Is the Humanoid Robotics Market growing now?

Why does the humanoid robotics market suddenly feel real?

The humanoid robotics market feels different now because companies have started shipping robots by the thousands, booking real revenue and putting machines into customer operations.

A couple of years ago, most evidence came from prototypes and carefully controlled demos. The latest numbers describe a different industry. Interact Analysis estimates that global humanoid production jumped from fewer than 2,000 units in 2024 to more than 20,000 in 2025. Counterpoint's latest H1 2026 tracker already counts more than 22,000 shipments in six months.

Companies are also starting to disclose results that look more like a business. UBTECH sold 1,079 full-size humanoids in 2025 and generated RMB 820 million from that activity, up from just RMB 35.6 million the previous year. Agility Robotics says Digit has moved more than 100,000 totes in its commercial GXO deployment and now has more than $300 million of multi-year contracted orders.

We still need to be careful with the word "market." Research robots, entertainment robots, government-backed data-collection fleets and factory workers are all being counted together. The market is clearly getting much bigger, but the commercially useful part remains considerably smaller than the headline shipment numbers suggest.

Are humanoid robot shipments actually exploding right now?

Yes, humanoid robot shipments are growing extremely fast right now, even though nobody can measure the exact global total with much confidence yet.

Counterpoint Research estimates that more than 22,000 humanoid robots shipped globally during H1 2026, nearly 300% more than a year earlier. It expects full-year shipments to exceed 50,000, around 210% above its 2025 estimate.

The interesting wrinkle is that the newest trackers do not agree on the absolute number. Smart Analytics Global recently estimated 19,100 H1 shipments, while China's Humanoid Robot Scene Application Alliance reportedly counted more than 30,000 shipments from Chinese vendors alone. The definitions clearly differ enough that quoting a single global total to the nearest thousand would give us false precision.

The direction is much easier to establish. Interact Analysis independently found that production increased roughly tenfold in 2025. Counterpoint then found another near-300% year-over-year increase in H1 2026. Whatever definition we use, humanoid production has gone from hundreds of machines to thousands and then tens of thousands remarkably quickly.

Measure	Earlier level	Latest level	What it tells us
Interact Analysis production	Fewer than 2,000 in 2024	More than 20,000 in 2025	Roughly 10x in one year
Counterpoint H1 shipments	Roughly 5,500 implied one year earlier	More than 22,000	Nearly +300%
Counterpoint full-year estimate	Roughly 16,000 in 2025	More than 50,000 expected in 2026	About +210%

Are humanoid robots actually doing useful work yet?

Only a minority of humanoid robots are doing economically useful work today, although that minority is growing quickly.

This is probably the most important number in the whole market. Interact Analysis estimates that only around 10% of the more than 20,000 humanoids produced in 2025 went into real-world applications with a clear operational use. Most were used for R&D, AI training, data collection or entertainment.

Counterpoint's newer H1 2026 data shows some movement in the right direction, but the gap remains large. Entertainment and performance accounted for about 34% of shipments, while data production and research represented another 27%. Intelligent manufacturing reached roughly 13% and warehousing and logistics about 5%.

So out of more than 22,000 H1 shipments in Counterpoint's dataset, only around 4,000 went into manufacturing or logistics. That is still a huge increase from the fewer than 100 industrial humanoids Interact Analysis estimates were deployed in 2024. A fast-growing productive market is sitting inside an even faster-growing research, training and entertainment market.

Is China basically carrying the humanoid robotics market?

Yes, China currently carries most of the humanoid robotics market by unit volume, and the gap with the rest of the world is enormous.

Counterpoint's latest ranking has AgiBot at roughly 9,700 H1 shipments and Unitree above 7,000. Those two companies alone account for about three quarters of global shipments under Counterpoint's methodology. Galbot, UBTECH and Leju complete the top five, giving five Chinese companies a combined 86% share.

Interact Analysis reached a similar conclusion from its 2025 production data. It estimated that Chinese manufacturers produced more than 90% of the world's humanoids and that around 75% of global units were delivered to Chinese end users.

The scale is also becoming visible inside individual companies. Unitree disclosed that it sold 5,215 humanoids in 2025. UBTECH went from selling three full-size humanoids in 2024 to 1,079 in 2025. AgiBot has moved into production measured in several thousand units rather than dozens.

American companies are following a different path. Figure, Agility Robotics and Apptronik currently have much lower volumes, but their best deployments are concentrated around manufacturing and logistics customers such as BMW, GXO, Toyota, Mercedes-Benz and Catalyst Brands.

If we are talking about robots shipped, China is winning by a huge margin today. Narrow the question to proven industrial deployments, though, and the race looks considerably closer.

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

Are humanoid robots finally doing real factory and warehouse work?

Yes, humanoid robots are now doing real work in factories and warehouses, but only a handful of deployments have reached the level where we can measure meaningful operating history.

BMW gives us one of the cleanest examples because the automaker published actual operating data. Figure 02 worked for around 1,250 hours at BMW's Spartanburg plant over ten months, moved more than 90,000 sheet-metal components and supported production of more than 30,000 BMW X3 vehicles. The robot worked ten-hour weekday shifts on a repetitive task requiring precise positioning before welding.

Agility Robotics has produced another useful benchmark. Digit has moved more than 100,000 totes inside GXO's Flowery Branch logistics facility under a multi-year commercial agreement. That tells us far more than watching a humanoid complete a task five times on a conference stage.

The deployments are also getting more varied. Figure 03 is now being used at BMW for a harder sequencing workflow involving picking differently positioned parts and moving a wheeled cart. Figure has separately signed a commercial agreement with Catalyst Brands, which owns businesses including JCPenney, Aéropostale and Brooks Brothers, to deploy humanoids in distribution operations.

These deployments remain small next to conventional industrial automation. The International Federation of Robotics counts hundreds of thousands of traditional industrial robots installed globally each year. Humanoids are nowhere near that scale yet. Still, there are now enough operating machines to say that industrial humanoids have entered commercialization rather than merely preparing for it.

Company	Customer	What has actually happened
Figure	BMW	1,250 operating hours and 90,000+ components handled with Figure 02; Figure 03 now testing a harder logistics task
Agility Robotics	GXO	100,000+ totes moved under a multi-year commercial deployment
Figure	Catalyst Brands	Commercial distribution and logistics agreement
Agility Robotics	Toyota, Schaeffler and others	Digit operating in additional manufacturing environments

Are customers coming back after trying humanoid robots once?

Some customers are already expanding their humanoid robot experiments, which is encouraging, but repeat purchasing has not become normal yet.

BMW is a particularly useful case. The company finished the long Figure 02 production deployment and then brought the newer Figure 03 into Spartanburg for a different logistics task. BMW is also testing humanoid technology in Germany through Hexagon Robotics' AEON platform. That looks more like continued exploration of a useful technology than a publicity pilot that disappears after the press release.

GXO followed a similar path with Agility Robotics. The relationship moved from testing Digit into a multi-year Robots-as-a-Service deployment. GXO has also evaluated humanoids from other suppliers, including Apptronik, suggesting that it sees a broader automation category worth investigating.

What we cannot find yet is the pattern that would settle the debate: dozens of customers ordering five humanoids, finding that the economics work, and then coming back for 50 or 500.

For now, repeat engagement is appearing faster than repeat orders at massive scale. That is progress, but it is not the same thing.

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

Are humanoid robot sales finally big enough to count?

Yes, humanoid robot sales are now large enough to create real companies and meaningful order books, although the sector is still tiny beside mature automation markets.

UBTECH offers the clearest public financial evidence. According to its 2025 results, revenue from full-size embodied intelligent humanoid robots reached RMB 820 million, around $118 million, up 2,203.7% year over year. Humanoids contributed roughly 41% of UBTECH's total revenue and became its largest business line.

That is particularly striking because the revenue came from 1,079 units. It gives us an implied average of roughly RMB 760,000 per robot, or around $110,000, although the figure can include products, services and project components and should not be treated as a retail price.

Agility Robotics gives us a second type of evidence. In regulatory documents related to its planned $2.5 billion public-market transaction, the company disclosed more than $300 million of multi-year contracted Digit v5 orders. Agility has said these contracts cover roughly 1,000 robots across more than 30 customers.

The numbers are no longer trivial. One manufacturer can now generate more than $100 million of annual humanoid-related revenue, while another can build a contracted order book above $300 million. We have moved well past a market consisting entirely of grants and prototype sales.

Is China's humanoid robot boom based on real customer demand?

China's humanoid robot boom contains plenty of real demand, but government-backed training centers and policy support are currently boosting the numbers enough that we should not treat every shipment as proof of commercial ROI.

A recent Financial Times investigation looked at China's growing network of robot training centers. Local governments and affiliated organizations buy humanoids, employ people to teleoperate them through physical tasks, collect the resulting data and feed or sell that data back into the robotics ecosystem.

There is a rational reason to build this infrastructure. Physical AI needs huge amounts of real-world training data, and unlike language models, robotics companies cannot simply scrape decades of human movement data from the internet.

The problem comes when we interpret those purchases. A robot bought by a subsidized data center tells us that China is investing heavily in embodied AI. It tells us much less about whether an ordinary factory would buy the same robot from its own operating budget because it saves money.

The latest shipment mix reinforces that concern. More than 60% of Counterpoint's H1 volume still came from entertainment, performance, data production and research.

China is building the world's first genuinely large humanoid ecosystem, including factories, suppliers, training infrastructure and customers. We just need to separate that achievement from the narrower question of whether market demand can already stand without policy support.

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

Are humanoid robots productive enough to replace workers now?

No, today's humanoid robots generally cannot replace a normal worker across a full job, but they are becoming useful enough to take over specific repetitive tasks.

The BMW deployment shows the distinction well. Figure 02 repeatedly moved and positioned sheet-metal parts with millimeter-level precision inside an actual car-production process. That proves a humanoid can create useful output for long shifts. BMW has not published evidence showing that Figure 02 was cheaper, faster or more productive than the human alternative.

The same applies to Digit. Moving 100,000 totes is serious operational work, but Agility and GXO have not disclosed enough data on labor savings, supervision, downtime and total operating cost for us to calculate a clean human-versus-robot payback period.

UBTECH has been unusually open about the remaining performance gap. Company executives have said Walker robots can still operate at roughly 30% to 50% of human efficiency on some factory tasks.

That explains where humanoids are finding work first. Material movement, repetitive picking, machine tending, parts loading and inspection do not require a machine to behave like a person all day. A humanoid only needs to perform one constrained task reliably enough to earn its place in the workflow.

The "robot replaces an employee" story is still ahead of the technology. The "robot replaces one unpleasant part of an employee's job" story is already happening.

Are humanoid robots getting cheap enough to make economic sense?

Humanoid robots are getting dramatically cheaper at the low end, but industrial humanoids still need better productivity and reliability before the economics become obvious for most employers.

Unitree has pushed the price floor down aggressively. Its G1 starts around $13,500, giving universities, developers and smaller businesses access to hardware that would have cost many times more only a few years ago.

That price should not be confused with the cost of a production-ready industrial worker. More capable humanoids can still cost well into six figures once hardware, software, integration, maintenance and support are included.

Agility's order book gives us an interesting way to frame the economics. More than $300 million of potential three-year contract value across roughly 1,000 Digit v5 robots works out to around $100,000 per robot per year if everything is deployed and contracted value is realized. That price can make sense for a machine working multiple shifts in a high-cost labor market. It becomes much harder to justify if the robot spends significant time idle, needs frequent intervention or works much slower than a person.

The cost curve is heading in the right direction. Cheap Chinese hardware is pushing component prices down, manufacturers are redesigning robots for die casting and injection molding rather than expensive CNC machining, and higher production volumes spread engineering costs across more units.

Price is gradually becoming less of a blocker. Useful autonomous hours per dollar are now the metric we should watch.

Can humanoid robot makers really manufacture at scale?

Humanoid robot makers can now manufacture at a scale that looked unrealistic a couple of years ago, although installed capacity is beginning to run ahead of proven customer demand.

Figure provides one of the clearest examples outside China. Its BotQ factory has a first-generation capacity of up to 12,000 humanoids per year. During its recent Figure 03 ramp, the company said it moved from producing roughly one robot per day to demonstrating a one-per-hour production cycle in less than four months. More than 350 Figure 03 units had already been built when Figure disclosed the milestone.

Agility Robotics' Oregon RoboFab is designed for up to 10,000 Digit robots annually. UBTECH reported annualized full-size humanoid capacity above 6,000 units by the end of 2025. Hyundai plans to build Boston Dynamics Atlas robots at much larger scale, with a target of around 30,000 units per year by 2028.

China has already gone further on actual output. AgiBot shipped around 9,700 units in H1 alone according to Counterpoint, while Unitree exceeded 7,000.

This creates a new problem for the industry. Manufacturing is improving faster than anyone has proved demand for productive humanoids. If AI capability and customer ROI keep improving, today's factories will help push costs down very quickly. If they stall, the industry could end up with far more robots than customers know what to do with.

Company	Manufacturing evidence	Scale
Figure	BotQ first-generation line	Up to 12,000 robots/year
Agility Robotics	RoboFab	Up to 10,000/year
UBTECH	Reported annualized capacity	More than 6,000/year
AgiBot	Latest H1 shipments	Around 9,700 in six months
Hyundai / Boston Dynamics	Planned Atlas production	Around 30,000/year target by 2028

Is too much money chasing humanoid robotics right now?

Probably. Investment in humanoid robotics and physical AI is rising much faster than proven commercial revenue, and current valuations already assume a huge future market.

Crunchbase counted $18.8 billion of global robotics startup funding by mid-2026. That had already passed the $15 billion raised in all of 2025 and the previous $14.1 billion annual record from 2021. The dataset covers robotics more broadly than humanoids, but Crunchbase identifies embodied AI as one of the main drivers of the jump.

Humanoid companies themselves are raising extraordinary amounts. Figure raised more than $1 billion at a $39 billion post-money valuation. Apptronik has expanded its Series A to more than $935 million. XPeng's robotics subsidiary recently raised more than $900 million at a valuation above $6.3 billion.

Public markets are even hotter in China. Unitree's stock surged roughly 460% on its Shanghai debut, briefly valuing the company around $50 billion. Yet Unitree sold 5,215 humanoids in 2025, with much of that volume going toward research rather than productive labor.

This is where the market gets uncomfortable. Humanoid robotics can be growing extremely fast while humanoid robotics stocks and private valuations run even faster. Investors are already assigning tens of billions of dollars to companies whose current revenue remains tiny compared with the labor and automation markets they hope to disrupt.

We are comfortable saying that the industry is booming. Whether investors are paying sensible prices for that boom is a much harder case to make.

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

Is humanoid robot AI improving fast enough?

Humanoid robot AI is improving quickly, but general-purpose intelligence remains several years behind the hardware ambitions currently being priced into the market.

Figure's Helix system can now coordinate hands, arms, torso and locomotion for more complex physical tasks. Apptronik has opened Robot Park facilities where fleets of Apollo robots collect physical-world data for models being developed with Google DeepMind. Boston Dynamics is also working with DeepMind, while Chinese companies including AgiBot, Unitree and UBTECH are building their own embodied-AI stacks.

The goal is straightforward to describe and extremely difficult to achieve: show a robot a new environment, explain what needs to be done and have it figure out the physical details without weeks of engineering.

We are nowhere near consistent human-level performance on that test. Unitree founder Wang Xingxing recently said the breakthrough where robots can reliably perform most unfamiliar tasks from natural-language instructions could still be five to ten years away, although he left open the possibility of faster progress.

Interact Analysis is cautious for the same reason. Its latest humanoid study says autonomy, sufficient ROI and multi-task generalization still form a difficult combination and places the broader commercial inflection point after 2032.

Tesla offers another reality check. The company is building Optimus production infrastructure, but its Q2 disclosure removed the earlier language saying Optimus would reach volume production in 2026. Tesla now says the first Fremont units will initially be used for training-data collection and functionality development.

Hardware is becoming plentiful surprisingly fast. Intelligence is still the part that decides whether all those machines become workers.

Where is humanoid robot demand actually coming from?

Humanoid robot demand currently comes from research, entertainment, manufacturing, logistics and data collection, with factories and warehouses becoming more important but still representing a minority of shipments.

Counterpoint's latest breakdown puts entertainment and performance at roughly 34% of H1 shipments, data production and research at 27%, service and guidance at 19%, intelligent manufacturing at 13% and warehousing and logistics at about 5%.

The customer examples underneath those percentages are becoming more diverse. AgiBot sells smaller humanoids into entertainment and research while placing its G-series robots on electronics and automotive-component lines. Galbot has robots working around smart pharmacies and has tested an industrial model with CATL. UBTECH says more than 80% of its full-size humanoid sales are now tied to industrial scenarios covering automotive, logistics, electronics, semiconductors, aerospace and industrial data collection.

In the U.S., commercial activity remains more concentrated around logistics and manufacturing. BMW is using Figure, GXO is using Agility, Mercedes-Benz has tested Apptronik, Toyota is working with Agility, and Catalyst Brands is bringing Figure into distribution.

Homes remain much further away. Several companies are developing household robots, but the home combines children, pets, stairs, fragile objects, clutter and constantly changing tasks. A factory can redesign a workflow around a robot. A household expects the robot to deal with the workflow it finds.

For the next phase of market growth, warehouses, factories and structured commercial environments look far more important than the dream of a robot in every kitchen.

What could stop the humanoid robotics market from growing?

The humanoid robotics market could hit a serious slowdown if productive deployments fail to catch up with the enormous amount of manufacturing capacity, funding and policy support entering the industry.

The first pressure point is simple economics. A robot that costs six figures and works at half human speed can still be useful in some jobs, but the addressable market shrinks quickly once supervision, maintenance and downtime are added.

Demand quality is another concern. China's government-backed training centers are helping manufacturers build scale and collect valuable data, but that spending cannot substitute forever for private customers buying robots because the payback works.

Capacity could become painful as well. Figure, Agility, UBTECH, AgiBot, Unitree, Hyundai, Tesla and others are all preparing for much larger production. Orders have to rise fast enough to absorb those machines.

Then there is the intelligence problem. A market of robots that can repeat a narrow factory movement can become sizeable. The trillion-dollar visions require robots that can learn new jobs, cope with surprises and work safely around people without constant engineering support.

Valuations make any disappointment more visible. A $50 billion robotics company does not merely need a good robot. It needs a very large future business.

None of those problems currently points to shrinking demand. They tell us where the growth could break if the technology fails to keep up with the money.

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

So, is the Humanoid Robotics Market growing now?

Yes. The Humanoid Robotics Market is growing extremely fast right now, but productive commercial deployment is still several steps behind the headline growth in shipments, funding and manufacturing.

The physical expansion is difficult to argue with. Interact Analysis saw production rise roughly tenfold in 2025. Counterpoint now estimates H1 shipments above 22,000, almost 300% higher year over year, and expects more than 50,000 for the full year. Chinese vendors are shipping robots by the thousands. UBTECH has turned full-size humanoids into a business generating more than RMB 800 million of annual revenue. Agility has built a contracted order book above $300 million. Figure is manufacturing hundreds of its latest-generation robot and installing capacity for thousands.

Commercial evidence has improved too. BMW has accumulated more than a thousand operating hours with Figure. GXO has moved more than 100,000 totes with Digit. Customers are testing second generations and new tasks rather than abandoning humanoids after their first experiment.

The restraint comes from what those figures still leave out. More than 60% of Counterpoint's latest shipments remain tied to entertainment, research and data production. Interact Analysis found that only around 10% of 2025 production reached clearly productive real-world applications. China's state-supported training ecosystem is also boosting demand before normal commercial economics have been proved.

So our answer is a strong yes, with a specific meaning. Humanoid robotics is already a hypergrowth hardware and investment market. The smaller market for humanoids doing economically productive work is growing quickly too, from a tiny base, but mass labor substitution has not arrived.

The next milestone is no longer whether companies can build humanoid robots. They clearly can. What we need to see now is hundreds of independent customers buying more robots after the first deployment because the machines save enough money to justify the next order.

If that begins happening at scale, today's boom will look like the start of a major automation market. If it does not, shipment growth will eventually collide with the much smaller number of jobs these robots can actually perform.

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.