How's Figure AI doing these days?

Is Figure AI doing well right now?

Figure AI is making impressive progress in humanoid robotics, but its business is still much smaller than its $39 billion valuation suggests.

Founded by Brett Adcock in 2022, Figure has developed three generations of humanoid robots, built its own AI system called Helix, and started manufacturing robots at its BotQ facility. It has also demonstrated real industrial work at BMW and recently tested household tasks in 30 unfamiliar homes.

The problem is that these achievements haven't yet translated into a large, independently verified customer base. An October 2026 Forbes investigation reported that Figure had manufactured approximately 1,000 robots, while the number actually working for paying customers remained unclear. Figure also declined to disclose its revenue.

Figure's machines and software are advancing fast. Customer adoption is harder to see.

Why did Figure AI's valuation jump to $39 billion?

Figure AI's valuation rose fifteenfold in just 19 months, largely because investors are betting that humanoid robots could become a huge market.

In February 2024, Figure raised $675 million at a $2.6 billion valuation, backed by investors including Microsoft, Nvidia and Jeff Bezos. By September 2025, the company had announced more than $1 billion in Series C commitments at a $39 billion valuation, with Parkway Venture Capital leading the financing.

That's an increase of $36.4 billion in less than two years.

During that period, Figure made real progress. It deployed robots at BMW, introduced its own AI models, developed Figure 03 and worked toward higher-volume manufacturing. But with no published revenue figures, there's no evidence that its sales grew anywhere near as quickly as its valuation.

The financing itself is worth a closer look. Forbes reported that Parkway and Align Ventures assembled substantial portions of the investment through special-purpose vehicles. This is a legitimate financing method, although the resulting valuation reflects the price agreed in a private transaction rather than a broadly traded market price.

Figure's $39 billion valuation is ultimately a bet on future dominance. Investors are paying for the possibility that the company will eventually sell or rent tens of thousands of useful humanoids, long before that business exists at scale.

Milestone	Valuation	What happened
Company founded, 2022	Undisclosed	Brett Adcock launches Figure
Early financing, 2023	Undisclosed	Figure raises institutional capital
Series B, February 2024	$2.6B	$675M raised
BMW factory pilot, 2024	No new verified valuation	Initial industrial testing
Figure 02 factory operations, 2025	No new verified valuation	Extended production work
Series C, September 2025	$39B	More than $1B in commitments
Figure 03 manufacturing ramp, 2026	$39B last verified round	Hundreds of new-generation robots manufactured
Latest fundraising discussions	Undisclosed	Additional investment discussions, without a confirmed new priced round

If you want more recent data on this point, see the latest figures in our Humanoid Robots report ↓.

Did Figure AI's BMW robots actually work?

Figure AI's robots really did perform useful factory work at BMW, although the deployment was far smaller than the production figures might suggest.

BMW confirmed that Figure 02 robots operated at its Spartanburg, South Carolina, factory during a ten-month program in 2025. In a February 2026 update, the automaker reported more than 90,000 sheet-metal components handled, approximately 1,250 operating hours and contributions to the production of more than 30,000 BMW X3 vehicles.

We can put those numbers into perspective. Dividing 90,000 components by 1,250 operating hours gives an average of 72 components per hour, or one every 50 seconds. That's a meaningful amount of physical work, even though the average includes operating periods that may not represent uninterrupted production.

The surprising detail is how few robots were involved. BMW told Forbes that only one Figure robot worked at a time during the original deployment, with another available for rotation and charging.

Figure returned to Spartanburg in June 2026 with Figure 03, demonstrating more complex logistics tasks such as manipulating carts and organizing parts. However, BMW later described the program as a non-production test involving two or three robots.

By the time Forbes investigated the relationship in October, BMW said the second test had ended and no Figure robots were operating at the factory. The two companies were still collaborating on evaluations.

BMW has also started testing humanoids from other suppliers, including Hexagon's AEON at its Leipzig plant in Germany.

Figure has demonstrated that its robots can handle real automotive tasks, but BMW hasn't turned those experiments into a large, permanent Figure robot fleet.

How many Figure AI robots are actually working today?

Figure AI has manufactured roughly 1,000 humanoids, but we still don't know how many are doing paid work for customers.

That uncertainty is especially striking because Figure has become much more transparent about manufacturing. In April 2026, the company reported producing more than 350 Figure 03 robots. Its subsequent production ramp pushed total manufacturing into the neighborhood of 1,000 units, according to Forbes.

But manufactured robots serve several purposes. Figure uses them for internal research, AI training, hardware testing, household experiments and commercial operations. We cannot treat the total as an installed customer fleet.

The publicly documented customer numbers are much smaller. BMW's original deployment used one working robot at a time, while its follow-up test involved a handful of machines. Catalyst Brands has an agreement with Figure, but the number of robots actively operating under that arrangement has not been disclosed.

Forbes also examined reports of a relationship with UPS. The Teamsters union said it had received no required notification of robot testing affecting its members, while UPS declined to comment.

So the manufacturing count is real enough; the paid fleet is still a mystery.

Can Figure AI really manufacture thousands of robots?

Figure AI has made a major leap in manufacturing, but its factory hasn't yet proved it can sustain production at thousands of robots per year.

The company's BotQ facility increased its demonstrated production rate from one Figure 03 robot per day to one per hour in fewer than 120 days, according to its April 2026 manufacturing update. That's a 24-fold improvement.

Figure also reported more than 9,000 actuators and 500 battery packs produced, with over 150 networked workstations supporting its manufacturing processes.

Some of its quality-control figures are particularly revealing. Figure reported a 99.3% first-pass yield for battery packs and more than 80% for completed robots at final inspection.

The battery result is strong. The complete-robot figure suggests that roughly one in five units initially needed some form of correction before passing inspection, which isn't especially surprising during a manufacturing ramp but leaves room for improvement.

We should also be careful with the one-robot-per-hour claim. At 24 hours a day, that cycle time would theoretically produce 8,760 robots annually. Figure hasn't demonstrated continuous operation at that pace.

The machinery and quality controls are there. Now Figure has to keep those lines moving without letting costs or defects creep up.

Is Figure AI's Helix getting smart enough to handle real homes?

Figure AI's Helix 2.5 is a genuine breakthrough in robot learning, but its 56% task-success rate shows why a dependable household robot is still some distance away.

In September 2026, Figure tested three household tasks across 30 Bay Area homes that its robots had never encountered during training. The tasks involved tidying living rooms, folding towels and making beds.

The results were unusually useful because Figure compared two versions of the system under the same training and evaluation conditions. Without pretraining on its Index human-video dataset, Helix completed 9% of tasks. With Index pretraining, success reached 56%.

That's a 47-percentage-point improvement, or approximately 6.2 times the original success rate.

Figure also reported using half as much task-specific training data as a representative earlier Helix 02 behavior, while transferring those skills to 30 unfamiliar homes. The evaluation required full task completion, without credit for partially successful attempts.

There are limits to what this proves. The robot had already been trained to perform the three selected behaviors in other environments. It wasn't being asked to respond to any household instruction a person might invent.

The 56% result is also company-reported rather than independently reproduced. Figure hasn't published enough information on per-task reliability, operating speed and human supervision to assess how the same system would perform as a consumer service.

Getting a robot to complete a chore around half the time is a real research achievement. For someone paying to have their home cleaned, however, that reliability would still be frustrating.

Could Figure AI's Index dataset become its biggest advantage?

Figure AI's Index project may be more important than the robots themselves because it gives Helix access to millions of examples of how humans interact with the physical world.

Figure introduced Index publicly in August 2026 after running the program privately for several months. At launch, it reported more than 264,000 app downloads, over 44,000 weekly active users and 16 million uploaded videos.

The company had already paid approximately $15 million to contributors.

Index was processing around 30 minutes of video every second at launch, equivalent to 43,200 hours of human activity per day. The rate subsequently increased to approximately 35 minutes per second.

Index sidesteps one of robotics' biggest headaches: collecting training data exclusively through expensive physical robots. Thousands of contributors can record activities in kitchens, workshops, offices and other environments without Figure having to send humanoids into each location.

The videos still need to be converted into useful robot training data. Human movements and robot movements aren't interchangeable, especially when force, balance and object contact are involved.

Figure addresses this through filtering, deduplication, annotation and model training. The company says it also rebalances the dataset to avoid overrepresenting similar activities.

As seen above, Index pretraining dramatically improved Helix's ability to complete tasks in unfamiliar homes. Whether collecting even more data continues to produce similar gains will determine how valuable this advantage becomes.

Is Figure AI spending too much money on computing power?

Figure AI is committing extraordinary amounts of money to AI computing infrastructure before proving that its robot business can generate revenue on a comparable scale.

In September 2026, Figure announced a partnership with Nscale covering access to as many as 100,000 Nvidia Vera Rubin GPUs. The initial deployment is targeted for the second half of 2027.

The agreement includes an initial $3.5 billion computing commitment, with plans to potentially expand beyond $6 billion.

For perspective, the initial commitment equals almost 9% of Figure's last confirmed $39 billion valuation. It also comes on top of the costs of designing humanoid hardware, operating BotQ, collecting training data and supporting customer deployments.

These figures describe infrastructure commitments rather than cash already spent. The GPUs have not yet been deployed under the announced schedule.

There's a technical case for that bet. Figure is collecting tens of thousands of hours of human activity every day, and its latest research suggests that larger pretraining datasets can improve Helix's ability to handle unfamiliar situations.

More computing capacity would let it train on a wider range of physical experiences.

But we're still missing the financial link between a smarter model and a profitable robot. A robot that becomes 20% more capable does not automatically generate 20% more revenue, particularly if customers still need to supervise it.

Figure's investment could eventually give it an advantage that smaller robotics companies cannot afford to reproduce. It could also become an enormous financial burden if paid deployments grow slowly.

If you want more recent data on this point, see the latest figures in our Humanoid Robots report ↓.

Is Figure AI actually making money, and who pays for its robots?

Figure AI has commercial agreements with recognizable companies, but its revenue, profitability and paying customer base remain largely undisclosed.

BMW has confirmed real industrial work, although it hasn't disclosed how much, if anything, it paid Figure for its deployments. Catalyst Brands announced a commercial agreement for humanoids at its Reno distribution facility in May 2026, but the companies haven't published contract revenue or an active fleet count.

Brookfield has a broader relationship covering investment, property access and opportunities for collecting real-world data. These connections give Figure access to different environments, from automotive manufacturing and warehouse operations to residential properties.

A paid production contract, a robot evaluation, a data-collection partnership and an investor relationship have very different financial implications.

Brookfield is also an investor in Figure and is connected to Catalyst Brands. That makes the relationship useful for opening doors, but it doesn't necessarily demonstrate demand from completely independent customers choosing Figure over competing suppliers.

One potentially useful financial reference appeared in investor materials discussed by Forbes. An Align Ventures memo described a scenario in which deploying 100,000 robots could generate $9 billion in annual recurring revenue.

That works out to $90,000 per robot per year, or $7,500 per month.

Those figures are hypothetical. They tell us what investors have considered possible under certain pricing and deployment assumptions, rather than what Figure currently charges or earns.

Revenue wouldn't settle the profitability question anyway. Humanoid robots require expensive components, ongoing maintenance, software development, insurance and customer support. Figure's manufacturing and computing commitments add further expenses.

Without published revenue, operating costs, gross margins or repeat orders from independent customers, we cannot establish whether Figure's existing contracts are profitable or whether demand is growing beyond initial trials.

Why did Figure AI stop working with OpenAI?

Figure AI and OpenAI ended their robotics partnership after less than a year, and the companies disagree sharply about why it happened.

Their collaboration began alongside Figure's $675 million funding round in February 2024. The plan was to combine Figure's humanoid hardware with OpenAI's AI models, creating robots capable of understanding instructions and performing physical tasks.

Figure later decided to develop its own AI internally.

Brett Adcock said Figure's team was making faster progress independently and no longer needed OpenAI's technology. That became part of the company's broader strategy of controlling both its hardware and intelligence systems.

OpenAI gave Forbes a conflicting account in October 2026. It said Figure had repeatedly failed to meet agreed deliverables and provide the hardware and software support required for the collaboration.

Figure disputed those allegations.

Because the underlying development records aren't public, we can't determine which company bears responsibility for the split. What we can evaluate is what happened afterward.

Figure continued developing Helix and eventually demonstrated whole-body autonomy and more generalizable household behaviors. Its September 2026 research provides concrete evidence that the internal AI team has made progress.

The split therefore hasn't stopped Figure from advancing its technology. Whether building its own foundation models is financially wiser than relying on outside AI providers remains much harder to answer.

Is Figure AI ahead of Tesla Optimus and other humanoid robots?

Figure AI is one of the strongest humanoid robotics contenders, but nobody has demonstrated enough large-scale commercial success to claim an undisputed lead over Tesla, Agility Robotics, Apptronik and other serious competitors.

Figure has several credible strengths. Its BMW deployment produced independently confirmed factory results, its manufacturing operations have expanded quickly, and Helix has demonstrated increasingly flexible robot behavior.

Tesla is following a different strategy with Optimus. It has enormous experience manufacturing complex products, including vehicles, batteries and motors, and can test robots inside its own factories. Its challenge is proving that Optimus can do useful work at scale beyond controlled demonstrations and internal deployments.

Agility Robotics has concentrated on warehouse and logistics work with Digit. Its narrower range of applications could make it easier to prove customer value, although the company still needs to expand deployment volume and demonstrate attractive economics.

Apptronik has developed Apollo and secured industrial relationships including Mercedes-Benz and Jabil. Boston Dynamics brings decades of robotics engineering experience, while Unitree and UBTECH are pushing humanoid development from China.

The difficulty is that manufacturers disclose different metrics. Some emphasize robots manufactured, others report walking demonstrations, operating hours, customer partnerships or task success.

We can compare their strategies and verified achievements. There's no credible robot-hours leaderboard yet, much less one based on paid, productive work.

Company	Main focus	Current advantage	What remains unclear
Figure AI	General-purpose humanoids	Helix AI, BMW experience, growing manufacturing capacity	Paid fleet size and profitability
Tesla Optimus	Vertically integrated humanoids	Manufacturing resources and internal factory access	External commercial deployment
Agility Robotics	Logistics and warehouses	Focused industrial applications	Large-scale customer economics
Apptronik	Industrial humanoids	Mercedes-Benz and Jabil relationships	Sustained deployment volume
Boston Dynamics	Advanced mobile robotics	Whole-body control and robotics experience	Atlas commercial fleet scale
Unitree Robotics	Affordable humanoid platforms	Hardware pricing and accessibility	Autonomous task reliability
UBTECH Robotics	Industrial humanoids	Automotive manufacturing applications	Independently verified operating economics
NEURA Robotics	Cognitive and industrial robots	Industrial automation integration	General-purpose autonomy at scale

Could cheaper Chinese robots destroy Figure AI's advantage?

Chinese humanoid manufacturers could put serious pressure on Figure AI's hardware pricing, especially if increasingly capable robots become available for a fraction of the cost.

Unitree is the clearest example. Its G1 humanoid was introduced at a starting price around $16,000, far below the amounts commonly associated with advanced industrial humanoid systems.

Of course, a $16,000 robot doesn't necessarily deliver the same autonomy, reliability or productive output as a more expensive machine. Hardware configurations and software capabilities vary enormously.

But cheaper machines could still reshape the market.

Companies such as Unitree and UBTECH are developing manufacturing capabilities while benefiting from China's extensive electronics, motors and component supply chains. If several manufacturers eventually produce capable humanoid bodies at competitive prices, customers may become reluctant to pay large premiums for hardware alone.

Figure's best defense is its software.

Helix, its human-video training system and its experience with real-world tasks could become more valuable as mechanical components become easier to source. A robot that works reliably without supervision can justify a higher price even when cheaper hardware alternatives exist.

The risk is that competitors improve their AI systems while maintaining a manufacturing-cost advantage.

If you want more recent data on this point, see the latest figures in our Humanoid Robots report ↓.

Are Figure AI's robots safe enough to work around people?

Figure AI hasn't yet shown enough independent safety evidence to establish that its humanoids are ready for widespread, unsupervised use around ordinary people.

The issue attracted attention after former Figure safety engineer Robert Gruendel filed a lawsuit in November 2025. He alleged that he had been dismissed after raising concerns about robot safety, including dangerous forces and an incident involving damage to a refrigerator door.

Figure denied wrongdoing, disputed the allegations and said Gruendel had been terminated for poor performance. The company also brought counterclaims involving company documents and related matters.

A federal court addressed a discovery dispute in July 2026, but that ruling did not establish the truth of the underlying safety allegations.

Beyond the lawsuit, Figure faces the same practical safety challenges as other developers of powerful humanoid machines.

BMW reported that its original factory deployment required changes to safety arrangements, including barriers and partitions. Those precautions are normal in industrial automation, where machines operate in controlled environments and workers follow defined procedures.

Homes are much less predictable. A household robot might encounter children running nearby, pets moving underfoot, fragile objects and unexpected physical contact.

Figure's Helix research shows progress in interpreting unfamiliar surroundings, but task-completion rates are only one part of the problem. Customers will also need convincing evidence that robots can recognize dangerous situations, stop reliably and recover safely when something goes wrong.

Could Figure AI robots really replace human workers?

Figure AI's humanoids could eventually replace substantial amounts of repetitive physical labor, but current deployments support a much narrower conclusion.

Consider BMW's manufacturing experiment. Figure 02 accumulated about 1,250 operating hours over ten months while handling metal components.

That equals approximately 31 standard 40-hour workweeks. But a robot's operating hours cannot automatically be counted as equivalent human labor, especially when its task represents only one part of a production process.

The economic calculation is also more complicated than comparing a robot's purchase price with a worker's salary.

A company needs to consider maintenance, charging, software, supervision, downtime and integration. Robots become attractive when they can deliver enough productive hours at a lower total cost than the human labor or conventional automation they replace.

Using the hypothetical $90,000 annual robot subscription discussed earlier, a humanoid generating 4,000 productive hours per year would cost $22.50 per hour in subscription fees alone.

At 2,000 productive hours, that figure doubles to $45 per hour. At 1,000 hours, it becomes $90.

Those are illustrative scenarios rather than Figure's actual pricing or verified utilization rates.

They show why uptime matters so much. A costly robot working effectively across multiple shifts could make economic sense, while the same machine operating only a few hours per day might struggle to compete.

Is Figure AI really worth $39 billion today?

Figure AI's $39 billion valuation looks extremely aggressive when measured against its disclosed commercial achievements.

We can test that valuation using hypothetical revenue multiples.

At 10 times annual revenue, Figure would need $3.9 billion in yearly sales to support a $39 billion valuation. Even at a generous 20 times revenue, it would still need approximately $1.95 billion.

Using the $90,000 annual revenue per robot assumed in the investor scenario reported by Forbes, those two outcomes require roughly 43,300 and 21,700 paying robots respectively.

Figure has reportedly manufactured around 1,000 humanoids in total, including machines used for research and internal development.

So even the more optimistic 20-times-revenue scenario would require a paying fleet more than twenty times larger than the company's reported total manufacturing output.

One more snag: a robotics company may deserve a lower revenue multiple than a highly profitable software business because manufacturing, maintenance, hardware depreciation and customer support can consume substantial capital.

The following table shows how demanding the valuation becomes under different assumptions.

Revenue multiple	Revenue needed	Paying robots at $90K/year
5×	$7.80B	86,667
8×	$4.88B	54,167
10×	$3.90B	43,333
12×	$3.25B	36,111
15×	$2.60B	28,889
20×	$1.95B	21,667
25×	$1.56B	17,333
30×	$1.30B	14,444

If you want more recent data on this point, see the latest figures in our Humanoid Robots report ↓.

What's the biggest risk for Figure AI now?

Figure AI's biggest risk is that its robots keep getting better without becoming reliable and affordable enough for customers to deploy in large numbers.

We've already seen the gap in several areas. BMW's experiments showed genuine industrial value, but follow-on work remained limited. Helix 2.5 improved household generalization considerably, yet still completed only 56% of the evaluated tasks. BotQ demonstrated much faster manufacturing, although sustained production at full theoretical capacity hasn't been established.

Figure is also trying to do an extraordinary number of difficult things at once.

It designs its own robots, manufactures key components, builds foundation AI models, collects human training data, develops industrial and household applications, and plans billions of dollars in computing commitments.

There are advantages to controlling all these pieces. Figure can optimize its hardware and software together instead of relying on suppliers whose products weren't designed for its robots.

The downside is how much capital and engineering effort this approach requires.

Its next major achievement should therefore be commercial rather than promotional: a customer deploying a meaningful number of robots, paying for their work, measuring their productivity and ordering substantially more.

So, how's Figure AI really doing these days?

Figure AI is doing extremely well in robotics development, but it's still a long way from proving that its $39 billion valuation makes financial sense.

The technological progress is hard to dismiss. Figure has manufactured roughly 1,000 humanoids, operated robots inside a real BMW production environment, improved manufacturing speed dramatically and developed an AI system capable of transferring learned tasks into unfamiliar homes.

Its Index project is also worth watching closely. Collecting human activity at global scale and using it to improve robot intelligence could become a serious competitive advantage if the gains continue.

However, the commercial picture is much weaker. Figure has not disclosed revenue, profit margins or a clear paid-deployment count. Its best-known industrial relationship has produced successful but limited trials, and no independently confirmed large customer fleet has emerged.

Figure remains one of the most promising companies in humanoid robotics, with a valuation that already assumes tremendous future success.

The next stage will be decided by customer orders, robot uptime and operating costs. If Figure starts deploying thousands of reliable humanoids for paying customers, its current valuation will become much easier to defend. Until then, its engineering achievements are real, while the scale of the business investors are paying for remains largely a projection.

If you want more recent data on this point, see the latest figures in our Humanoid Robots report ↓.