the country has developed more than 400 humanoid products, more than half of the global total. That headline number greatly overstates the competitive field. Many projects remain local-government platforms, university spinouts, prototypes, or companies whose main business sits elsewhere. We focus on independent vendors with at least one of four things: documented shipments, repeated deployments, material revenue, or enough capital and manufacturing progress to challenge the leaders.

Unitree, AgiBot and UBTECH form the top tier. Unitree and AgiBot each shipped more than 5,000 humanoids in 2025 according to company disclosures and Omdia estimates. UBTECH shipped fewer units, yet it generated the strongest documented industrial humanoid revenue and built the deepest roster of large factory customers. Galbot sits just behind them after a run of credible industrial progress.

Leju, Robotera, LimX Dynamics, Booster Robotics, Noetix and EngineAI make up the next group. Their positions vary sharply. Leju and Booster have shipped meaningful volumes in education and research. Robotera claims batch deliveries in logistics. LimX has raised enormous sums but still discloses little commercial scale. Noetix and EngineAI have attracted attention through affordable machines and dynamic demonstrations, although their evidence of durable paid use remains thinner.

We exclude Fourier Intelligence from the core ranking because rehabilitation remains central to its business. We treat Matrix Robotics as an emerging watchlist company until its roughly 1,000 orders turn into larger verified deliveries.

Funding totals remain approximate because several Chinese rounds were undisclosed and databases count them differently. The table shows publicly identifiable capital without pretending that every company can be compared to the nearest million. Unitree and AgiBot figures come from Tracxn; Galbot’s from CB Insights; the newest Leju, Robotera, LimX, Noetix and EngineAI amounts come from company announcements or reported rounds.

Company	Main position	Publicly identifiable funding	Why it belongs
Unitree	Low-cost bipedal platforms	About $250 million before its IPO	Thousands of shipments, profitable operations and a large developer base
AgiBot	Full-stack embodied AI	At least about $84 million, probably more	Top-tier volume, broad product range and the strongest public data strategy
UBTECH	Industrial and service humanoids	Public company; private total is no longer comparable	Best documented industrial revenue, orders and customer quality
Galbot	Wheeled and heavy-duty manipulation robots	About $887 million	Live industrial deployment, large agreement and exceptional capital strength
Leju Robotics	KUAVO research and industrial robots	At least about $207 million disclosed	Meaningful shipments, Huawei ties and growing industrial ambitions
Robotera	Logistics and industrial humanoids	More than $200 million in its latest round	Batch-delivery claims and deployments with major logistics groups
LimX Dynamics	General-purpose humanoids	$400 million raised over six months	One of the best-funded challengers and strong locomotion research
Booster Robotics	Compact research and competition robots	More than $10 million tracked	Global developer sales and unusually strong competition adoption
Noetix Robotics	Affordable compact humanoids	About $145 million in its latest round	Early mass deliveries and aggressive low-cost positioning
EngineAI	Dynamic full-size and compact bipeds	At least $15 million tracked; later rounds partly undisclosed	Fast hardware development, strong visibility and improving capital access

Has the Chinese humanoid race already narrowed to three companies?

The race has narrowed to Unitree, AgiBot and UBTECH, with Galbot close enough to punish any slowdown.

The top three lead for different reasons. Unitree has the best combination of shipments, revenue, profit and price. AgiBot has matched its shipment scale while building a broader embodied-AI stack. UBTECH has converted fewer robots into much more industrial revenue per unit and a large order book. The result is three specialized leaders rather than one dominant company.

Shipment concentration shows how far the field has separated. Omdia counted a little over 13,000 global humanoid shipments in 2025 and placed more than 10,000 of them with Unitree and AgiBot alone. Unitree’s later IPO filing revised its own figure upward, which leaves the exact number-one position debatable. The leading pair’s distance from most Chinese rivals is already several multiples.

UBTECH belongs in the top group despite lower volume because its machines sell into a different market. Its Walker series targets production lines, carries more integration work and produces far higher revenue per delivered robot. Galbot is the only challenger showing enough recent industrial progress to threaten that three-company structure.

If you want more recent data on this point, please see our latest humanoid robotics market report.

Who ships the most humanoid robots right now?

Unitree holds the narrowest possible shipment lead, while AgiBot leads cumulative production momentum.

Omdia estimated 5,168 AgiBot shipments and 4,200 Unitree shipments in 2025. Unitree then disclosed more than 5,500 humanoids in its IPO prospectus, roughly 31% above Omdia’s estimate. Using the company filing puts Unitree around 6% ahead of AgiBot’s Omdia number. The gap is tiny, far too small to support claims of outright volume dominance.

AgiBot’s strongest claim concerns cumulative production. The company reached 5,000 robots late in 2025 and said it passed 10,000 roughly one quarter later. Production and shipments are different measures, especially because AgiBot sells full-size bipeds, smaller humanoids and wheeled task robots. Even with that caveat, doubling cumulative output in about three months shows the fastest visible factory ramp among the leaders.

UBTECH delivered about 1,000 full-size humanoids, far below the leading pair by count. Leju, Galbot, Booster and Noetix also crossed or approached four-digit volumes according to industry and company reports, but many of those units are smaller research, education or service machines. Counting every body as an equivalent unit hides large differences in price, payload and intended work.

Shipment measure	Unitree	AgiBot	UBTECH
Common 2025 market estimate	4,200	5,168	About 1,000
Later company disclosure	More than 5,500 shipped	More than 5,100 shipped	About 1,000 delivered
Best current scale evidence	Audited IPO filing	Fastest cumulative production ramp	Full-size industrial deliveries
Main caveat	Most demand came from research and displays	Product categories are broader and less comparable	Lower volume reflects more expensive industrial systems

Who is actually making money from humanoid robots?

Unitree is the clearest financial winner in Chinese humanoid robotics today.

Its IPO filing reported RMB1.708 billion in 2025 revenue, up 335%, with attributable net profit of about RMB288 million. During the first nine months of the year, humanoids generated just over half of core product revenue and gross margin approached 60%. No close Chinese rival has disclosed the same combination of scale, growth and positive earnings.

The quality of Unitree’s revenue is less impressive than the headline. Roughly 74% of humanoid sales during the first nine months came from research and education, another 17% from displays and entertainment, and only about 9% from industrial applications. Unitree has built a highly successful hardware company, but most buyers are still purchasing development platforms. Proven labor payback remains rare.

AgiBot said its revenue rose from RMB60 million to RMB1.05 billion in 2025. That 17.5-fold increase would make it the fastest financial scaler in the leading group, although the number comes from management and lacks an audited public filing. AgiBot also mixes industrial robots, interactive humanoids, smaller machines and data services, so we cannot cleanly isolate the economics of its full-size bipeds.

UBTECH generated RMB820.6 million from full-size humanoid robots and delivered 1,079 units, according to its annual reporting. The simple ratio is roughly RMB760,000 of revenue per robot, more than four times Unitree’s disclosed average humanoid selling price during the first nine months of 2025. UBTECH has proved customers will pay much more for integrated industrial systems, yet the group still reported a large annual loss.

If you want more recent data on this point, please see our latest humanoid robotics market report.

Who is growing fastest now?

AgiBot is expanding fastest overall, while UBTECH is growing fastest inside industrial factories.

AgiBot compressed years of early production into a much shorter second ramp. It took the company until late 2025 to reach its first 5,000 robots, then roughly three months to add the next 5,000. Its reported revenue also climbed more than seventeenfold. No rival matches that combination of production acceleration and company-reported commercial growth today.

UBTECH’s humanoid revenue increased more than twentyfold from a very small base. That growth carries more commercial weight than a surge in research-platform shipments because the company recognized hundreds of millions of renminbi from full-size industrial systems. Its remaining order book suggests another step up if factories accept the robots and deployments expand beyond initial batches.

Unitree grew from a much larger base and still increased revenue by 335%. Its pattern is financially healthier than the raw percentage ranking suggests: prices fell, humanoid volumes rose sharply, gross margin remained high and the company stayed profitable. AgiBot wins the speed comparison today, but Unitree has produced the least fragile growth.

Galbot deserves attention here. Recent industrial deployments and agreements have moved its story from fundraising toward operational demand, although the company still lacks annual revenue disclosure comparable with the top three.

Which Chinese humanoid robots are doing real factory work?

UBTECH still leads factory commercialization, but AgiBot and Galbot have closed much of the credibility gap.

UBTECH’s Walker robots have entered automotive, electronics and other industrial programs involving BYD, Geely, FAW-Volkswagen, Foxconn, Airbus and Texas Instruments. Its annual humanoid revenue and roughly RMB1.4 billion order book provide stronger proof than customer logos alone. The disclosed tasks include material handling, sorting, inspection and production-line logistics, which are narrow enough to automate reliably and valuable enough for factories to buy.

The machines remain well below human versatility. UBTECH has previously said suitable Walker tasks reach around 30% to 50% of human productivity, and deployments still depend on structured workstations, integration and supervision. Even so, a thousand paid industrial units can carry more commercial weight than several thousand robots sold for research.

AgiBot recently provided the most useful public endurance test among Chinese competitors. Eight robots worked for 64 hours across a six-day livestream at Longcheer Technology, participating in the inspection and production flow for 17,625 tablets. The company reported a 99.99% task success rate. We should treat that percentage as vendor-reported, yet the duration, active factory and measurable throughput make the event more informative than a short staged clip.

Galbot has now entered the same conversation. Its S1 is operating on CATL production lines with up to eight hours of runtime and a heavy-duty dual-arm payload specification. The company has also announced a 1,000-unit agreement with Patek Precision for electronics manufacturing. Delivery timing and acceptance data remain unclear, but Galbot’s latest evidence is industrially relevant and large enough to alter the ranking.

Unitree trails on paid factory use. As noted above, industrial buyers represented only about 9% of its humanoid demand during the disclosed period. Its machines are capable bodies and widely used development tools, but the company has yet to show a comparable portfolio of long-running production deployments.

If you want more recent data on this point, please see our latest humanoid robotics market report.

Who can build reliable humanoids at scale?

Unitree has the strongest manufacturing system, AgiBot has the fastest production ramp, and UBTECH has the most mature industrial configuration.

Unitree entered humanoids after years of producing quadrupeds, giving it supplier relationships, actuator knowledge, testing processes and after-sales experience before demand surged. Its prospectus links in-house work on motors, reducers, control systems and other components to lower costs and high margins. More than 5,500 annual humanoid shipments provide stronger manufacturing evidence than a factory-capacity target.

AgiBot has demonstrated exceptional ramp speed across a wider product family. Its current portfolio spans full-size humanoids, compact education robots, wheeled industrial systems, quadrupeds and manipulation hardware. At the latest WAIC, it added the A3 Ultra, X2 Edu, G2 Max and OmniHand 3 Ultra-M while operating more than 30 robots across the venue. Breadth improves supply-chain leverage, although it also makes aggregate production less comparable with a pure biped count.

UBTECH’s Walker S2 prioritizes factory uptime over athletic movement. Autonomous battery swapping, industrial navigation, production-system integration and coordinated manipulation make it more deployment-ready for repetitive shifts. Its manufacturing capacity has moved ahead of completed deliveries, so the next test is utilization: how much of that capacity turns into accepted robots, recognized revenue and repeat orders.

Smaller challengers have found narrower paths to scale. Booster says its compact robots serve hundreds of customers across more than 20 countries, and 38 of 59 teams at RoboCup 2026 used Booster machines. Noetix delivered 105 robots in one month shortly after beginning mass production. These figures show functional assembly operations, though they remain far below the top tier in product value and commercial depth.

Who has the strongest robot brain?

AgiBot currently has the strongest public evidence in embodied AI and training data.

Its AgiBot World dataset contains more than one million real-world trajectories across 217 tasks and five deployment scenarios. The associated research reported a 30% average improvement over policies trained on Open X-Embodiment, success rates above 60% on difficult tasks and a 32% advantage over the referenced RDT approach. The work came from contributors tied closely to AgiBot, but the dataset, methodology and benchmark results are more inspectable than most competitor claims.

The company has also pushed beyond locomotion into manipulation. Its latest product release pairs full-size platforms with a heavy-payload task robot, a new dexterous hand and manipulation foundation models. That product sequence shows a coherent strategy: collect diverse trajectories, train general policies and deploy them across different bodies.

UBTECH’s advantage lies in industrial data. Factory deployments can produce highly valuable examples of repeated handling, inspection and coordination tasks, and the company has built BrainNet and multi-robot systems around that environment. Public evidence remains less benchmarked than AgiBot’s research, so we can see the strategic value without knowing how large the software performance gap really is.

Unitree has historically led with the body. Its new H2 Plus collaboration with NVIDIA directly addresses that weakness through Jetson Thor, Isaac GR00T, simulation and teleoperation tools. The platform could make Unitree the default physical layer for outside AI researchers. For now, it serves as a development reference and does not prove that Unitree owns the best autonomous policy.

Which humanoid gives buyers the best value?

Unitree offers the strongest price-to-hardware ratio in China by a wide margin.

The G1 entered the market around $16,000, while the R1 pushed the entry point close to $6,000. Unitree’s filing gives the more important trend: its average humanoid selling price fell from RMB260,700 in 2024 to RMB167,600 during the first nine months of 2025, a decline of roughly 36%, while sales volume expanded sharply. Lower prices arrived alongside higher scale and strong margins, a pattern consistent with manufacturing efficiency.

That value proposition is aimed mainly at researchers, universities, developers and entertainment operators. Buyers receive a capable programmable body; reliable labor payback remains unproven. Unitree’s lead is therefore strongest in acquisition cost and platform accessibility, while UBTECH and Galbot compete on the economic value of completing industrial tasks.

UBTECH’s implied humanoid revenue per delivered unit sits around RMB760,000, including integration and services. Galbot’s heavy-duty S1 adds up to eight-hour operation, hot-swappable batteries and high payload capacity, features that could justify a higher price where moving heavy materials reduces injuries or labor requirements. Neither company discloses enough customer-level payback data to prove that its more expensive robot produces the best total cost of ownership.

Who has the best customers and the hardest lead to copy?

UBTECH has the strongest customer moat today, while Unitree owns the deeper hardware ecosystem.

UBTECH’s relationships with major automakers, electronics manufacturers and industrial groups give it repeated access to environments where reliability failures have direct financial costs. Integrating a robot into plant software, safety procedures, workstations and maintenance routines creates switching costs that a cheaper standalone robot cannot immediately overcome. Its order book reinforces that position, although renewals and larger follow-on deployments will decide whether the moat becomes durable.

Unitree’s defensibility comes from a different loop. Vertical integration lowers costs, high shipment volume spreads engineering expense, and broad university adoption attracts researchers who build software on its machines. Recent work on motion tracking, teleoperation and even experimental surgical assistance has used Unitree hardware, extending the platform beyond the company’s own software team.

AgiBot’s moat could eventually be stronger than both. Large datasets, multiple robot bodies and a shared manipulation stack can create compounding returns if each deployment produces useful training data. The openness of AgiBot World also weakens exclusivity because rivals can learn from the dataset. Its real advantage will depend on proprietary field data and whether its models outperform on customer tasks. Publishing the largest public collection cannot sustain the moat by itself.

Capital is accelerating the challengers without proving leadership. Galbot has raised close to $900 million, LimX has raised $400 million in six months, and Robotera announced a round above $200 million. Those war chests can fund factories, data collection and field teams, but Unitree reached profitability with much less disclosed private capital. Funding currently measures attack capacity more reliably than competitive position.

The biggest replication threat may come from established companies. XPeng plans to exceed 1,000 IRON robots per month before its broader launch, while automotive and technology groups already control factories, AI teams and distribution. The startup leaders have a head start, but none has secured the category against large incumbents.

If you want more recent data on this point, please see our latest humanoid robotics market report.

Who has gained the most ground lately?

Galbot has made the largest recent leap, although AgiBot still carries the strongest sustained momentum.

Galbot previously looked like a heavily funded manipulation startup with limited proof of scale. Its S1 is now operating inside CATL’s battery production lines, the partnership extends into aftermarket service, and a separate agreement covers 1,000 robots for electronics manufacturing. Together, these developments address deployment, service and volume, three bottlenecks that fundraising alone could not solve.

AgiBot’s momentum has lasted longer and covers more dimensions. It completed a multi-day factory demonstration with measurable throughput, expanded internationally and released four new products at the latest WAIC. More than 30 AgiBot robots operated across the conference venues, a modest commercial use case but useful evidence of fleet management and repeated public interaction.

Unitree recently strengthened its weaker software position through H2 Plus and NVIDIA’s embodied-AI stack. Its IPO registration has also cleared a major regulatory step, giving it a route to finance larger model development and manufacturing. Useful progress, yes. But it does not change the ranking.

UBTECH opened a new consumer front with the UWORLD U1 series and reported more than 10,000 orders. We treat that figure cautiously because orders can include reservations, distributor commitments or purchases that have not been delivered or paid in full. The launch expands UBTECH’s addressable market, but it carries less weight in our ranking than recognized industrial revenue.

Can we trust the shipment and order numbers?

The broad ranking is reliable, but many precise gaps are less certain than they appear.

Unitree’s IPO filing and UBTECH’s public reporting provide the strongest evidence because they disclose revenue, units, margins and business risks under securities rules. Omdia offers a common market methodology, yet its Unitree estimate missed the company’s later shipment disclosure by roughly 30%. Even professional estimates struggle when private vendors define products differently and disclose selectively.

The vocabulary itself creates confusion. “Produced,” “rolled off the line,” “shipped,” “delivered” and “deployed” describe different stages. A production milestone cannot be treated automatically as accepted customer deliveries. The same caution applies to consumer reservations, industrial framework agreements and order pipelines.

Demonstrations also vary in evidentiary value. A six-day factory livestream with disclosed throughput tells us more than a choreographed dance. A teleoperated task proves the body can execute the motion but says little about autonomy. Recent reporting on humanoids continues to find human operators behind supposedly intelligent behavior, especially in unstructured settings.

We therefore weight audited revenue, accepted deliveries, repeated operating hours, named customers and follow-on orders most heavily. Capacity targets, reservation totals, viral videos and company benchmarks still matter, though they receive much less weight until outside evidence confirms them.

Who leads each part of China’s humanoid market?

China currently has three category leaders and several credible specialists; no company sweeps every segment.

Unitree dominates affordable bipedal hardware and research adoption. AgiBot leads public embodied-AI data and broad platform development. UBTECH remains strongest in paid factory deployments. Galbot has become the most serious heavy-duty manipulation challenger, while Booster has carved out a compact research and competition niche. That is why different datasets can name different “leaders” without making the overall comparison meaningless.

Segment	Current leader	Why it leads
Affordable humanoid hardware	Unitree	Lowest entry prices, high volumes and strong component integration
Research and developer adoption	Unitree	Large installed base and growing NVIDIA-compatible tooling
Public embodied-AI data	AgiBot	One-million-trajectory dataset and published manipulation results
Broad product portfolio	AgiBot	Full-size, compact, wheeled, quadruped and manipulation systems
Industrial factory commercialization	UBTECH	Highest documented humanoid industrial revenue and strongest customer roster
Heavy-duty manipulation	Galbot	Best mix of payload, runtime and fresh industrial traction
Logistics challenger	Robotera	Batch-delivery claims and deployments with China Post and SF-linked operations
Compact competition robots	Booster Robotics	Broad RoboCup adoption and international developer sales
Healthcare and rehabilitation	Fourier Intelligence	Deepest domain experience, although it sits outside the core general-purpose ranking

Which Chinese humanoid company is ahead overall?

Unitree is still ahead overall today, AgiBot is close enough to overtake it, and UBTECH remains the strongest industrial specialist.

Unitree wins because it has already combined the five things most competitors hold separately: thousands of shipments, large revenue, real profit, falling prices and a widely adopted hardware platform. Its IPO filing provides more verifiable operating detail than any private Chinese rival. As seen above, the NVIDIA partnership gives it a credible route to improve the software layer without surrendering its hardware advantage.

Its lead is narrow because the company’s current demand mix remains heavily concentrated in research, education and display uses. Factory adoption is the clearest missing piece. Should Unitree convert its installed developer base into reliable autonomous manipulation, the company could widen the gap quickly. Failure there would confine it to being the best robot-body supplier, short of leading useful humanoid labor.

AgiBot ranks second and poses the greatest immediate threat. It has near-peer shipment scale, the fastest visible production ramp, a much broader product family and the strongest public data strategy. As seen above, its cumulative production passed 10,000, while the Longcheer test added unusually concrete factory evidence. AgiBot moves into first place once it proves that these machines generate repeat paid demand and that its software advantage survives outside company-organized tests.

UBTECH takes third. Its industrial revenue, order book and customer list are better than Unitree’s or AgiBot’s, and its Walker systems are closer to purpose-built factory equipment. Lower volumes, high prices and continued losses keep it behind overall. The new consumer orders could change its scale dramatically, but delivered units and recognized revenue will matter more than the headline order count.

Galbot now holds fourth and has shortened the distance to UBTECH. As pointed out above, it has secured a prestigious live industrial environment and a large electronics-manufacturing agreement, while its capital base is much larger than Unitree’s pre-IPO funding. The remaining gap is financial proof: Galbot has yet to disclose annual humanoid revenue or accepted deployments comparable with the top three.

The next group remains several steps behind. Leju and Robotera have credible deployment stories, LimX has exceptional financial backing, and Booster has built a strong compact-platform niche. None matches the leaders across scale, customers, economics and product maturity at the same time.

Rank	Company	Why it holds this position now
1	Unitree	Best verified combination of shipments, revenue, profit, price and platform adoption
2	AgiBot	Closest volume rival, fastest sustained momentum and strongest public embodied-AI strategy
3	UBTECH	Clear factory-commercialization leader with the best industrial customers and revenue
4	Galbot	Rapidly improving industrial challenger with credible factory deployment, a large agreement and huge funding
5	Leju Robotics	Meaningful shipment scale, strong Chinese technology partners and improving industrial positioning
6	Robotera	Promising logistics deployments and batch-delivery claims, with less independently verified financial evidence
7	LimX Dynamics	Excellent locomotion technology and enormous fresh capital, but limited disclosed commercial scale
8	Booster Robotics	Strong compact research ecosystem and global developer reach, though focused on a smaller-value segment
9	Noetix Robotics	Early mass-production progress and aggressive pricing, with shallow evidence of repeated paid use
10	EngineAI	Impressive dynamic hardware and visibility, but commercial deployment still trails its demonstrations

If you want more recent data on this point, please see our latest humanoid robotics market report.

Who is the author of this content?

NEW MARKET PITCH TEAM

We track new markets so founders and investors can move faster

We build living "market pitch" documents for emerging markets: AI, synthetic biology, new proteins, and more. Instead of outdated PDFs or hallucinated LLM answers, our clients get a clean, visual, always-updated view of what's really happening: key players, deals, regulations, and signals that matter. Learn more about us.