Figure AI

IWhat it does

Figure AI is a private company in San Jose that builds humanoid robots: machines with two arms, two legs and hands, sized to work in spaces built for people. It is not publicly traded. You cannot buy it on any exchange, and the shares that change hands on private-market sites are restricted.¹³ It is a different company from Figure Technology Solutions, the blockchain-based lender, which has no link to the robot maker.

The current robot is Figure 03, shown in October 2025. It has cameras in the palms and touch sensors in the fingertips that pick up forces of about 3 grams. The company says it was redesigned for mass production, with fewer parts and a switch from machined metal to die-casting, moulding and stamping.¹⁸ The robot is run by Helix, Figure’s own vision-language-action model. That means one neural network takes in camera images and instructions in plain language and outputs motor commands.¹⁸

The pitch is simple. Labour is the largest cost in warehouses, factories and homes, and a general-purpose robot could do many jobs without being rebuilt for each one. Whether that works is the open question. Today the evidence is a handful of named customer sites, a lot of demonstration video, and company statements that are hard to check from outside.¹¹¹²⁷

Think of the company as three businesses stacked together. One is a hardware maker that has to cut cost per robot. One is an AI lab that has to teach the robot new jobs. One is a service firm that has to keep fleets running at customer sites. Each is hard alone. Doing all three at once with investor money is the bet. A Figure 03 is designed to charge by standing on a pad, to be washed, and to sit next to people, which tells you the target includes homes as well as factories.¹⁸

Production is done in-house at a plant called BotQ. In March 2025 Figure said its first line could build up to 12,000 humanoids a year and that the supply chain could scale to 100,000 robots over four years.¹⁷ In April 2026 it said it had delivered more than 350 Figure 03 units and had lifted output from about one robot a day to about one an hour.⁶ Those are company numbers. An hourly rate over a short run is not the same as thousands of robots per year that customers pay for and keep using.

The company keeps adding cost lines that are not robots. In August 2026 it launched Index, a programme that pays people to record real tasks at home or work, and the company’s own post describes it as an effort to build a very large physical dataset.¹¹¹⁹ The company says it has committed to spend over $1 billion in the next 12 months on data and compute.¹⁹ In September 2026 it signed a compute deal with the cloud provider Nscale for at least $3.5 billion of computing capacity.¹⁶⁸ Both are bets that more data and more compute make the robot’s brain good enough. They are also large commitments for a company that has raised about $1.9 billion in total.³

IIHow it makes money

Figure has not published revenue, customer counts, prices or a profit and loss account. I found no public filing, which is what you would expect before a listing. Everything below about money is therefore either a company announcement, a press report, or my estimate, and I label which.³¹²

The intended model is selling or leasing robots to businesses, plus ongoing software and service. Two customer agreements are public. BMW has used Figure robots at its Spartanburg, South Carolina plant, and BMW’s own release, as reported, said a Figure 02 supported production of more than 30,000 BMW X3 vehicles over ten months, with Figure 03 now moving into logistics work.⁴ Catalyst Brands, owner of JCPenney, announced a commercial partnership on 26 May 2026, starting at its Reno, Nevada distribution centre. Financial terms, contract length and robot numbers were not disclosed.¹²

The home market is the large prize in the pitch but has no confirmed revenue.

The one revenue-like figure in circulation is a target of about $9 billion in revenue by 2029, which industry trackers say comes from fundraising materials. That is a goal, not a result. The same article estimates the 2026 revenue base as almost certainly under $100 million, and calls that its own inference.³ I use under $150 million as a deliberately loose estimate for 2026, because the number of robots in paid service is unknown and many of the 740 reported robots may be Figure’s own units.¹⁵

Money facts and how reliable they are

IIIThe market and the competition

The question is how large the humanoid market becomes and who captures it. Nobody can answer that with data yet. Sector money is flowing: one report says humanoid startups raised $8.6 billion so far in 2026, about 1.8 times all of 2025.⁸ That tells you investors are enthusiastic, not that customers are buying.

Competition comes in four groups. First, Chinese makers. Unitree listed on Shanghai’s STAR Market in August 2026. It reported 2025 revenue of 1.70 billion yuan (about $252 million) and said it shipped over 5,500 humanoids in 2025, so it has real sales at lower prices.¹⁰ Second, other US start-ups. Agility Robotics agreed in June 2026 to go public by merging with a blank-cheque company at a $2.5 billion pre-money value, and says it has more than $300 million of milestone-based orders and Digit robots at nine customer sites with over 65,000 operating hours.¹⁴ The deal article I opened does not give its revenue.¹⁴ Third, large companies with their own humanoid programmes, such as Tesla with Optimus. I could not verify Tesla’s current status in this research, so I do not rate it. Fourth, software-first players that make the robot brain and may sell to anyone.

Size estimates are unreliable. Early home robots, factory arms and warehouse systems each have different buyers and prices, and a humanoid has to take work from all of them. I do not publish a market-size number because I found no source I would trust. What matters for Figure is narrower: how many robots customers buy per year, at what price, and whether they keep paying for software. None of the three is disclosed.¹²³

Customers also have a cheaper comparison. A fixed conveyor, sorter or wheeled robot already does much of warehouse work for less money. A humanoid has to beat those on total cost, uptime and safety, not only look impressive. I cannot judge that from public data, and neither can the reader, because Figure has released no uptime or cost per task figures.

IVThe moat and how it could be disrupted

I rate the moat contestable. Figure has real advantages. It builds its own hardware and software, so it can tune one to the other. It has large investors and strategic backers that double as customers or suppliers: Nvidia, Brookfield, Intel, Qualcomm and Salesforce took part in the Series C, and Brookfield also holds a large stake in Catalyst Brands.¹¹² It is ahead of most Western peers on production claims.⁶

But none of that is durable the way a network effect or a regulatory licence is. A robot’s brain can be copied by rivals with enough data and compute. Figure itself ended its earlier tie-up with OpenAI in February 2025, about a year after the funding round that brought OpenAI in as an investor, after the CEO said language models were getting smarter but also more commoditized.²²¹²² That cuts both ways: it shows Figure values owning the stack, and it shows the underlying AI capability is not owned by anyone for long.

Switching costs could build up if a customer integrates a fleet into a workflow, but nothing public shows large fleets yet. BMW referred to one robot working at a time during a 2025 review, not a fleet, and Fortune reported that the CEO’s description of daily operations was broader than BMW’s.⁷ More recent reports describe a longer pilot of about ten months and a move to Figure 03 logistics, which is progress but still small.⁴

The ways it could be disrupted are concrete. A cheaper Chinese maker could undercut on price, as Unitree already sells at scale.¹⁰ A rival could reach good-enough task performance on an existing, cheaper form factor such as wheels. A customer could demand open data and let robot makers compete on price. A safety incident or a lawsuit could freeze home sales. A former head of product safety sued in November 2025, alleging he was fired for raising concerns that the robots could generate force enough to fracture a skull. Figure says he was fired for performance, calls the claims false, and has countersued. The case is pending and the allegations are unproven.⁵

VThe numbers we can see

There are no audited accounts. What exists is a short list of company statements and press reports. The table below lists them with their type so you can see how much is hard evidence.

On funding: Figure’s February 2024 round was $675 million at a $2.6 billion valuation. The September 2025 Series C raised more than $1 billion at $39 billion post-money, led by Parkway Venture Capital.²¹ The jump of about 15 times in 19 months was driven by hopes of demand, not reported sales. One blog totals funding at about $1.9 billion, excluding Nscale’s stake.³

On commitments: Nscale agreed to supply at least $3.5 billion of compute and took an undisclosed equity stake. Critics note it books the revenue and holds shares, and that Nvidia invests in both Nscale and Figure, so demand looks more circular.⁸ The $3.5 billion is nearly twice all money raised to date, spread over several years from H2 2027 as GPUs arrive.¹⁶³ Where the cash comes from is not public. A report from The Information says Nvidia discussed another $1 billion investment, with Figure seeking money at around a $38 billion pre-money value. No deal is confirmed.⁹

On the secondary market: indicative quotes ranged from about $162.71 a share on 16 June 2026 to about $204 in October 2026, per one blog.³ I treat these with caution. Figure published a notice on 8 July 2026 saying any transfer needs board approval, that unapproved sales are void, and that it believes Forge Global and others facilitated unauthorized transfers.¹³ So secondary prices are thin, may involve unenforceable claims, and are not a clean read on value.

A simple illustration, which is my own arithmetic and not a company figure: if a robot sold for $100,000, then $6 billion of 2031 revenue would mean about 60,000 robots a year. That is five times BotQ’s stated first-line capacity of 12,000 a year.¹⁷ The company says it plans to grow well beyond that, but it also needs buyers and suppliers to match. Prices have not been published, so treat $100,000 as a placeholder, not a fact.

On operations: the CEO said about 740 robots were deployed in June 2026 against roughly 660 human staff, of whom about 314 are salaried.¹⁵ That the robot count is self-reported and may include internal use means it does not prove paid fleet size.

What is known and how firm it is

VIWho runs it and who owns it

Brett Adcock founded the company in 2022 and is CEO. He also co-founded Archer Aviation and the hiring marketplace Vettery.⁵ A separate Adcock company, Hark, was reported to have raised $700 million at a $6 billion valuation in May 2026, which shows his attention and fundraising are not confined to Figure.⁸

Ownership is not public. The Series C list is the best guide: Parkway Venture Capital led, with Brookfield Asset Management, Nvidia, Macquarie Capital, Intel Capital, Align Ventures, Tamarack Global, LG Technology Ventures, Salesforce, T-Mobile Ventures and Qualcomm Ventures.¹ Earlier backers included Jeff Bezos, Microsoft and OpenAI.² Nscale is now a shareholder.¹⁶ I cannot see the cap table, the preferences on each share class, or the CEO’s stake, so I cannot judge how much of a sale price would reach common holders.

Three governance points are worth knowing. The CEO’s public claims have at times been broader than the customer’s description, as the BMW episode shows.⁷ The company is in litigation with a former safety head.⁵ And it controls share transfers tightly.¹³ None of these is proof of anything wrong. They are the things an owner should watch.

VIIValuation: is the last price supported?

Equity value in US$ billions. Not a share price: a private company has none. Latest mark: primary round, post-money (Series C).

What the last valuation assumes

At $39 billion post-money, the round price needs roughly $20 billion of revenue by 2031 at my 5x multiple, 15% discount rate and 25% dilution. That is about 200 times the roughly $0.1 billion I guess it makes now, with no public revenue to check. The terminal multiples of 3x to 8x are my own judgment for a mature hardware and software business. I did not verify mature public automation comparables in this research, so I do not claim that those multiples are market-tested. The only public prices I verified are for Unitree and Agility, both far above any normal multiple.

Bear, base and bull

• Bear. 2031 revenue of $1.5B (my estimate), valued at 3x revenue, discounted at 20% for five years, 30% dilution, plus $0.5B of cash. Equity value about $1.8B. This is a world where the robots stay niche and Unitree and others take price.
• Base. 2031 revenue of $6B (my estimate; below the reported $9B target for 2029³), valued at 5x revenue, discounted at 15% for five years, 25% dilution, plus $1.0B of cash. Equity value about $12B.
• Bull. 2031 revenue of $20B, valued at 8x revenue, discounted at 12%, 20% dilution, plus $1.5B of cash. Equity value about $74B. Needs large fleets in warehouses and homes.

Comparables used

Where I come down

The September 2025 price of $39 billion is more than three times my base case of $12 billion. It sits between my base and my bull case of $74 billion. To believe it, you have to believe Figure sells many tens of thousands of robots a year by 2031 at prices that keep decent margins, that its compute spend pays back, and that safety problems stay manageable. The market may be right. Private investors paid that price knowing more than I can, Nvidia was reportedly in talks to invest again, and Unitree trades far higher on revenue than my multiples. Manias also reprice quickly, as Unitree showed when it fell by more than half from its peak.⁹¹⁰ The price is over a year old and may be stale. The reported $38 billion pre-money level in the Nvidia talks suggests no big change, but it is unconfirmed.⁹

What I cannot judge

I cannot see revenue, margins, cash, burn, robot prices, the cap table, share-class terms, or how many robots are in paid service. I also cannot judge whether the demos reflect average performance. Secondary quotes are unreliable and the company says many are unauthorized.¹³ Because public data is this thin, treat my range of about $2B to $74B as wide and uncertain, not as a forecast. This is research, not investment advice.

VIIIThe IPO

No IPO plan reported. I found no public or confidential filing and no credible report of a planned listing in the sources I opened. Figure’s own notice in July 2026 restricts share transfers and warns about secondary sales.¹³ Pre-IPO broker pages call it a private company.³ I did not search SEC EDGAR directly in this session, so a confidential filing cannot be excluded. Competitors are listing: Unitree on the STAR Market in August 2026 and Agility through a SPAC planned as AGLT.¹⁰¹⁴

What to watch

• A public S-1 or a credible report of a confidential filing
• Whether Nvidia’s reported $1B investment closes, and at what valuation⁹
• Disclosure of revenue, fleet size and contract terms for BMW and Catalyst Brands¹²
• How Unitree and Agility trade after listing, as the public price signal for the sector¹⁰¹⁴
• Outcome of the safety lawsuit⁵
• Funding of the Nscale commitment ahead of the H2 2027 deployment¹⁶

IXWhat could go wrong

What most likely destroys this business is not a rival’s better robot. It is running out of runway before robots earn money. Figure has committed to at least $3.5 billion of compute with Nscale and says it will spend over $1 billion in the next 12 months on data and compute, on about $1.9 billion raised.¹⁶³¹⁹ If revenue stays small and the next investors balk, the fixed costs outrun the cash. A reported Nvidia talk is a sign that someone may still fund it, but it is unconfirmed.⁹

For the current price to be right, several things must go right at the same time. Robots must work reliably on real tasks. They must be safe next to people. They must be cheap enough that customers save money. Customers must reorder in large numbers. Figure must build them at the pace it says. And the market must still pay a rich multiple for that revenue when an investor wants to sell. If I give each of those even a 70 to 80 percent chance and treat them as independent, the combined chance is only about 15 to 25 percent. They are not fully independent, but that is why the price needs a lot to go right.

Cash needs deserve numbers. Using my own rough guess: if Figure spends $0.5 billion a year on people and plants, plus the data programme, plus the first compute payments in 2027, the roughly $1.9 billion raised could be used up in two to three years without new money. I do not know actual burn or cash, so this is an order-of-magnitude illustration only.³¹⁶

An owner can only take out what the business earns after paying for compute, data and manufacturing. Today that is negative. Cash returns to a shareholder only through a sale or listing, and until then transfers need board approval.¹³

The next best use of the same dollar is a listed company with sales already. Unitree trades publicly with real revenue, though it was at about 125 times 2025 revenue at one September close and fell more than half from its first-day peak, which shows how volatile this theme is.¹⁰ A reader comparing alternatives should weigh liquidity as well as growth.

Other risks: a safety incident or recall; export or tariff limits on parts, since Figure vertically integrates but still uses outside suppliers¹⁸; the dependence on a few strategic investors who are also suppliers⁸; and the legal risk from the pending lawsuit.⁵ I cannot judge whether the robot’s skills are as good as the demonstration videos suggest, because I have not seen independent testing.