It is a Tuesday morning in Spartanburg, South Carolina, and a humanoid robot is putting a sheet-metal panel onto a chassis without looking up. It does this 89,999 more times. Then a colleague rolls it into a side bay and retires it - because the next one is better.
§01A start-up that ships robots, not slides
Figure is a four-year-old company headquartered in San Jose with roughly 180 employees, one general purpose product line, and the kind of cap table normally reserved for nation-states. NVIDIA invests. So does Microsoft, Intel, Salesforce, T-Mobile, LG, Qualcomm, Brookfield, and a man named Jeff Bezos. The most recent round - a Series C that closed in September 2025 - cleared a billion dollars at a post-money valuation of thirty-nine billion. That was a fifteen-fold jump from the prior round eighteen months earlier. The robots have not, technically, started shipping at volume.
This is the part where a reasonable person raises an eyebrow. Figure invites the eyebrow. Then it shows you tape.
In the last eighteen months the company has built two production-grade humanoids, retired one of them after racking up more than a thousand runtime hours on a working car line, unveiled an in-house foundation model that drives two robots at once, and broken ground on a Bay Area factory designed to ship 100,000 humanoids a year. Almost nobody outside robotics has heard of any of that. Almost everyone inside robotics has an opinion about it.
§02The economy is short two warm bodies
The pitch, stripped of pretension, sounds almost rude. There is not enough labor to do the unwanted, repetitive, physically punishing jobs that prop up modern supply chains. Warehouse picking. Sub-assembly. Overnight stocking. Cleaning. There is a shortage today and a deeper shortage queued up for the 2030s as demographic curves do what demographic curves do.
You can solve that two ways. You can pay much more for the work, which raises prices on everything else, or you can build a machine that does the work. The humanoid bet is that the second option is finally tractable - because language models, vision transformers, and modern actuators have, by chance, arrived in the same five-year window.
Figure was not the first to spot this. It was simply willing to bet a billion dollars of other people's money that the path was hardware and software, vertically integrated, with no flinching at either end. Tesla is doing something similar with Optimus. Apptronik is doing it with Apollo. Boston Dynamics is doing it with Atlas. The pack is real. So is the cliff they are all running toward.
§03Brett Adcock, third venture in
Brett Adcock has the resume of a man who refuses to learn the wrong lesson. He co-founded Vettery, a recruiting marketplace, and sold it. He co-founded Archer Aviation, an electric vertical-takeoff aircraft company, took it public, and stepped back. Then, in 2022, he started Figure. As of early 2026 Forbes put his net worth at roughly $19 billion. He is forty.
The bet behind Figure was not "robots are cool." The bet was: if you assemble a single team that owns hardware, electronics, control software, and a foundation model under one roof, you can iterate faster than the firms doing only one of those things. So Adcock hired from Boston Dynamics for hardware, Tesla for manufacturing, Apple for industrial design, and Google DeepMind for AI - and refused to outsource any of it.
The result, two years later, was Helix.
Figure - the short version
§04Figure 03, and the model that lives inside it
Figure 03 is five-foot-eight and weighs sixty-one kilos. It carries twenty kilos in either hand. It runs for about five hours, then sits on the floor and charges wirelessly through its feet - which is the kind of design detail that earns standing ovations from electrical engineers and shrugs from everyone else. It is covered in soft textiles and multi-density foam, because at some point it will run into a person, and the company has decided in advance how it would like that meeting to go.
Its hands register three grams of pressure - lighter than a paperclip. Its cameras run at twice the frame rate of the previous generation, with a quarter of the latency and a sixty percent wider field of view. None of this matters, by itself. It matters because of what runs on top.
That is Helix - Figure's proprietary Vision-Language-Action model. Tell the robot, in English, to clear a table. Helix takes the request, looks at what is on the table, decides what to grab first, and moves the right wrist, fingers, torso and head, in continuous high-rate control, to get the job done. It also does the trick that nobody else has done in public: it can drive two robots at once, on the same task, with objects neither of them has ever seen before. That is not a parlor trick. That is the difference between "a robot" and "a workforce."
§05BMW, Brookfield, and the receipt drawer
Robotics is a field with a long, embarrassing relationship with the demo video. Companies show a robot performing one task, beautifully, in one rehearsed environment, and ask you to assume the rest. Figure has played that game and also moved past it.
The longest-running receipt is BMW. Figure 02 was deployed at the company's Spartanburg, South Carolina plant - the biggest BMW plant in the world - for roughly eleven months. It logged more than 1,250 hours of real production runtime and loaded more than 90,000 parts onto vehicles being built for paying customers. When Figure 03 was ready, the older model was retired.
The second receipt is more recent and, depending on your priors, more interesting. In May 2026 Figure livestreamed a fleet of its robots processing packages around the clock for the better part of a week. The company then ran a ten-hour competition between one of its robots and a human picker. The robot reportedly hit 98.5 percent of human throughput. The number is, at the very least, suggestive. It is also, depending on the union contract, terrifying.
From a seed round to thirty-nine billion in 28 months
Three rounds. One curve. The startup convention of "build first, raise later" turned upside down and given a stiff drink. Sources: PR Newswire, TechCrunch, Sacra.
§06BotQ, or how to scale a body
Figure's stated mission is to develop general-purpose humanoids that take on labor shortages, unsafe work, and the long tail of jobs no one wants. The harder, less photogenic part of that mission is BotQ - the company's manufacturing facility, designed for 12,000 robots in year one and scaling to 100,000 a year. For comparison, that is more units than many actual car plants ship of actual cars.
This is the part the AI press tends to skip. Software does not bend metal. A foundation model does not solve supply contracts. Figure is choosing, deliberately and at expense, to learn to be a hardware manufacturer at automotive scale. Whether it can is, candidly, an open question. Tesla discovered that "production hell" is a real and humiliating place. The cap table backing Figure assumes, with a straight face, that it can.
If it works, the unit economics matter more than the press release. A humanoid that costs roughly $20,000 - the reported target for the home version of Figure 03 - and works two shifts a day in a warehouse pays for itself in months, not years. That is the math the Series C investors did before they wrote the check.
§07The argument against shrugging
It is fashionable to roll one's eyes at humanoid robots. They have been five years away since 1965. The honest version of the story is that for the first time, the three ingredients - the brain, the body, and the bill of materials - are all roughly here at once, and a small number of well-capitalized companies are racing to put them together.
If Figure is right, the consequences are not subtle. Warehouses staffed by humanoids. Factories that build the factories that build the humanoids. Eventually a robot folding laundry in someone's house, which is the use case Brett Adcock keeps pointing at and the one his investors keep underlining. If Figure is wrong, it will join a famous list of well-funded companies that taught everyone else how the work was supposed to be done.
Either outcome is interesting. One of them, you should probably be following.
Back, then, to Tuesday morning in Spartanburg. The first robot is gone now - retired, posed somewhere in a hallway, probably labeled. The line is still moving. A different robot is at the station. It is lighter. Its hands are softer. It is being told, in English, what to do. It is not looking up.
The links file
Official channels, primary sources and watch-this-now videos. Open in a new tab and judge for yourself.