Field report NE-78 / P7 · sheet /strategy
Corporate StrategyFiled from the territory
Inside Prometheus: Bezos's $41 Billion Physical AI Bet Takes Shape
Bezos's $41 billion Prometheus has raised $18 billion, built one multi-physics model and hired 250 people from xAI, Waymo and the U.S. Army's research ranks. The talent map reveals its targets.
- Grid
- NE-78 / P7
- Surveyed
- November 19, 2026
- Reading
- 5 min read
- Correspondent
- Sophie Lindqvist

Field notes
- Prometheus is valued at $41 billion after raising roughly $6 billion and then $12 billion more earlier this year
- Bezos coleads the startup — his first executive role since leaving Amazon's CEO post in 2021
- Fast Company identified 130 of roughly 250 employees, the first public database of the secretive firm's staff
- The company has built one multi-physics model and a neural surrogate, including a Mars helicopter redesign experiment
- Advisers include a U.S. Army ground systems chief scientist and a former Anduril attorney
Prometheus, the Jeff Bezos-backed AI startup valued at $41 billion, employs roughly 250 people working toward what Bezos calls an "artificial general engineer" — and the composition of that workforce now reveals more about the company's direction than anything its leadership has said publicly. Fast Company identified 130 of them by combing LinkedIn and personal websites, assembling what appears to be the first public database of its kind.
What does this change for the race in physical AI (models trained on physical-world data rather than text)? It exposes the hiring blueprint — and the government ambitions — of one of the field's best-funded players.
What the money says
Prometheus originally raised about $6 billion, then another $12 billion earlier this year, per Semafor's June reporting. Backers include JPMorgan, BlackRock, Arch Venture Partners cofounder Robert Nelsen, Powerlaw — a closed-end fund that has also invested in SpaceX, Colossal and Waymo — and RoboStrategy, a venture fund that says it owns shares in the company.
Bezos is not merely bankrolling the venture. He coleads it, his first executive role since stepping down as Amazon's CEO in 2021. His cofounder is Vik Bajaj, a Stanford professor who previously led science operations at Verily, Alphabet's health technology arm. Prometheus was incubated in 2024 at Foresite Labs, which Bajaj leads, with support from Arch Venture Partners. The company used a code name, "FL2024-008 INC," in lobbying disclosures and H1-B visa applications — a measure of how tightly it guards its operations. Its website contains only links to its LinkedIn and X pages.
What has actually been built?
At a Semafor event on November 17, Bezos and Bajaj said most of the funding has gone to compute. The company has built one single multi-physics model and a neural surrogate (an AI approximation of numerical simulation), and has used the technology to design a range of hardware — including an exercise in which its models redesigned the Mars helicopter.
That is a demonstration, not a product line. The company has reportedly eyed a network of spin-offs in which it might one day own stakes, though nothing has materialized publicly. Bezos has also said the company is not focused on robotics, despite the term's usual association with physical AI. He likens the ambition to a far more advanced version of CAD (computer-aided design) powered by AI.
"My time at Amazon is spent on AI. My time at Prometheus is spent on AI — Prometheus is really an AI company," Bezos told CNBC's Andrew Ross Sorkin earlier this year. "And [at] Prometheus . . . the tools that we're building are going to help companies like Blue immensely."
A trademark application stretches the ambition across structural, thermal, fluid, acoustic and electromagnetic analysis, digital twin technology (virtual replicas of physical systems), 3D printing, semiconductor manufacturing and factory layouts.
Who is building it?
The identified hires suggest deliberate breadth: expertise in ocean engineering, energy engineering, mathematics, high-performance computing, aerospace and chemistry — a hint at future manufacturing and healthcare customers. Staffers have come from xAI, Google, Foresite-backed Xaira, Epic Games, Meta's Reality Labs, Waymo and Red Bull Advanced Technologies. Several arrived via General Agents, an applied AI lab Prometheus quietly acquired last year; its former CEO, Sherjil Ozair, is now on Prometheus's technical staff.
The government thread is hard to miss. One technical adviser also serves as chief scientist for the U.S. Army's ground systems research. Another adviser, Chase Johnson, is a former Marine Corps innovation unit member and former attorney at defense tech company Anduril; his LinkedIn bio suggests Prometheus could transform the "defense industrial base."
"Most fundamentally, the biggest challenge is just the sheer breadth of what we're doing. Engineering is vast. There's no such thing as an engineer," one early-stage employee told Fast Company anonymously. "This idea of building something that's general enough to cover all of that space, that's been a really, really huge challenge — and a really interesting one."
A government relations employee framed the pitch more narrowly: "We are exclusively focused on design, engineering, and manufacturing, and I think that is an advantage. Focus is always an advantage."
Why Bezos wins twice
Bezos may be Prometheus's first customer. Blue Origin is developing reusable rockets and has ambitions including orbital data centers; Bezos said at the Semafor event that Blue Origin already uses some Prometheus tech. Amazon manufactures in-house and invests heavily in warehouse robotics. Bezos has separately invested in Physical Intelligence, a robotics-focused AI startup, though Prometheus says there is no formal relationship between the two.
Who is competing?
Anima Anandkumar and Benedikt Jenik launched Accelerated Understanding after declining an offer to lead Prometheus — an offer Reuters reported came from Bajaj with a major stake and a million-dollar salary. Their startup builds models that predict in four dimensions at any resolution. "Even if one were to use LLMs to automate execution, the real world stays the bottleneck," Anandkumar said. "If we instead teach AI models to directly simulate and understand physics, that bottleneck goes away."
Paul Eremenko of P-1 AI, whose Archie tool works inside engineering teams, says he does not compete with OpenAI or Anthropic for talent. "The prototypical AI researcher or AI engineer for us is somebody who comes from a physical world background," he said — a mechanical, aerospace or electrical engineer who became interested in AI.
Prometheus expects to expand its workforce "exponentially," one employee said, even as hiring stays highly selective. Whether a jet engine's decade-long development cycle can shrink to a few years, as the company envisions, remains unproven anywhere outside its own simulations.
via fastcompany.com/rss (Source)
More from Sophie Lindqvist
Show full bio
Correspondent covering money, funding and market shifts around corporate strategy and chief-executive leadership at CEO Frontier.
483 articles
Adjacent sheets
- VCs question neolab math as funding targets reach $50 billion
- AI Dealmaking Frenzy: Nvidia, SpaceX and Amazon Reshape Startup M&A
- Rox Raises $100 Million at $1.3 Billion Valuation to Challenge Salesforce
- Companies replace blanket AI access with budgets and controls
- US Series A Funding Hits $42 Billion, But Deal Count Stalls