Who will sell the first million humanoid robots?

Who will sell the first million humanoid robots?

AgiBot is our best pick today to become the first company to sell one million humanoid robots, with Unitree close behind and Tesla carrying the biggest upside if Optimus suddenly works at automotive scale.

That ranking looks very different from the attention each company gets. According to Counterpoint Research’s latest global shipment tracking, AgiBot shipped around 9,700 humanoid robots in the first half of 2026, giving it more than 43% of the market. Unitree shipped more than 7,000 and held about 31%. Together, two Chinese companies controlled roughly three quarters of global humanoid shipments.

Tesla still sits in another category. Its Fremont Optimus line is being designed around eventual annual capacity of one million robots, an output level that could overwhelm every startup in the market. But capacity only becomes relevant once Tesla has a robot that outside customers actually want to buy in huge numbers.

So our choice comes down to three very different paths. AgiBot already has the largest commercial volume. Unitree could reach much broader markets by making humanoids dramatically cheaper. Tesla could arrive later and crush both on manufacturing scale. Right now, AgiBot has done the most to earn the lead rather than announce it.

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

What should count as selling one million humanoid robots?

We should count humanoid robots delivered to paying external customers, because production targets, reservations and robots used inside the manufacturer’s own factories can make this race look much further along than it really is.

Unitree gives us a useful benchmark. When confusion spread about its 2025 numbers, the company explicitly clarified that more than 5,500 humanoids had been “actually sold and delivered to end customers.” It separately reported production above 6,500 units. That distinction is exactly what we need.

AgiBot requires a little more care. Its factory announced the rollout of its 15,000th embodied robot in June 2026, while Counterpoint estimated 9,700 humanoid shipments during the first half of the year. Production and shipment figures therefore should not be treated as interchangeable.

Tesla creates an even bigger distortion. A factory designed to make one million Optimus robots per year sounds as though Tesla is already approaching the finish line. Yet the company is still building that manufacturing system, and its first large fleets are expected to support Tesla’s own development and data collection.

For the rest of this analysis, a million means cumulative humanoid robots commercially delivered outside the manufacturer.

Metric	Count toward one million?	Why
Robot delivered to a paying customer	Yes	Commercial sale completed
Robot sold through a distributor	Yes	Commercial channel delivery
Signed order	No	Delivery may never happen
Reservation or deposit	No	Demand indication only
Robot produced but still in inventory	No	No customer delivery yet
Robot used by its manufacturer	No	Internal deployment
Announced factory capacity	No	Production capability, not sales

Why does one million humanoid robots suddenly sound possible?

Selling one million humanoid robots has become a serious question because the market has moved from prototype-scale volumes into tens of thousands of annual shipments, while manufacturing costs and entry prices are falling at the same time.

Counterpoint estimates global humanoid shipments exceeded 22,000 units during the first half of 2026, up almost 300% from a year earlier. It expects full-year shipments to pass 50,000. Even if those forecasts prove optimistic, the market is already several times larger than it was only a short time ago.

Production ramps are also getting faster. AgiBot says it needed almost two years to reach its first 1,000 robots, around another year to move from 1,000 to 5,000, and just three months to go from 5,000 to 10,000. By June, its 15,000th embodied robot had rolled off the line.

Figure is showing the same acceleration from a much smaller base. Its BotQ factory increased Figure 03 production from roughly one robot a day to one an hour in less than four months, while producing more than 350 units during the ramp.

Meanwhile, Unitree has pushed humanoid pricing far below the six-figure prices people once associated with the category. Its G1 starts around $13,500, while the smaller R1 has been advertised below $5,000 internationally.

The scale is still tiny beside cars or conventional industrial robots. China alone installed around 295,000 traditional industrial robots during 2024. But a humanoid market moving toward 50,000 annual shipments is close enough to industrial scale that one million cumulative sales no longer requires science-fiction assumptions.

Who is actually winning the humanoid robot sales race right now?

AgiBot is currently winning the global humanoid shipment race, and Unitree is the only company close enough to call the contest competitive.

Counterpoint puts AgiBot at roughly 9,700 humanoid shipments in the first half of 2026, compared with more than 7,000 for Unitree. AgiBot therefore shipped roughly four times as many humanoids in six months as the entire global market was shipping annually only a short time earlier.

The ranking holds up across different datasets. Omdia estimated AgiBot shipped 5,168 humanoids in 2025 and took around 39% of the market. Unitree later confirmed more than 5,500 humanoids actually sold and delivered during that same year.

Those sources differ slightly in methodology, which is normal in such a young market. The useful conclusion is much harder to dispute: AgiBot and Unitree are already shipping thousands of humanoids, while most American competitors are still measured in hundreds of robots, pilot fleets or future capacity.

UBTech is worth separating from that group because its industrial commercialization is already substantial. The company says it sold 1,079 full-size embodied humanoids in 2025 and generated RMB820 million from them. More than 80% went into manufacturing, logistics, electronics, semiconductor, aerospace and industrial-data applications.

Figure had produced more than 350 Figure 03 robots by late April 2026. Agility Robotics has disclosed a multi-year agreement covering 1,000 Digit robots. Both look serious, but neither is currently in the same unit-volume league as AgiBot or Unitree.

Company	Latest useful volume evidence	Current position
AgiBot	~9,700 H1 2026 shipments	Clear volume leader
Unitree	>7,000 H1 2026 shipments	Close second
UBTech	1,079 full-size embodied units sold in 2025	Strong industrial seller
Figure	>350 Figure 03 produced during early ramp	Fast manufacturing buildout
Agility Robotics	1,000-unit multi-year agreement disclosed	Real enterprise demand, smaller scale
Tesla	Commercial external ramp still ahead	Huge potential capacity, little sales evidence

Are AgiBot’s thousands of humanoid robot shipments actually meaningful?

AgiBot’s lead is commercially meaningful, although a large share of today’s humanoid demand still comes from research, data collection and entertainment rather than workers replacing humans on factory floors.

Counterpoint estimates entertainment and performance plus data-production and research applications still accounted for more than 60% of global humanoid shipments during the first half of 2026. Intelligent manufacturing represented around 13%, and warehousing and logistics another 5%.

That mix explains part of AgiBot’s rapid growth. Counterpoint says the company’s X-series robots are widely used in China’s entertainment-performance rental market, as well as research, education and data production. Those buyers can justify purchasing robots without proving that each machine saves tens of thousands of dollars in labor every year.

AgiBot is also building a more industrial business. Its G2 robots have been deployed in clustered production-line applications through partners including Longcheer Technology and Joyson Electronics. At its latest production milestone, AgiBot showed a G2 working in a tablet-production quality-inspection process alongside factory employees.

The takeaway is fairly specific. AgiBot has already learned how to manufacture, distribute and support robots by the thousands, which creates a real head start. We should be more cautious about interpreting those thousands as proof that general-purpose humanoid labor already has mass-market economics.

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

Could Unitree win by making humanoid robots ridiculously cheap?

Unitree may have the strongest route to one million sales if low-cost humanoids become a much bigger category than expensive industrial workers.

The price difference is already enormous. Unitree lists the G1 from around $13,500, and its smaller R1 has appeared at roughly $4,900 before configuration, shipping and other costs. Many competing full-size industrial humanoids are expected to cost several times more.

A $5,000–$15,000 robot opens completely different buying behavior. Universities can buy multiple machines for teaching and research. AI companies can build data-collection fleets. Developers can experiment without securing a six-figure equipment budget. Entertainment businesses can rent robots out. Eventually, some consumers can buy one simply because they want one.

Unitree has also proved that its lower prices can coexist with real manufacturing volume. The company delivered more than 5,500 humanoids to customers in 2025 and produced more than 6,500. Counterpoint then estimated more than 7,000 shipments during the first half of 2026 alone.

This gives Unitree a path that Figure, Agility and most Western humanoid startups currently cannot copy. It can grow the market downward by reducing the amount of economic usefulness a robot needs before someone is willing to buy it.

The biggest question is whether cheap humanoids remain mostly research and entertainment products or become genuinely useful machines. If capability catches up while Unitree keeps its cost advantage, one million cumulative units could arrive surprisingly fast.

Can Tesla jump from almost no Optimus sales to one million robots?

Tesla can still win this race because its planned Optimus manufacturing scale is so much larger than everyone else’s, but today we would not put it ahead of companies already delivering thousands of humanoids.

Tesla is converting former Model S and Model X production space in Fremont for Optimus manufacturing and has described the first major line as being designed around eventual capacity of one million robots annually. Future Texas capacity has been discussed at an even larger scale.

That changes the arithmetic completely. AgiBot might enter a year with a cumulative lead of tens of thousands of robots, yet Tesla running a 500,000-unit annual Optimus line could erase that gap within weeks.

Tesla also has manufacturing experience no humanoid startup can match. It already manages huge supplier networks, high-volume electronics, batteries, motors, castings, logistics and large automated factories. The company produced hundreds of thousands of vehicles during the first half of 2026 and has tens of billions of dollars of liquidity available for capital spending.

Still, Optimus brings manufacturing problems that vehicle experience does not magically solve. Humanoid hands need compact actuators, tactile sensing and precise control in very little space. Joints need high torque, low backlash and long operating lives. McKinsey recently identified actuators, force-and-torque sensing and some perception components as potential bottlenecks for the whole industry.

Tesla also needs demand. Cars arrive with an obvious customer job: transportation. A million humanoid robots need a million reasons to be useful.

For now, Tesla has the most explosive upside in the race. AgiBot and Unitree have much stronger evidence that customers are already paying for their machines.

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

Why are Chinese humanoid robot companies scaling so much faster?

Chinese humanoid manufacturers currently benefit from a dense supply chain built around electric vehicles, industrial automation and consumer electronics, and that advantage becomes more important as production moves from thousands toward hundreds of thousands of robots.

McKinsey’s latest humanoid supply-chain work shows how deep that base is. China accounts for about 40% of precision-bearing capacity in the relevant supply chain, around 40% of encoder and position-sensor capacity, roughly 30% of driver boards and power electronics, and about 90% of permanent-magnet processing.

The geography helps as well. Robotics manufacturing clusters around Hangzhou, Ningbo, Shenzhen and Suzhou allow companies to source motors, gear systems, CNC housings, circuit boards, sensors and battery assemblies within tightly connected supplier networks.

AgiBot’s production curve shows what that can look like in practice. Its move from 5,000 to 10,000 robots took only three months after the company had needed around a year to move from 1,000 to 5,000. Unitree has reached thousands of annual humanoid deliveries while selling its entry products at prices many Western companies cannot approach.

The surrounding industrial base is already enormous. China had roughly 2.03 million conventional industrial robots operating in factories after installing another 295,000 during 2024, according to International Federation of Robotics figures cited by McKinsey.

Tesla can build a highly integrated supply chain of its own, and Figure is doing the same for critical modules. China gives dozens of companies access to much of that ecosystem before they have built Tesla-sized businesses. In a race where every actuator has to be made millions of times across thousands of robots, that is a serious structural advantage.

Are factories actually ready to buy hundreds of thousands of humanoid robots?

Factories are becoming credible humanoid customers, but current deployments still fall far short of the fleets needed to carry one manufacturer to one million sales.

BMW’s Figure 02 trial is one of the strongest real examples. According to BMW, the robot spent about 1,250 operating hours in its Spartanburg plant over ten months, worked ten-hour shifts five days a week, moved more than 90,000 sheet-metal components and supported production of more than 30,000 BMW X3 vehicles. The job required repeatable positioning with millimeter-level precision.

UBTech offers another useful test because its commercial volume is already concentrated in industry. The company says more than 80% of its 1,079 full-size embodied humanoids sold in 2025 went into areas including automotive production, logistics, electronics, semiconductors, aerospace and industrial data collection.

Agility Robotics has moved Digit into commercial environments with companies including GXO, Schaeffler and Toyota, while one disclosed multi-year agreement covers 1,000 robots.

These are much stronger examples than a robot folding laundry on stage. They also show how far the market still has to go. A 1,000-unit contract is large enough to stand out today, yet a company would need one thousand contracts of that size to reach one million robots.

Factories can probably provide the first major wave of serious economic demand. Reaching the final few hundred thousand units will be much easier if humanoids also spread into warehouses, retail, hospitality, education and eventually homes.

Are humanoid robots useful enough today for mass adoption?

Humanoid robots can already do valuable repetitive work, but today’s autonomy and reliability are still too narrow to justify millions of deployments.

The BMW example is impressive precisely because the task was specific. Figure 02 repeatedly retrieved and positioned sheet-metal parts. That produced measurable factory value, yet it did not require the robot to wander around the plant discovering arbitrary jobs by itself.

The pattern repeats elsewhere. Agility focuses heavily on material movement. UBTech emphasizes manufacturing workflows. AgiBot is putting its G-series robots into defined production processes. Companies keep choosing jobs where environments, objects and actions can be constrained.

Homes are much harder. A household robot might encounter a half-open drawer, an unfamiliar detergent bottle, a child leaving an object on the floor, a pet crossing its path and dozens of objects it has never manipulated before. Human environments generate endless exceptions.

That gap between controlled usefulness and general autonomy is probably the biggest obstacle to a million commercial sales. Hardware factories can scale surprisingly quickly once designs stabilize. Intelligence, reliability and recovery from mistakes may improve much less predictably.

If those capabilities stall, specialized machines will keep winning many jobs. China installed around 295,000 conventional industrial robots in 2024 because purpose-built automation already performs welding, assembly, packaging, logistics and machine tending extremely well.

Humanoids therefore need more than impressive movement. The robot eventually has to become cheaper or more flexible than the alternative for enough tasks that customers keep ordering fleets after the pilots finish.

Could research and entertainment sell more humanoid robots than factories?

Research, education, entertainment and AI-data work are currently buying far more humanoid robots than factories, which helps explain why AgiBot and Unitree have opened such a large early lead.

Counterpoint estimates those research, data-production, entertainment and performance applications represented more than 60% of global shipments in the first half of 2026. Manufacturing accounted for about 13%.

That mix changes what “commercial success” looks like. A university does not need to calculate a three-year labor payback before buying a Unitree G1. A robotics lab wants a physical platform for training algorithms. An AI company may need hundreds of machines generating manipulation data. An event operator can earn money by renting a robot for performances.

AgiBot has leaned directly into this market. Counterpoint describes its X-series as widely adopted in China’s entertainment-performance rental business, alongside research, education and data-production applications. The company has lately expanded its robot-rental model outside China as well, including markets in Europe and the Middle East.

These categories could carry the industry through tens of thousands and eventually hundreds of thousands of robots. We are less convinced they can absorb one million machines by themselves.

The likely progression is that research and entertainment keep building early manufacturing volume while industrial deployments become much larger. Consumer or broad commercial use would then make the final stretch much easier.

Does someone need to crack the home humanoid robot market first?

A company can probably reach one million humanoid sales without dominating homes, but a genuinely useful household robot would dramatically shorten the race.

The addressable market changes once individual consumers start buying. Large manufacturers might eventually operate thousands of humanoids per site, yet there are hundreds of millions of middle- and high-income households globally. A tiny penetration rate would create more demand than today’s entire humanoid industry.

Price is already moving toward consumer territory faster than capability. Unitree can sell basic humanoids for thousands rather than hundreds of thousands of dollars. Other companies are building specifically for home use. Tesla has repeatedly described household work as part of the long-term Optimus opportunity.

The harder problem is getting a robot to perform enough household tasks reliably that owners keep using it after the novelty disappears. Factories can redesign a workstation around a robot. A home will not be rebuilt every time the machine encounters a strange cupboard.

That makes home adoption the biggest upside scenario rather than something we need for our base case. AgiBot or Unitree could accumulate one million sales through several commercial categories over enough years. Whoever makes household humanoids genuinely useful could reach the same number much faster.

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

How fast would AgiBot or Unitree have to grow to reach one million humanoid sales?

AgiBot or Unitree probably needs several more years of roughly 60–100% annual growth before one million cumulative humanoid sales becomes realistic.

Suppose a company enters the next phase shipping 20,000 humanoids a year. If annual shipments never grow, selling one million takes 50 years. At 50% annual growth, cumulative sales are still only around 640,000 after seven years. At 60%, the figure reaches roughly 860,000.

An 80% growth rate changes the picture. The same company would pass one million cumulative robots during roughly the seventh year. Doubling shipments each year gets there around year six.

Those numbers sound aggressive, but the current market is growing from an exceptionally small base. Counterpoint’s latest estimate has global shipments up almost 300% year over year in the first half of 2026. Nobody should extrapolate that rate indefinitely, although it shows why decade-long forecasts can become huge so quickly.

Production learning can also compound. As seen above, AgiBot needed about a year to move from 1,000 to 5,000 robots and only three months to add the next 5,000.

The million-unit winner therefore does not need to be producing one million robots annually anytime soon. Sustained growth from tens of thousands into hundreds of thousands can get one company over the cumulative line.

Starting annual shipments	Annual sales growth	Rough cumulative result after 7 years
20,000	0%	140,000
20,000	50%	~640,000
20,000	60%	~860,000
20,000	80%	~1.5 million
20,000	100%	~2.5 million

Can Figure, UBTech, Agility or XPeng still beat AgiBot, Unitree and Tesla?

Figure, UBTech, Agility and XPeng can all become major humanoid companies, but none currently has a better one-million-unit path than our top three.

Figure has probably produced the strongest combination of sophisticated autonomy and documented Western factory work. BMW’s Spartanburg trial gives it real operating evidence, and BotQ has already accelerated Figure 03 manufacturing from about one unit per day to one per hour. Figure says the first BotQ line is designed for up to 12,000 robots annually and has a goal of producing 100,000 over four years.

That is a serious manufacturing plan. AgiBot may already be operating around that order of magnitude today.

UBTech deserves more attention than it usually gets outside China because its 1,079 full-size embodied humanoid sales in 2025 generated RMB820 million, and more than 80% went into industrial uses. It also sells much larger volumes of smaller and less autonomous humanoid products. The company has therefore proved customers will pay for both premium industrial robots and cheaper humanoid formats.

Agility has one of the clearest enterprise-focused strategies, but RoboFab’s planned capacity of roughly 10,000 Digit robots per year remains small beside the output eventually required for a million-unit race.

XPeng is the most dangerous later entrant. Its robotics business recently raised more than $900 million at a valuation above $6.3 billion. XPeng expects IRON to enter mass production around the end of 2026, followed by broader deliveries in China and overseas markets in 2027. It also brings automotive supply chains, manufacturing expertise and AI development under one corporate roof.

We would watch XPeng more closely than a hypothetical startup nobody has heard of yet. Still, every quarter that AgiBot and Unitree ship thousands of robots increases the amount of catching up required.

What could stop the humanoid robot market before anyone sells one million?

The million-unit race falls apart if customers discover that specialized robots can automate most valuable jobs more cheaply and reliably than humanoids.

That competition already exists at huge scale. China now operates roughly two million conventional industrial robots. Factories installed another 295,000 there during 2024 alone, according to International Federation of Robotics data highlighted by McKinsey.

Those machines have several advantages. A fixed robotic arm does not need to balance. An autonomous mobile robot does not need legs. A purpose-built warehouse system can move goods all day without manipulating hundreds of unfamiliar objects. Simpler machines usually cost less to build and maintain.

Humanoid form wins when the surrounding environment was designed for humans and changing that environment would be expensive. Stairs, shelves, doors, workstations and tools all become arguments for human-shaped machines. Flexibility also becomes valuable if one robot can move between several jobs instead of performing the same motion forever.

The other major risk is reliability. A humanoid that completes a task 95% of the time sounds impressive in a demo. In a factory running thousands of cycles, a 5% failure rate can be disastrous. Service costs, remote intervention and downtime can quickly destroy the labor-saving calculation.

We are therefore much more confident that annual humanoid sales will keep growing from today’s small base than we are about the exact year someone reaches one million. The milestone depends on the machines becoming economically boring: reliable enough that companies order another thousand without needing a technology showcase around the deployment.

Who will actually sell the first million humanoid robots?

AgiBot is the company we would bet on today to sell the first million humanoid robots, with Unitree close enough that a sustained price advantage could flip the ranking and Tesla capable of overtaking both if Optimus reaches true automotive-scale production.

AgiBot has the strongest current starting point. Counterpoint’s latest tracking puts it at roughly 9,700 humanoid shipments in the first half of 2026 and more than 43% global share. Its product range already covers research, entertainment, services and industrial applications, while its production milestones show a manufacturing ramp that keeps accelerating.

Unitree has the more disruptive price curve. More than 5,500 humanoids were actually sold and delivered in 2025, and Counterpoint estimates more than 7,000 shipments in only the first half of 2026. Selling robots from roughly $5,000 to $15,000 gives Unitree access to buyers that cannot justify premium industrial machines. If cheap humanoids improve fast enough, Unitree could eventually have the broadest market of anyone.

Tesla remains the hardest company to rank. Fremont is being built around a scale that could make today’s entire humanoid market look microscopic. Tesla also has money, automotive manufacturing experience and enormous internal environments where it can train and test Optimus. What we still lack is meaningful evidence of external customers buying Optimus robots in volume.

Figure, UBTech, Agility and XPeng belong in the next group. Figure has some of the best real factory evidence. UBTech already sells expensive industrial humanoids. XPeng brings the manufacturing base of a large EV company and has just raised more than $900 million for robotics. Any of them could move up quickly.

But the lead today belongs to AgiBot. It has already moved beyond thousands of prototypes into thousands of shipments while the overall market is approaching tens of thousands of annual units. That head start does not guarantee one million.

The real race starts when annual volumes move past 100,000. AgiBot currently looks most likely to get there first.

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.