Bezos's Prometheus Raises $12 Billion to Automate Heavy Engineering

Jeff Bezos has a new $12 billion answer to a question almost nobody in artificial intelligence has been able to answer cleanly: what does AI actually build once it stops writing text?

His startup, Prometheus, came out of stealth on Thursday with a $12 billion raise that values the company at $41 billion. Bezos is co-CEO alongside Vik Bajaj, a former co-founder of Verily, Google's life-sciences arm. It is Bezos's first operating role at the top of a company since he handed Amazon to Andy Jassy in 2021, and he is putting his own money in beside JPMorgan, Goldman Sachs, BlackRock, and the venture firms DST Global and Arch Venture Partners.

The pitch is a phrase: an "artificial general engineer." Where the large language models from OpenAI and Anthropic were trained by ingesting a sea of internet text, Prometheus says it is ingesting the physical world — the laws of physics plus testing data from manufacturers — to speed up the design and production of things like jet engines, skyscrapers, smartphones, and drug compounds.

Bezos put the ambition in a ratio. Something that today was going to take 100 engineers 10 years to build, if you can change that to taking 10 engineers one year to build, you're just going to get way more things built, he told Semafor.

It is worth being precise about what was disclosed and what wasn't, because the gap is the story. Prometheus employs about 150 people across San Francisco, London, and Zurich, runs its own large GPU cluster, and buys more compute on top of that. Bezos said compute is absolutely in short supply and probably will be for a while. Beyond that, the founders are keeping the product under wraps. Asked whether the manufacturing data feeding the models is proprietary, Bezos gave an answer that sidestepped the moat question entirely: What we're doing is so difficult, the last thing I'm worried about is moats.

That confidence is doing a lot of work for a company that won't say what it has shipped. A $41 billion valuation on an undisclosed product is a bet on Bezos and Bajaj, not on demonstrated output — and the precedent cuts both ways. The same engineering-heavy, capital-heavy thesis sits behind a wave of "physical AI" deals investors have been chasing on the argument that the real world creates defensibility that pure software can't.

The part of Bezos's argument most likely to age strangely is about jobs. He has spent recent months pushing back on the fear that AI destroys work, predicting the opposite — what he calls labor scarcity, a world where demand for workers outruns supply. Significant productivity in the economy is going to raise the standard of living, he said, sketching two-earner households that become one-earner households by choice. It is a tidy theory from a man whose other company, Amazon, where he remains executive chairman and largest individual shareholder, has cut tens of thousands of jobs over the past year while accelerating its own automation.

There is also a second act the founders wouldn't discuss: reporting that Prometheus is tied to a far larger effort to raise capital and buy or build industrial companies outright — a Berkshire-style portfolio that would run on Prometheus's models. If that is the real plan, the $12 billion announced Thursday is a down payment, not the headline.

For now the company has a name from Greek myth, a roster of marquee financiers, and a sentence about jet engines. Bajaj summarized the data sourcing in a way that captures how much is still taken on faith: We create that data for the most part ourselves, but we also obtain it where we can from other sources. Whether the artificial general engineer designs anything the rest of us can see — that remains, for $41 billion, an open question.