39 billion US dollars, the threefold bubbles of Figure, the global top leading player in the embodied intelligence track

This global robotics star headquartered in California with a 300,000-square-foot campus had its valuation bubble exposed in October by a Forbes article titled Figure’s Brett Adcock Sold Investors A Humanoid Future. Now He Has To Build It.

The issues can be roughly summarized into three layers, and the first layer is the marketing bubble:

In 2025, Figure once carried out a pilot cooperation with BMW. Its founder Adcock publicly claimed that they had "a fleet of robots performing end-to-end operations". But as a pilot project, the actual implementation situation on BMW's side was that only one Figure robot was working, and there was another one on site for rotation charging.

In June this year, Figure released another promotional video showing robots working in the BMW factory, in which 4 robots appeared. But according to BMW's disclosure, only 1 robot actually completed the work, and the other 3 did not participate at all. To ensure smooth shooting, Figure also dispatched 5 to 10 employees to cooperate on site.

As for the actual workload of Figure robots, BMW confirmed that they had assisted in the production of tens of thousands of cars and moved about 90,000 components.

The second layer is the financing bubble:

Last year, Figure sought to raise USD 15 billion in financing with a pre-money valuation of USD 38 billion. The term sheet reserved a USD 1.2 billion share for two relatively unknown venture capital firms: Parkway Venture Capital and Align Ventures. But neither of the two firms had sufficient assets to take on such a large quota.

The two firms later made up their respective shares through investment pools set up exclusively for this single company. The minimum single investment amount accepted by Align was only USD 100,000.

After the financing was completed, Figure's valuation reached USD 39 billion, an increase of about 15 times over the previous year. It ranked among the top 20 most valuable private companies in the world, almost on a par with its well-known partner BMW.

A core basis driving the rise in valuation was the robot fleets that had not yet been deployed by two major clients: one was BMW, and the other was UPS.

The third layer is the technology bubble:

Recently, Figure has shifted the focus of its narrative to household scenarios.

They showed a video of robots being trained in an Airbnb residence, and stated that by next year, Figure robots would possibly enter ordinary people's homes to help with housework.

But Rodney Brooks, co-creator of the Roomba vacuum cleaner, believes that humanoid robots are still far from being able to work safely in home environments: "I would certainly not let a humanoid robot stay around my grandchildren."

A wrongful termination lawsuit filed by Rob Gruendel, former head of product safety at Figure, in November 2025 described the internal disputes at the company surrounding these risks. The lawsuit states that early last year, Figure founder Adcock asked him what needed to be achieved to get robots into households.

Gruendel then developed a roadmap focusing on identifying nearby objects to avoid colliding with people or other living beings. However, Adcock did not respond to the messages about the roadmap. When he tried to push for risk reduction measures internally, another executive at Figure told him: "If we do that, Adcock will fire all of us."

Even so, it came as no surprise that Figure started raising funds again.

A member of Figure's board of directors revealed that the company is completing "billions of dollars" in financing in the form of an extension of its Series C round; the company refused to confirm whether it is currently conducting financing. Hill also said that Figure's Series D round will have a much higher valuation.

Its founder Adcock holds the same view. As always, the planned scale is expanding along with the capital.

"Some crazy things are happening at the company right now," he said. "I made a prediction at the beginning of this year: my challenging goal is to bring a robot into a household it has never been to before, to complete long-term continuous work, such as washing a whole basket of clothes from start to finish, or making a plate of scrambled eggs. I am very confident about that."

There is a huge gap between one robot that has completed a task once and a robot fleet that can operate stably.

False lies are easy to identify, but when Figure mixes real facts with false narratives, it forms an embodied intelligence business that can be scaled, replicated and monetized.

The founder who excels in inflating valuations

Brett Adcock, the founder of Figure, has spent the past decade starting businesses that almost all profit from valuation bubbles.

In 2018, he sold the recruitment website Vettery for USD 110 million, making his first USD 30 million.

In 2021, he pushed the air taxi company Archer Aviation to go public via SPAC. At that time, Archer had not yet completed a single commercial flight, but it had already raised USD 858 million at a valuation of USD 1.7 billion.

Figure's narrative is even grander.

When the company was founded, humanoid robots could only carry out small-scale tests and controlled demonstrations in factories. 19 months later, Figure's valuation rose from USD 2.6 billion to USD 39 billion.

Last year, Figure sought to raise USD 15 billion in financing with a pre-money valuation of USD 38 billion. The financing documents even reserved a USD 1.2 billion quota for two small-scale venture capital firms. The two firms only completed their investment later through special purpose vehicles.

Figure's final valuation reached USD 39 billion, close to that of established automakers like BMW.

What investors bet on is not how much revenue Figure has generated today, but how many robots it may deploy in the future.

Adcock publicly stated that Figure has the opportunity to deploy 100,000 robots at BMW and another unnamed company in the next four years. A memo sent to investors of the special purpose vehicle stated that at the pricing assumptions at that time, deploying 100,000 robots in four years would mean USD 9 billion in annual recurring revenue.

The robots have not been deployed yet, but the revenue has been calculated in advance.

This algorithm of estimating the future based on vague business projections is not difficult to understand. The humanoid robot market is large enough, and its boundaries are vague enough. Morgan Stanley predicts that the market size may reach USD 5 trillion by 2050; Citi's forecast is USD 7 trillion; RBC gives a range of USD 9 trillion to 12 trillion.

As long as the end point is far enough, a single demonstration along the way can be interpreted as "commercialization is coming soon".

Adcock is also very good at using this kind of narrative. He says that humanoid robots will eventually enter every household, with a monthly rent of 500 or 600 US dollars, helping people do laundry, tidy up the kitchen, and pick up toys scattered by children.

All these things may happen someday.

But today's robots have not yet learned to use scissors stably, and there is no set of human behavior data on the Internet that can be directly used to train humanoid robots.

Figure's valuation has already entered a much larger world in advance; its products have not.

The ambiguous cooperation bubble

The BMW pilot project ended at the end of last year, and Figure robots assisted in the production of tens of thousands of cars and moved about 90,000 components.

But Figure founder Adcock once claimed on social media that Figure provided BMW with "a fleet of robots performing end-to-end operations". This statement was denied by BMW, and in fact only one Figure robot was working.

In June this year, Figure returned to BMW's factory in Spartanburg, South Carolina. Adcock promoted it on social media: in the video with strong electronic music, the updated Figure robot dragged heavy vans in the BMW factory.

Talking about this project, Adcock told Forbes that there were "six to eight" Figure robots on site, and he was "quite sure" that BMW was paying for these robots. "I think BMW is paying for all the robots now," he said. "All our customers have a certain number of robot quotas and contracts, and we are working hard to deliver more robots to them."

BMW stated that this project was a non-production test involving two to three robots. The test has ended, and there are no Figure robots running in BMW factories at present. Among the four robots in the promotional video, BMW said that only one actually completed the work, and the other three did not participate.

Figure also dispatched 5 to 10 employees to ensure the smooth progress of the video shooting. As for the relationship between the two companies, BMW only stated that the two sides "are still continuously evaluating humanoid robots and exchanging relevant experience".

UPS, another speculated major client, also does not seem to be actually using Figure's humanoid robots. A spokesperson for UPS truck drivers said that UPS has not notified the union of any Figure robot tests. The union stated that as long as the test may affect employees, UPS must give advance notice.

Figure is a real existing company, but it is wrapped in a set of figures far exceeding its actual scale. It has produced about 1,000 humanoid robots, but it is not clear how many of them have been deployed and where they are deployed. The question of who is paying for these robots also remains unanswered.

The previous cooperation between Figure and AI giant OpenAI also had similar conflicting statements.

In 2024, OpenAI participated in Figure's Series B financing, and the two companies then jointly developed the model driving Figure robots. But less than a year later, the cooperation fell apart.

OpenAI stated that Figure missed many agreed delivery deadlines and could not provide the hardware and software support required by researchers. OpenAI finally concluded that Figure was not a viable robotics partner, and told Adcock that it planned to increase investment in self-developed robots.

Figure does not accept this statement. Figure said that its team was advancing faster in robot model training and was unwilling to hand over this work to OpenAI, so it terminated the cooperation. Both sides interpreted the breakup as a result of their own active choice.

Figure, which began to shift its deployment focus to household scenarios this year, has amplified technology safety issues.

In a lawsuit filed in November 2025, Rob Gruendel, former head of product safety at Figure, stated that the company had internally discussed how to bring robots into households. The roadmap he developed included identifying nearby objects to prevent robots from colliding with people or other living beings.

The lawsuit states that Adcock did not attend relevant meetings and did not respond to the roadmap. Gruendel also stated that when he pushed for risk reduction measures, another executive said to him: "If we do that, Adcock will fire all of us."

He also described a test: a Figure 02 was set to hit with maximum force, and the impact force was more than twice the force required to break an adult's skull. In another malfunction, the robot punched the steel refrigerator door, leaving a crack about a quarter of an inch long, almost hitting an employee.

Figure denied some of these claims, stating that Gruendel was fired for poor performance, and the company will comply with the highest safety standards. The lawsuit is still ongoing.

After robots leave the laboratory and enter actual deployment scenarios, hidden safety hazards will quickly be amplified into issues more important than technological advancement.

What part of Figure are Chinese companies chasing

When Chinese companies benchmark against Figure, the first thing they value is its technical performance.

At this year's Spring Festival Gala, more than 20 Unitree robots completed tumbling, ejection and continuous movements at the same time. The difficulty of this demonstration does not lie in whether a single machine can do it, but whether so many mass-produced robots can move in unison.

Unitree's currently outstanding capabilities lie in the ontology, motion control and hardware consistency.

The prospectus shows that in the first three quarters of 2025, Unitree's humanoid robot revenue accounted for more than 50% of its total revenue, with overall revenue of about USD 200 million, and the gross profit margin of related products reached 63%.

But Unitree's main customers are still in the scientific research and education market. The scientific research market can accept higher prices, and does not require robots to run continuously for a long time like factory equipment. A 63% gross profit margin shows that Unitree has found a good product market fit, but it does not directly prove that humanoid robots have entered large-scale industrial applications.

The tennis robot from Galaxy General is also very representative.

Galaxy General rented tennis courts during the Spring Festival, used motion capture equipment to collect data, and then carried out repeated reinforcement learning training. Its value does not necessarily lie in proposing a new academic path, but in quickly integrating hardware, data and algorithms together.

The robot needs to judge the direction of the incoming high-speed ball, adjust its posture, complete the swing, and then restore balance after the movement. This process is closer to real system engineering than releasing a video of a hand grasping an object.

Agibot is promoting full-body motion control. In the past, the upper limb operation and lower limb movement of robots were usually handled by two separate systems. Now more and more companies are trying to use one model to handle movement, balance and operation at the same time.

This route is similar to Figure's Helix.

Figure has publicly introduced a three-layer control architecture of low frequency, intermediate frequency and high frequency, trying to use a unified full-body model to control the robot. What Chinese companies are catching up with is exactly this capability that evolves from "hands can move, legs can walk" to "the whole body moves in coordination".

But the easiest thing to replicate is the movement, not the system.

For the same tennis movement, it is not difficult to make a one-minute video; it is completely different in difficulty to let the robot work every day on different courts, under different light and different incoming ball speeds.

For the same task of moving parts in the BMW factory, it is not difficult to let a single machine complete one movement; only when it can run continuously for several months and be repaired quickly after failure, will customers treat it as production equipment.

The embodied intelligence field is generally short of data now. Humans can obtain massive amounts of text and video from the Internet, but the data required by robots is more about "how to hold a cup steadily", "how to fold a towel well", "how to move your hand to avoid the rebounding refrigerator door".

Collecting such data is very expensive and very slow.

So half of Figure's story is real: its robots are indeed making progress, and its unified control model does have technical content. The other half is automatically filled in by the capital market. The market complements a single movement into a capability, complements a capability into industry space, and then converts the industry space into the company's valuation.

The domestic industry is now going through this process.

Statistics show that there are more than 20 humanoid and embodied intelligence companies in China with a valuation of more than 10 billion RMB or USD 1.5 billion. At the same time, the top company Unitree has a revenue of about USD 200 million, and the real commercialization of embodied intelligence still takes time.

Putting these two figures together, you can see how fast the industry is developing.

The capital has arrived, but the actual demand is still on the way.

Figure has produced about 1,000 robots, but the outside world does not know how many of them are actually deployed, how many are working, and who is paying for them continuously.

Unitree is moving faster in hardware, Galaxy General is moving faster in complex movements, Agibot is accelerating in full-body control, and other companies are seeking breakthroughs in dexterous hands, data collection and model training.

This is a real technology competition.

But if everyone finally uses the smoothest demonstration video to prove the largest commercial value, the competition will gradually turn into another thing: who is better at telling stories.

This article is from WeChat official account "Digital Intelligence Surge", author: Zhou Yi, published with authorization from 36Kr.