The 150 person startup, which calls its work "physical AI" — applying large model techniques to robotics and engineering design — says it is building what Bezos terms an "artificial general engineer," a system meant to take over a human R&D engineer's work, but has not disclosed…
The $12 billion Jeff Bezos and his co-founder Vik Bajaj just raised for their 150-person startup Prometheus is unusually large for a company that has not yet named a product, a customer, or a timeline. The new round, reported by Ars Technica, values Prometheus at $41 billion on top of an initial $6.2 billion raised last year, and brings JPMorgan Chase, Goldman Sachs, and BlackRock onto the cap table alongside a sizable personal check from Bezos. The bulk of the cash is earmarked for compute and for producing proprietary training data, the two most expensive line items in modern AI development, before any public roadmap has appeared.
Prometheus describes its work as "physical AI," a category the founders use to mean applying the deep learning techniques behind large language models to robotics, manufacturing, and engineering design. In on-record remarks, Bezos has framed the target as an "artificial general engineer," a system meant to take over the work of a human research and development engineer the way a steam engine or a mechanical plow once redrew what labor could produce. Bajaj, a physicist and longtime Bezos collaborator who has held board roles at technology companies including Matterport, has argued that engineers' design tools have been largely unchanged for decades and that better AI tooling could let small teams produce designs in days that today take months.
The terms are worth marking carefully, because they carry a lot of weight. "Artificial general engineer" is not a phrase with a settled technical meaning in the machine learning research community. It is a Bezos coinage for a goal, not a description of a built system. Per Ars Technica's account of the founders' remarks, Prometheus has not published an architecture, named a target industry, announced a first customer, or given a delivery date for any specific product. The funding round is real and the investor list is real, and the company is otherwise a stated mission backed by a stated team.
That framing is not, on its own, a reason to dismiss the bet. In his CNBC remarks, Bezos said the goal of the work is technological breakthroughs that produce "civilizational wealth," and not just returns for a single firm. The reader can weigh that against the more measurable fact that his track record on founder-stated timelines for physical-product ventures, Blue Origin most visibly, is mixed. A $41 billion private valuation against a 150-person team with no published product is the kind of receipt-versus-rhetoric gap that will resolve only one way: with either a named customer, a published technical paper, or a hire that maps to the compute and data line items the round is funding.
Separately, Bezos and Bajaj are reportedly working to raise a roughly $100 billion investment fund that would back companies able to use or be tied to the Prometheus platform. That vehicle was reported by The New York Times and cited in the Ars piece; it has not been confirmed by Prometheus or by the named banks, and is best read as fund-raising activity rather than a closed pool of capital.
The first independent signal of progress at Prometheus will not be a press release. It will be a named customer, a published technical paper, or a senior hire from a robotics or chip company that maps to the compute and proprietary data line items the round is funding. Until one of those appears, the story is a $41 billion receipt, a stated goal, and the specific gap between them.