Factory humanoids: which startup is ahead?

Are humanoid robots actually working in factories now?

Humanoid robots are already doing real factory work, and Agility Robotics, Figure AI, UBTECH, and AgiBot have gone considerably further than companies still showing prototypes.

Figure's robots helped BMW manufacture more than 30,000 X3 vehicles at its Spartanburg plant. Agility's Digit robots have accumulated over 65,000 hours of real-world operation. In China, UBTECH sold 1,079 full-size humanoids in 2025, while AgiBot reached 15,000 robots produced by June 2026.

These achievements have different meanings. Figure has detailed evidence of production-line work. Agility has operated across several customer facilities and signed commercial agreements. AgiBot has demonstrated much greater manufacturing volume, while UBTECH has disclosed substantial revenue from humanoid products.

The real question is which company can turn these achievements into thousands of robots doing paid, productive factory work. For now, no startup has publicly proved that level of deployment.

Who's actually winning the factory humanoid robot race right now?

Agility Robotics currently has the strongest overall claim to industrial humanoid leadership, although Figure AI is particularly strong in automotive production and AgiBot is far ahead in reported manufacturing volume.

We put more weight on sustained customer operations than on robot production announcements. Agility has commercial relationships with GXO, Schaeffler, and Toyota Motor Manufacturing Canada, alongside more than 65,000 operating hours. Figure has detailed production results verified by BMW, but its publicly documented customer footprint is narrower.

AgiBot and UBTECH complicate the comparison. Both have moved substantial numbers of humanoids into the market, although their published figures cover different robots and uses. We cannot assume that every unit sold or manufactured is working on a factory floor. Numbers that mix different robots blur the picture, so we compare who is in front and who is slipping in our report on humanoid robots.

There's also a big difference between signing a customer and completing a deployment. A paid agreement establishes demand. Measured throughput tells us what a robot has actually accomplished.

Startup	Main robot	Strongest evidence	Where it stands
Agility Robotics	Digit	65,000+ operating hours	Strongest commercial deployment record
Figure AI	Figure 02/03	BMW production-line results	Strongest documented automotive precision work
AgiBot	G2 and other models	15,000 cumulative robots produced	Manufacturing volume leader
UBTECH	Walker series	1,079 full-size units sold in 2025	Strong financial commercialization evidence
Apptronik	Apollo	Mercedes-Benz and Jabil partnerships	Major industrial challenger
NEURA Robotics	4NE-1	Industrial partnerships and development programs	Commercial deployment still developing
Unitree	G1/H1/H2	Commercially available hardware	Limited proven factory productivity
Fourier Intelligence	GR series	Humanoid development and commercial programs	Limited sustained factory operating evidence

Is Agility Robotics really ahead of Figure AI in factories?

Agility Robotics is ahead of Figure AI in disclosed commercial deployment breadth, but Figure has delivered some of the most convincing production-line results in the humanoid industry.

Agility's first major commercial breakthrough came in June 2024, when GXO signed a multiyear Robots-as-a-Service agreement following a successful pilot. Digit moved containers between automated equipment at GXO's Flowery Branch warehouse in Georgia. The company subsequently expanded its relationships with Schaeffler and Toyota Motor Manufacturing Canada.

Toyota's February 2026 agreement is particularly interesting. After roughly a year of testing, Toyota contracted seven Digit robots for its Canadian automotive operations. The machines are intended to unload containers of parts from automated transport equipment.

Meanwhile, Figure's BMW project demonstrated sustained handling of sheet-metal components on an active automotive production line. BMW independently confirmed the deployment's duration, task, and output.

Agility's commercial case strengthened further with over $300 million in multiyear Digit 5 orders, disclosed in its SEC-filed investor materials. Those commitments concern approximately 1,000 robots under three-year service contracts, subject to milestones and product requirements.

The orders have not all become deployed robots or recognized revenue. Still, Agility has built a broader combination of industrial experience, customer agreements, and future commitments.

How much factory work has Figure AI actually done for BMW?

Figure AI has proved that its humanoid can perform repetitive work on a real automotive production line, handling more than 90,000 sheet-metal components during its BMW program.

According to BMW's account of the Spartanburg project, Figure 02 completed approximately 1,250 operating hours and took around 1.2 million steps. The program supported the production of more than 30,000 BMW X3 vehicles, with the robots working ten-hour shifts during the full deployment period.

Figure 02 picked sheet-metal components from containers and positioned them for welding. BMW reported that the machines could repeatedly position parts with millimeter-level precision. Conventional industrial robots subsequently handled the welding.

The operating figures work out to roughly 72 components per reported robot-hour. That's a useful measure of the work completed, although it includes operational time rather than representing the robot's maximum instantaneous handling speed.

Figure also disclosed a concrete engineering problem uncovered during the deployment: the forearm was its biggest source of hardware failures. The company redesigned the wrist electronics for Figure 03, removing an internal communications board and simplifying the wiring.

BMW's decision to continue with Figure 03 is encouraging. Its June 2026 demonstration introduced a harder logistics-sequencing task involving sorting components and pulling a wheeled cart. However, Figure has not yet published comparable long-term throughput figures for that newer operation.

Does Agility's 65,000 operating hours prove Digit is more reliable?

Agility's Digit has accumulated far more publicly disclosed industrial operating time than Figure's BMW robots, giving Agility a valuable reliability-testing advantage.

When Agility introduced Digit 5 in September 2026, it reported more than 65,000 hours of real-world experience across commercial operations. That's over 50 times the approximately 1,250 hours documented for Figure 02 at BMW.

The gap tells us how much operating experience each company has disclosed, although Agility's number combines multiple robots and facilities. We cannot use those two figures to calculate which individual machine fails less often. Missing reliability data makes comparisons hard, and we go through the biggest unsolved problems for humanoids in our report covering humanoid robots.

What Agility has learned from those hours is more important. Digit has spent years dealing with container handling, industrial traffic, charging requirements, and the everyday interruptions that are difficult to reproduce in a laboratory.

Figure's BMW experience offered a different benefit: detailed feedback from a demanding automotive production process. Its forearm redesign shows how factory failures can directly change the next generation of hardware.

Neither startup publishes enough comparable information about failures per thousand hours, maintenance expenses, or human interventions to declare a winner on reliability alone. Agility has the stronger disclosed operating history. That's the part we can actually establish.

Is AgiBot already bigger than every American humanoid startup?

AgiBot is producing humanoid robots on a scale that American startups have not yet matched publicly, although its production totals cover a broader range of products than industrial factory humanoids.

The Shanghai company reached 5,000 cumulative robots before advancing to 10,000 in March 2026 and 15,000 in June. According to AgiBot, the journey from 1,000 to 5,000 robots took about a year, while the next 5,000 required just three months.

Producing another 5,000 in approximately three months suggests an average pace of roughly 1,600–1,700 robots a month across that latest interval.

Figure, by comparison, reported producing more than 350 Figure 03 robots by April 2026. AgiBot's 15,000 cumulative units were more than 40 times that number, although the periods and product mixes are different.

There is also evidence of industrial use. AgiBot's G2 worked on a tablet quality-inspection line at Longcheer Technology, where eight robots were reported to be operating together. AgiBot also disclosed approximately 100 cumulative hours of livestreamed factory operation in late June.

The production numbers establish a major manufacturing advantage. But 100 hours of livestreamed factory operation are still modest compared with the longer industrial programs disclosed by Figure and Agility. We need more evidence before giving AgiBot the factory deployment crown.

Is UBTECH selling more factory humanoids than AgiBot?

UBTECH has the clearer financial record for humanoid sales, while AgiBot leads on reported shipments across a broader portfolio.

UBTECH sold 1,079 full-size humanoid robots in 2025, generating RMB820.6 million from full-size humanoid products and services. AgiBot, meanwhile, shipped 5,168 humanoids that year, according to Omdia figures cited in its June 2026 announcement, representing a reported 39% of global shipments.

The latest UBTECH interim report makes the comparison more interesting. During the first half of 2026, the company sold another 921 full-size humanoids, up from approximately 45 in the same period a year earlier. Revenue from the category increased from RMB38.2 million to RMB590.3 million.

That's growth of approximately 20 times in unit volume and more than 15 times in revenue. The difference suggests that the mix of products and services sold is changing as deliveries accelerate.

Dividing humanoid segment revenue by units sold gives about RMB761,000 per unit in 2025, compared with RMB641,000 in the first half of 2026. These are blended revenue-per-unit calculations, not actual robot selling prices, because the segment includes services.

UBTECH's full-size humanoids also serve commercial, educational, and other markets. AgiBot's shipments similarly cover multiple configurations. Neither disclosure tells us exactly how many robots are working full shifts in manufacturing plants.

Is UBTECH finally making money from humanoid robots?

UBTECH's humanoid business is growing extraordinarily fast, but the company remains unprofitable despite a sharp improvement in its financial results.

According to its interim results released in August 2026, total revenue doubled to RMB1.27 billion in the first half of the year. Full-size humanoid products and services generated RMB590.3 million, accounting for 46.5% of company revenue, compared with just 6.1% a year earlier.

That shift is striking. In twelve months, full-size humanoids went from a relatively small business to nearly half of UBTECH's sales.

The rest of the financial report gives us a clearer picture of the transformation. Other smart robot products and services declined 18.8% year over year, while total gross profit rose 160.9% and the company's gross margin increased from 35.0% to 44.7%.

UBTECH's net loss narrowed from RMB440.0 million to RMB338.8 million, a 23% improvement. The humanoid expansion is coinciding with better overall margins and smaller losses, although we cannot attribute the entire improvement to humanoid robots alone.

UBTECH is one of the few humanoid companies whose commercialization can be tracked directly through detailed financial statements. That's a substantial advantage for anyone trying to judge the business rather than just the technology. Factory-level profitability, though, remains unclear.

Could Apptronik overtake Figure and Agility in factory robots?

Apptronik could become a major competitor to Figure and Agility, especially through its relationships with Mercedes-Benz and Jabil, but Apollo still needs more public evidence of sustained factory work.

Mercedes-Benz began testing Apollo for repetitive industrial tasks, including moving components through manufacturing facilities. Jabil's partnership, announced in February 2025, goes further by combining potential robot manufacturing with factory testing. Partners like Mercedes-Benz and Jabil improve a challenger's odds; we look at who has the best shot at leading in our report on humanoid robots.

That arrangement gives Apptronik something valuable: a manufacturing partner that can help refine how Apollo is assembled while experimenting with the robots inside actual production environments.

The financial backing is substantial too. Apptronik announced a $520 million Series A extension in February 2026, bringing the round to more than $935 million. Its backers include Google, Mercedes-Benz, and John Deere.

Those relationships give Apptronik access to technical expertise and potential customers across automotive, electronics, and industrial equipment.

But we haven't seen Apollo publish an operating record comparable to Digit's cumulative hours or Figure's component-handling results at BMW.

Apptronik has the partners and capital needed to compete seriously. What it needs now is a customer deployment with hard productivity numbers. Until then, putting Apollo above Figure or Agility would be premature.

Are Chinese startups beating the US at building humanoid robots?

Chinese humanoid startups are ahead in reported manufacturing volume, although American companies such as Agility and Figure have published stronger evidence for certain sustained industrial jobs.

AgiBot's production increase provides the clearest manufacturing example. Its reported output grew from 5,000 to 15,000 cumulative robots in roughly half a year. UBTECH has also moved beyond small pilot quantities, reporting almost as many full-size humanoid sales in the first half of 2026 as during all of 2025.

American manufacturers are scaling too. Figure reported that its BotQ facility improved Figure 03 production cycle time from one robot per day to one per hour in less than 120 days, a 24-fold improvement.

One robot per hour is a demonstrated manufacturing cycle time, however, rather than proof that the factory continuously produces 24 robots every day.

Agility's RoboFab facility was designed for eventual capacity of up to 10,000 units annually, but published design capacity cannot be counted as actual production.

The contrast is becoming sharper: China has established substantial manufacturing and shipment volume, while the leading US startups are building commercial relationships around particular industrial applications.

Company	Reported figure	Period	What it tells us
AgiBot	5,168 units shipped	2025	Broad humanoid shipment volume
AgiBot	10,000 cumulative units	March 2026	Manufacturing scale
AgiBot	15,000 cumulative units	June 2026	Faster production ramp
UBTECH	1,079 full-size units sold	2025	Commercial sales volume
UBTECH	921 full-size units sold	H1 2026	Rapid sales growth
Figure AI	350+ Figure 03 units	April 2026	New-model production
Figure AI	One robot per hour	April 2026	Demonstrated production cycle time
Agility Robotics	Up to 10,000 units/year	Planned capacity	RoboFab manufacturing target

Which humanoid robot can actually handle the hardest factory jobs?

Figure AI has some of the strongest evidence for precise automotive manipulation, while Agility's Digit is more thoroughly proven in repetitive industrial logistics.

BMW's Spartanburg deployment required Figure 02 to pick and position sheet-metal components accurately enough for robotic welding. Its newer Figure 03 program tackles sequencing, where components may arrive in different orientations and the robot needs to adjust its hands, body, and position while working.

Agility has concentrated on tasks such as transferring containers between automated systems. These jobs require repeatable handling, reliable movement, and integration with existing logistics equipment.

AgiBot's G2 has shown another type of industrial work through tablet quality inspection, where robots must interact with electronics production processes. UBTECH's Walker robots target material handling, sorting, and inspection.

The complexity differs considerably. A robot can be excellent at moving standardized containers while struggling with thin, irregularly positioned metal components. Similarly, handling a part accurately doesn't mean a machine can inspect electronics or navigate a busy warehouse.

Figure currently has the more convincing public demonstration of complex automotive manipulation. Agility's strength is the longer record of deployed logistics work. If we had to choose one for precision automotive handling, Figure would be our pick. For established industrial logistics work, we'd choose Digit.

Can factory humanoids work entire shifts without human help?

Factory humanoids can now complete long periods of repetitive autonomous work, but no leading startup has publicly established that its robots can handle every part of a factory shift independently.

Figure's BMW deployment included ten-hour scheduled shifts during its production phase. Agility's Digit has worked in real logistics facilities over an extended period, including GXO's warehouse operations.

The missing information concerns what happens within those shifts. Companies rarely disclose comparable figures for time spent doing productive work, waiting, recharging, receiving assistance, or recovering from errors.

That distinction is particularly relevant to the newest robots. Figure 03's Helix 02 system coordinates movement and manipulation, allowing the robot to reposition itself while handling objects. Agility's Digit 5 includes an independent safety controller and a redesigned battery system.

Agility announced a planned 90-minute battery runtime with a nine-minute recharge for Digit 5, alongside approximately 22.7 kilograms of payload capacity. Early customer access is expected in 2027, so these specifications should be separated from the proven performance of Digit 4.

BMW's earlier Figure 02 project also required additional safety barriers, showing that working alongside employees still brings practical restrictions.

Factories have demonstrated that humanoids can operate for meaningful periods. Reliable full-shift independence across changing tasks remains unproven.

Are humanoid robots cheaper than factory workers yet?

Humanoid robots have not yet been shown to beat human workers on total operating cost across a broad range of factory jobs, although Agility's disclosures give us a useful starting point.

In its 2026 investor presentation, Agility illustrated approximately $500,000 in revenue to the company over five years under a Robots-as-a-Service arrangement. That averages $100,000 annually, although the actual contract payment schedules may differ. Renting robots by the year instead of selling them is a pricing choice; we compare which business models are winning in our humanoid robots report.

Another ownership-based illustration showed approximately $400,000 in cumulative revenue, including associated software and services.

These figures describe Agility's commercial assumptions rather than an independently measured customer cost. They don't tell us how many productive hours a robot delivers, how much integration costs, or how often human supervision is needed.

For example, a robot generating $100,000 in annual service payments would cost about $50 per productive hour if it completed 2,000 productive hours a year. At 4,000 hours, the same payment would equal $25 per productive hour. Those calculations exclude additional customer expenses.

The difference shows why utilization is so important. A humanoid that performs repetitive work across multiple shifts has a much stronger economic case than one used for a few hours each day.

Factories also have other automation options. BMW already uses conventional robotic arms and automated transport systems, so humanoids need to justify themselves against existing equipment as well as human labor.

For now, published customer data is too limited to establish a standard payback period for Digit, Figure, Apollo, Walker, or AgiBot's industrial robots.

Does making thousands of humanoid robots guarantee a market lead?

Producing thousands of humanoids gives AgiBot and UBTECH a major manufacturing advantage, but the companies still need to show how many machines customers keep using in productive jobs.

AgiBot's manufacturing ramp suggests that its supply chain, assembly processes, and delivery operations have improved rapidly. Figure is making a similar push through BotQ, while Agility has invested in RoboFab and commercial service infrastructure.

Each strategy tackles a different part of the same problem. A manufacturer needs repeatable hardware production, but customers also need reliable robots, repairs, software updates, and continued support.

UBTECH's financial results show why this distinction deserves attention. Humanoid revenue has increased dramatically, and company-wide margins have improved, yet the business still recorded a substantial loss in the first half of 2026.

Agility's service contracts provide another perspective. Its commitments are meaningful because customers are prepared to enter multiyear arrangements, although deployment milestones and performance requirements still have to be met.

The most valuable data would be the number of robots remaining in paid operation after twelve months, their average productive hours, and the rate at which customers order additional units.

None of the leading companies currently publishes those figures in a sufficiently complete and comparable form.

Which factory humanoid startup could become number one over the next few years?

Agility Robotics is best positioned to build on its current commercial lead, but Figure AI and AgiBot have credible ways to move ahead if they can turn their existing strengths into larger factory deployments.

Agility has already converted trials into commercial relationships, including Toyota's seven-robot arrangement. Digit 5 is designed to improve payload, safety, and battery performance, and the company is targeting availability for a wider set of customers by the end of 2027.

Its SEC-filed investor materials disclosed nine committed customer facility deployments and a goal of reaching up to 25 by the end of 2027. That would nearly triple its committed facility footprint, although the target is still forward-looking.

Figure's opportunity is more concentrated in advanced manipulation. BMW has already tested one generation of its robots in production and demonstrated another in a more complicated logistics workflow. The next milestone would be sustained Figure 03 operations with published throughput across additional facilities.

AgiBot has the opposite challenge. Its manufacturing output is well established, and Omdia estimates give it a leading global shipment share. A larger base of verified industrial customers and operating-hour data could make its commercial position considerably more convincing.

UBTECH also deserves attention. Its 2026 interim results show that full-size humanoids are becoming the core of its business much faster than in 2025. Apptronik could gain ground if Mercedes-Benz and Jabil move from testing to regular production deployments.

The next ranking change will probably come from customer operations rather than a spectacular product demonstration. A company that proves hundreds of robots can perform useful factory jobs consistently would have evidence that today's leaders still largely lack.

So which startup is ahead in factory humanoids today?

Agility Robotics is our overall leader in commercially deployed factory and industrial humanoids, with Figure AI ahead in detailed automotive production-line evidence and AgiBot leading in reported manufacturing volume.

Agility's strongest advantage is the combination of real industrial operating experience and customer commitments. Its work with GXO, Schaeffler, and Toyota shows that Digit has moved beyond isolated trials, while its multiyear orders provide a substantial basis for future deployment.

Figure makes the closest case in automotive manufacturing. As seen above, BMW confirmed that Figure 02 performed a precise, repetitive production task over an extended program. Figure 03 now needs to demonstrate that its more advanced capabilities can deliver similar reliability in regular factory operations.

AgiBot is the clear volume challenger. Its manufacturing scale is exceptional for a startup founded in 2023, and further evidence of robots working consistently in customers' factories could change the competitive ranking quickly.

UBTECH adds an important financial dimension: its latest reported humanoid revenue and sales growth show that commercial demand is increasing, even though profitability remains unresolved.

We favor Agility because its evidence combines actual operations, several enterprise customers, and contracted expansion. Figure is the closest competitor for proven automotive work, while AgiBot has the greatest disclosed manufacturing-scale advantage.

The race is far from settled. But these three companies have already established distinct leads that smaller competitors will need to overcome.