Factory humanoids: which startup is ahead?

Which factory humanoid startups are we comparing?

The factory humanoid race currently has eight relevant startups, but only four have built a serious case through real industrial work, manufacturing scale or large enterprise commitments.

We include Figure, Agility Robotics, AgiBot, Apptronik, NEURA Robotics, Sanctuary AI, Unitree and Persona AI. Each company is developing a human-shaped or human-compatible robot for factories, warehouses, distribution centers or other demanding industrial sites.

We exclude Tesla’s Optimus, Boston Dynamics’ Atlas and XPeng’s IRON because these projects sit inside much larger corporations. We also leave out the publicly listed UBTECH from the formal startup ranking, although its Walker robots remain an important benchmark in China. Companies focused mainly on homes, healthcare or research are outside the comparison unless industrial work is a major part of their strategy.

Funding totals are less precise than they appear. Some companies announce full rounds, while others disclose only selected investments. Chinese startup funding is particularly difficult to compare because several rounds have undisclosed amounts. The figures below show disclosed capital or the best credible minimum we could establish.

Startup	What it is building	Disclosed or estimated cumulative funding
Figure	General-purpose humanoids for factories, logistics and homes	More than $1.7 billion, including a Series C above $1 billion
NEURA Robotics	Cognitive industrial robots and the 4NE1 humanoid	More than $1.4 billion, with the latest round announced at up to $1.4 billion
Apptronik	Apollo humanoids for manufacturing and logistics	Nearly $1 billion
Agility Robotics	Digit humanoids for industrial material movement	More than $641 million before its proposed public transaction
Unitree	Lower-cost humanoids, quadrupeds and robot platforms	About $252 million
Sanctuary AI	Dexterous robots and physical-AI systems for industrial tasks	More than $140 million by credible industry estimates
AgiBot	Humanoids and other embodied-AI robots for industrial and service work	At least $84 million publicly disclosed, with additional undisclosed rounds
Persona AI	Rugged humanoids for shipbuilding and heavy industry	$27 million

Which factory humanoid startup is ahead right now?

Figure is currently ahead overall, but Agility Robotics has closed most of the commercial gap and now sits much closer than Figure’s valuation would suggest.

Figure has the broadest combination of strengths. Its robots have performed quantified automotive work, Figure 03 is back inside BMW, more than 350 units of the latest generation have already been produced, and the company controls both its Helix intelligence system and its BotQ manufacturing line. Figure therefore leads across more parts of the stack than any rival.

Agility leads where the market is most difficult to fake: commercial orders. The company says Digit is deployed with GXO, Schaeffler, Toyota Motor Manufacturing Canada and Mercado Libre. Agility has also announced more than $300 million of multi-year Digit v5 orders, subject to customers reaching contractual milestones, plus a pipeline of more than 30 potential customers. That new disclosure makes Agility a genuine co-leader rather than a distant second.

AgiBot has moved into third place in our ranking. Its production volume is far ahead of most Western startups, and its latest live factory operation finally gives the company measurable industrial evidence rather than another collection of polished robot videos.

Apptronik remains close behind AgiBot. Apollo has excellent partners and almost $1 billion in capital, but Apptronik’s latest public update still describes Apollo 2 mainly as a data-collection and training platform. That is promising work, although it remains one step behind repeated commercial output.

On a simple index where Figure equals 100, we would place Agility around 95, AgiBot near 73 and Apptronik around 68. The first two form a narrow leading group. The gap then becomes much larger.

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

Who has the strongest proof that factory humanoids can do real work?

Figure has the strongest long-duration automotive result, while AgiBot has produced the freshest large-scale factory test and Agility has the cleanest repeated warehouse output.

Figure’s earlier BMW deployment remains the best documented production result from a Western humanoid startup. Figure 02 worked 10-hour weekday shifts, loaded more than 90,000 sheet-metal parts, accumulated over 1,250 operating hours and contributed to the production of more than 30,000 BMW X3 vehicles. Figure also disclosed that the robot’s forearm was the most common hardware failure point, which gives the report more credibility than a flawless promotional case study.

AgiBot has now added evidence that deserves similar attention. During a six-day livestream inside a working tablet factory, several AgiBot humanoids completed 64,828 production tasks across more than four workflows. The company reported over 64 operating hours, a 99.99% task success rate and 17,625 finished tablets passing through the line. The numbers still come from AgiBot, but the continuous livestream made selective editing much harder.

Agility provides a narrower result with a longer commercial history. Digit has moved more than 100,000 totes at GXO’s Flowery Branch facility, transferring containers from autonomous mobile robots onto conveyors. The job sounds simple because it is simple. Factories often begin automation with repetitive, measurable work where failures and successful cycles can be counted clearly.

Apptronik has robots working at its own Robot Park and at partner sites, including Mercedes-Benz and GXO. However, Apptronik has not published an equivalent cumulative task count, operating-hour figure or customer throughput metric. Apollo 2 is currently generating the data that Apptronik and Google DeepMind need to build a stronger autonomous system.

Who has actually sold factory humanoids to customers?

Agility Robotics currently has the strongest commercial traction because it has converted several evaluations into formal deployments and now reports a substantial order book.

The clearest example is GXO. The relationship progressed from a proof of concept to a multi-year Robots-as-a-Service agreement, with Digit becoming part of daily warehouse operations. Agility later added commercial agreements involving Mercado Libre, Toyota Canada and Schaeffler. The company’s reported $300 million order total for Digit v5 gives us the first large contract figure disclosed by a major Western humanoid startup.

Figure has two especially valuable relationships. BMW expanded from Figure 02 production work to a new Figure 03 logistics workflow, while Catalyst Brands signed a commercial agreement covering its distribution network, starting at a facility in Nevada. Figure has not disclosed contract values or fleet sizes, so Agility still has the clearer revenue pipeline.

Apptronik calls its Mercedes-Benz relationship a commercial agreement, but the work itself remains a pilot covering tasks such as delivering assembly kits and inspecting components. Jabil is both a manufacturing partner and a testing environment for Apollo. These are strong relationships, though the public evidence still points toward validation and training rather than large recurring fleets.

NEURA says demand for its wider robotics portfolio is substantial, yet its humanoid-specific commercial figures remain difficult to isolate. Sanctuary has worked with Magna and previously completed a commercial placement with Canadian Tire, but it has not disclosed the kind of repeat industrial volume that would put Phoenix alongside Digit or Figure.

Startup	Strongest commercial evidence	What the evidence currently proves
Agility Robotics	More than $300 million of conditional multi-year Digit v5 orders and four named commercial customers	Strongest visible order book and clearest conversion from trials to paid deployments
Figure	Renewed BMW work and a Catalyst Brands commercial agreement	Deep automotive validation, with commercial breadth beginning to grow
AgiBot	Large production volume and a measurable live factory operation	Serious industrial capability, but limited visibility into contract values and repeat purchases
Apptronik	Mercedes-Benz, GXO and Jabil programs	High-quality customer access, with most public evidence still centered on pilots and training
NEURA Robotics	Large reported demand across its wider robotics portfolio	Strong company-level demand, but limited clarity around 4NE1 orders
Sanctuary AI	Magna development relationship and earlier commercial work	Credible industrial engagement at a smaller disclosed scale
Unitree	Broad hardware sales	Strong robot sales, with little proof of complete factory automation contracts
Persona AI	Shipyard-focused development program	Interesting specialist demand, still before commercial deployment

Which factory humanoid startup is moving fastest?

Figure currently has the strongest all-round momentum, while Agility is accelerating fastest commercially and AgiBot is expanding fastest in raw production.

Figure’s recent progress connects manufacturing, customers and robot intelligence. The company increased Figure 03 output from one robot per day to one per hour in under 120 days, deployed the new generation at BMW and added Catalyst Brands. Figure has therefore moved several parts of the business forward at the same time rather than relying on one funding announcement or one demonstration.

Agility’s momentum looks less dramatic on video, but its business has strengthened quickly. The company disclosed its Digit v5 order book, announced a proposed transaction valuing Agility at $2.5 billion and opened a new Fremont facility for physical-AI development. The fresh capital could give Agility more than $600 million in gross proceeds if the transaction closes.

AgiBot is expanding on a different curve. The company says it took roughly a year to move from 1,000 to 5,000 robots, then only three months to go from 5,000 to 10,000. Its cumulative production has since reached 15,000 robots across a portfolio that includes humanoids, quadrupeds, cleaning robots and dexterous systems. The mix prevents a direct comparison with Figure’s single-model count, but the acceleration is hard to dismiss.

Apptronik’s latest momentum is mostly preparatory. The company raised an additional $520 million, introduced Apollo 2 and expanded its 90,000-square-foot Robot Park. Apptronik is building the foundations for large deployments, though the resulting customer output is still to come.

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

Which factory humanoid is most ready for everyday work?

Agility’s Digit remains the most mature factory humanoid for a narrow, everyday industrial job.

Digit already operates through a complete system that includes the robot, Agility’s Arc workflow software, fleet controls, customer support and maintenance. GXO has kept the robot inside a live operation for more than a year, which says more about maturity than a robot completing ten different tasks once.

Agility’s focus has helped. Digit mainly moves containers and materials between existing machines. The robot does not need human-level hands or broad reasoning for that work. Agility has instead spent years improving walking, safety, facility integration and recovery when something goes wrong.

Figure 02 also proved that a more human-like machine could survive sustained automotive work. Figure then retired the full Figure 02 fleet and replaced it with Figure 03. That rapid redesign improves the technology, but it also resets part of the reliability record. The newest Figure robot is only beginning to build its own history inside customer sites.

AgiBot’s live tablet-factory operation raises its maturity score sharply, although 64 hours remains a short window compared with months of continuous customer use. Apptronik’s mix of teleoperation and autonomous execution shows that Apollo 2 is still being trained for broader independence.

No factory humanoid startup currently publishes audited uptime, intervention frequency, repair time or mean time between failures. We can identify the leader, but we cannot yet claim that any humanoid matches the dependability of mature industrial automation.

Which factory humanoid can handle the hardest work?

Figure currently shows the strongest combination of walking, balance and dexterous manipulation in a relevant factory setting.

Figure 03’s new BMW workflow involves selecting thin sheet-metal parts, repositioning its body, placing components accurately and pulling a heavy wheeled cart. Helix 02 coordinates the robot’s hands, arms, torso and feet while Figure 03 continues moving through the task. This is more demanding than transferring identical totes between two fixed points.

As seen above, BMW has already tested two generations of Figure robots. The latest workflow suggests that Figure is progressing from structured pick-and-place work toward jobs where the robot must move its whole body around large objects.

Sanctuary remains the strongest specialist in touch and hand dexterity. Phoenix uses tactile sensing and highly articulated hands, and Sanctuary says its seventh-generation robot increased the speed of automating new tasks by 50 times. The company has since introduced an eighth generation designed for better data collection and simpler manufacturing. These remain company-reported improvements, with little customer throughput attached.

Persona AI is targeting one of the hardest eventual applications: welding inside shipyards. That environment requires rugged hardware, precise tool use and safe operation around heavy structures. Persona’s $27 million financing gives the project credibility, though the company remains much earlier than Figure, Agility or AgiBot.

Which factory humanoid offers customers the best economics?

Agility currently offers the easiest factory-humanoid business case to understand, even though no startup has published enough data for a complete cost comparison.

Agility sells Digit through a Robots-as-a-Service model. The customer pays for an operating system rather than taking full responsibility for an experimental robot, while Agility remains involved in software, support and maintenance. GXO’s decision to move from a pilot to a multi-year agreement suggests that the arrangement cleared at least an internal economic threshold.

Unitree has the clearest hardware-price advantage. Its G1 humanoid launched at around $16,000, far below the prices usually associated with industrial humanoids. The comparison stops being useful once integration begins. A low-cost body still needs task software, safety systems, maintenance, supervision and customer support before it can replace paid work.

Figure is lowering costs through design and production. Figure 03 was rebuilt for higher-volume manufacturing, and Figure says its newer battery costs 78% less than the previous version while supporting five hours of peak operation. The company has not published a sales price, robot-hour cost or customer payback period.

Apptronik has said it ultimately wants to price Apollo around the cost of a car. That remains a target rather than a quoted customer price. Apptronik’s partnership with Jabil could eventually lower component and assembly costs faster than building every factory internally.

For now, Agility has the clearest purchasing model, Unitree has the cheapest visible hardware and Figure has the strongest cost-reduction program. We still lack the operating data needed to declare a total-cost winner.

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

Who can manufacture factory humanoids at real scale?

Figure has built the strongest demonstrated Western production system, while AgiBot leads the field in reported total robot output.

Figure designed BotQ for annual capacity of up to 12,000 humanoids. Capacity claims are common in robotics, but Figure has also shown actual output: more than 350 Figure 03 robots and a production rate that reached one unit per hour. That achieved rate makes Figure’s manufacturing claim more credible than a factory blueprint alone.

AgiBot says its 15,000th embodied-AI robot has now left the production line. The count covers several robot categories, so it cannot be read as 15,000 autonomous factory humanoids. Even with that limitation, AgiBot has demonstrated a supply chain and assembly operation at a scale no Western humanoid startup currently reports.

Agility’s Oregon RoboFab was designed for as many as 10,000 Digit units annually. The company’s current order book gives that capacity a clearer commercial purpose, while the proposed public transaction could finance the launch of Digit v5. Agility has not disclosed present annual output, making its achieved production scale harder to compare with Figure’s.

Apptronik is taking an asset-lighter route through Jabil. Jabil can manufacture Apollo through an existing global network and test the robots in its own factories. This could scale quickly once demand is firm, although Apptronik has yet to disclose a production count.

Funding strengthens these manufacturing plans without settling the ranking. Figure has already translated large financing into robots and an operating line. Agility has converted less capital into deeper commercial use. Apptronik and NEURA now have enormous financial resources, but both need more visible humanoid output before we credit them with comparable execution.

Who has won the strongest factory customers?

BMW gives Figure the single strongest factory reference, while Agility has built the most diversified group of paying industrial customers.

Automotive plants are unforgiving environments. Production interruptions are expensive, tasks repeat thousands of times and new equipment must fit around existing workers and machinery. BMW’s decision to bring Figure back with a new robot generation therefore carries more weight than a one-week innovation trial.

Agility’s customer mix gives it another kind of advantage. GXO brings contract logistics, Toyota Canada brings vehicle manufacturing, Schaeffler brings industrial components and Mercado Libre brings high-volume e-commerce. Agility has shown that the same basic robot platform can interest several industries rather than one unusually supportive design partner.

Apptronik may have the best customer list relative to its current maturity. Mercedes-Benz, GXO and Jabil provide factories, logistics sites and manufacturing expertise. Google DeepMind adds a leading AI partner. The missing piece is a customer result with enough scale to compare against Figure’s production work or Agility’s daily warehouse deployment.

NEURA has attracted backing or cooperation from companies including Bosch, Schaeffler, Amazon, NVIDIA and Qualcomm. That network is impressive, but investment, technical collaboration and robot purchasing represent different levels of customer commitment.

Which startup has the best robot intelligence and the hardest advantage to copy?

Figure currently has the most convincing integrated intelligence stack, while AgiBot has built the strongest open data asset and Apptronik has the most powerful external AI partner.

Figure’s Helix system connects vision, language, planning, whole-body movement and hand control. Helix 02 uses one hierarchy to turn camera input into coordinated physical movement, with lower layers handling balance and contact at high frequency. The system has performed multi-minute tasks across rooms and is now being applied to industrial sequencing work.

Figure’s wider advantage comes from owning the full loop. The company builds the robot, runs the AI system, gathers deployment data and changes the manufacturing design. Problems discovered in the field can feed directly into the next robot generation without waiting for a separate hardware supplier or foundation-model provider.

AgiBot has taken a more open approach. AgiBot World contains over one million robot trajectories covering 217 tasks and five deployment settings. In the accompanying research, policies trained on the dataset improved average performance by 30% over Open X-Embodiment baselines and exceeded 60% success on selected complex tasks. The tests were run by the AgiBot research team, but the open dataset gives outside researchers more material to inspect.

Apptronik’s Google DeepMind partnership could close the intelligence gap. Fleets of Apollo 2 robots are collecting data for Gemini Robotics across Robot Parks and customer locations. Apptronik gains access to world-class AI research, although part of its strategic advantage depends on a partner that also works across the wider robotics market.

Agility’s defensibility sits closer to operations. Years of locomotion work, safety engineering, customer integration and real warehouse data are difficult to reproduce quickly. A rival can copy the shape of Digit more easily than the accumulated knowledge required to keep a fleet working around people and other machines.

Is China already ahead in factory humanoids?

China currently leads factory-humanoid production volume and hardware cost, while US startups still provide more transparent evidence of long-running paid deployments.

AgiBot now combines two things that Chinese startups previously struggled to show together: industrial-scale hardware production and a measurable factory operation. Its recent tablet-line run narrows the evidence gap with Figure and Agility considerably.

Unitree creates pressure from below. The company has spent years manufacturing quadrupeds at scale, sells humanoid hardware at unusually low prices and has raised only a fraction of Figure’s capital. That manufacturing discipline could make Unitree a formidable supplier once its factory software and service layer mature.

UBTECH also strengthens China’s broader position, even though it falls outside our startup ranking. Its Walker series has received industrial orders and entered mass production, giving China another route from government-supported manufacturing into automotive and electronics factories.

The remaining weakness is disclosure quality. Chinese production totals often combine research robots, service robots, quadrupeds and humanoids. Order announcements may include training centers or demonstration facilities alongside ordinary commercial buyers. Factory managers need operating hours, intervention rates, completed cycles, maintenance costs and repeat orders.

Today, China is ahead at building robot bodies cheaply and quickly. Figure and Agility remain ahead at showing how a specific robot performs inside a named customer operation over time. AgiBot is now the company most likely to erase that distinction.

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

Is Figure spreading itself too thin by chasing home robots?

Figure’s move into homes creates a focus risk, but the factory business has continued progressing rather than being abandoned.

Figure’s public messaging now gives the home a much larger role. Recent Helix demonstrations include cleaning kitchens, tidying living rooms and resetting bedrooms. Figure 03 was designed with softer materials, improved sensing and other features that make sense around consumers.

Some of that work transfers back into factories. A robot that can handle unfamiliar objects, open doors, coordinate two hands and move through clutter could eventually automate logistics jobs that fixed industrial arms cannot reach.

The conflict appears in customer priorities. Factories mainly need uptime, safety, predictable cycle times, maintenance and a clear return on investment. Homes require broad intelligence, natural interaction, quiet movement and the ability to deal with thousands of unpredictable objects. Solving both markets could stretch engineering and management attention.

Figure’s latest industrial agreements keep the concern under control for now. The risk would become more serious if new factory deployments slowed while most new robots remained inside Figure’s own home-training fleet.

How much of the factory humanoid evidence can we actually trust?

The most credible factory-humanoid evidence comes from quantified customer operations and signed commercial orders, while demonstrations and theoretical factory capacity deserve much less weight.

Figure’s historical BMW output and Agility’s GXO throughput are useful because each metric tracks a repeated task inside a named customer site. Agility’s new order disclosure is also valuable, although the company clearly states that the orders depend on certain contractual milestones.

AgiBot’s livestream improves the credibility of its factory figures. Viewers could observe a continuous production environment instead of a short edited sequence. The result still lacks independent auditing, and a six-day program cannot prove years of reliability.

Apptronik’s Robot Park gives us evidence of active fleets, teleoperation and data collection. It says much less about autonomous customer productivity. NEURA’s funding and broader order backlog show demand for the company, but they reveal little about how many 4NE1 humanoids customers have committed to buy.

Private startups control most of the information in this market. They choose when the timer begins, what counts as a successful task and which failures appear publicly. We therefore give the most weight to repeated customer use, explicit contract language, achieved production and metrics that competitors could eventually reproduce.

The weakest evidence includes letters of intent, reservations, edited demonstration videos, capacity targets and combined production totals covering several robot types. These figures can still reveal direction, but they should never carry the same weight as paid work.

Which factory humanoid startups are actually ahead?

Figure is ahead overall today, with Agility close enough that one successful product cycle could reverse the ranking.

Figure wins through breadth. The company combines the most capable factory manipulation we have seen, an in-house intelligence system, achieved manufacturing output and a deep automotive reference. Figure has weaknesses, especially limited visibility into contract value and the distraction created by its home strategy, but no rival currently matches the whole package.

Agility wins the commercial argument. More than $300 million of conditional Digit v5 orders gives it the strongest visible demand in the category. Digit also has a longer record of routine paid work than newer humanoids. Agility would move into first place if Digit v5 turns those orders into large operating fleets and expands beyond material handling.

AgiBot now ranks third. Its enormous production lead can no longer be dismissed as hardware without real work, because the company has added a substantial live factory result. AgiBot still needs better disclosure around paying customers, fleet composition and long-term reliability.

Apptronik slips to fourth despite having better-known Western partners than AgiBot. Apollo 2 remains one of the most credible challengers, but Apptronik is currently investing in the data and infrastructure required for future autonomy. The ranking can change once that work produces measurable customer output.

NEURA has the capital and ecosystem to move quickly, although its humanoid remains much less commercially visible than its wider robotics platform. Sanctuary leads in selected dexterity technologies but lacks scale. Unitree is the hardware-cost threat. Persona is an early specialist with a credible industrial target and very little deployment evidence.

Rank	Startup	Why it holds this position
1	Figure	Strongest overall mix of factory performance, advanced manipulation, integrated AI and achieved manufacturing output
2	Agility Robotics	Best commercial order book, longest record of routine paid work and broadest group of named industrial customers
3	AgiBot	Largest production scale and a newly credible live factory result, with weaker contract and fleet transparency
4	Apptronik	Excellent partners, almost $1 billion in funding and a strong learning platform, but limited quantified customer output
5	NEURA Robotics	Exceptional funding and industrial ecosystem, with little public proof of repeated 4NE1 factory work
6	Sanctuary AI	Strong hands, tactile sensing and industrial relationships at a comparatively small commercial scale
7	Unitree	Major cost and manufacturing threat, but still lacking a complete factory deployment and support proposition
8	Persona AI	Promising focus on difficult shipyard work, currently at an early development stage

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.