On October 2, Business Insider reported that FieldAI has signed a term sheet to raise $700M at a $10B valuation. A year ago the company was valued at $2B. Techmeme carried the Business Insider summary, and The Next Web and Benzinga filled in the numbers below. The round has not closed, and no lead investor has been named.
FieldAI does not build robots. Founded in 2023 and based in Irvine, California, it sells what it calls a universal general-purpose brain: software meant to run humanoids, robot dogs, drones and industrial rovers. Its customers, per Benzinga, sit in construction, data centers and defense.
The number that needs splitting
The company reports more than $135M in revenue and signed contracts across 30-plus customers, including at least $35M added since June. The Next Web puts the June base near $100M, which makes the gain about 35% in four months. CEO Ali Agha’s description: very, very fast growth.
Read the unit before the size. $135M is revenue plus contracts, not revenue. The reporting does not say how much is recognized, how much is signed and undelivered, or how long the contracts run. Our own division puts the valuation at roughly 74 times that figure. It is a wide basis and not a price-to-sales ratio, so treat it as a ruler and not a multiple.
Where $10B sits
Two companies sit above it. Physical Intelligence is valued near $11B and Skild AI above $14B, Benzinga reports, and all three sell the same layer: the model that tells any body what to do. One step down, Europe’s biggest robotics round this year went to a company that builds robots. NEURA raised $1.4B at about $7B. The nearest European software rival, Genesis AI, is reported to be raising at roughly $3B, under a third of FieldAI’s number. For an exit comparison, Benzinga notes that Meta bought Assured Robot Intelligence in May.
Inside our own coverage the spread is wider. In July, Humanoid raised $152M at $1.35B, with its biggest customer also its supplier and investor. FieldAI’s mark is 7.4 times that. On September 3, Nscale committed $3.5B of compute to Figure, more than Figure had raised in equity over its life. The bill for training the brain now outruns the company. FieldAI sells from the other side of that bill, and it needs no factory to do it. The $700M would be 1.7 times the $405M it had raised in total through 2025, according to TechCrunch.
The earlier cap table reads like a map of who wants a seat at this layer. Benzinga lists Nvidia’s NVentures, Intel Capital, Bezos Expeditions, Gates Frontier, Samsung, Khosla Ventures, Temasek, Prysm Capital and Canaan Partners among the 2025 backers. TechCrunch described the pitch at the time as one robot brain that generalizes across robot types and environments, with physics built into the model so a robot can judge its own confidence. A year on, the pitch is the same and the customer count is the new part.
The bet underneath
The valuation only works if the brain and the body come apart. Bodies are already shipping. IDC counted nearly 25,000 humanoids in the first half and $740M of hardware revenue, and said 69% went to research, education, performance and government data centers instead of a production line (our write-up). Plenty of bodies, thin work. On September 25 we covered Tesla building Optimus at several hundred units a week with none doing paid work, which is the same gap from the factory side. A software layer that drops into any of them collects the value of the work without owning the metal. The two figures are not like for like: $740M is world hardware revenue for six months, $135M is one firm’s cumulative revenue and contracts. They sit together here for order of magnitude only.
Boston Dynamics showed the other end of the trade one day earlier, with an Atlas hand at 13 degrees of freedom built for drills and welding torches. That is a bet that one maker owns both body and brain. FieldAI bets they get sold separately.
What the reporting leaves out
No customer names. No word on whether any of the 30-plus are humanoid makers licensing the software. No contract lengths. No task list for what the brain does on a construction site, in a data center or on a defense job. A term sheet can also change before it closes. Funded to run humanoids is not the same as running humanoids at a customer’s site, and nothing published closes that gap.
What it means for the jobs
FieldAI names three sectors, not tasks, so this part is our inference. Construction sites, data centers and defense sites all contain walk-the-site work: inspection, survey, patrol, equipment checks. If those tasks go to robots, the exposure is not to one robot maker’s roadmap. It is to whichever brain ends up in the most bodies. That changes what to watch. Unit counts from hardware makers say where bodies went. Contract announcements from the software layer say where work went.
Three signals would turn this from a valuation into a displacement story: the round closing with a named lead, a named customer that employs the people doing these site jobs, and a split of the $135M into recognized revenue and contracts. On the hiring side, Benzinga says FieldAI recruits from Google DeepMind, Tesla, Nvidia and Boston Dynamics. Robot-learning engineers now have one more bidder holding a $700M term sheet.
For workers in the three named sectors the timing question is the open one. Nothing published gives a date by which the brain does a task unsupervised at a customer. Until a customer or a task list appears, the defensible read is that capital is arriving faster than evidence, and the evidence is what sets the clock on the jobs.
