Jeff Bezos raises $12B for AI that builds things

Bezos and Bajaj said they have assembled a set of data, drawn from both established laws of physics and testing results from manufacturing companies they declined to name. “We create that data for the most part ourselves, but we also obtain it where we can from other sources,” Bajaj said.

Some of their data “comes from collaboration with companies that are improving their processes,” Bezos said. When asked whether that data was proprietary, Bezos said: “What we’re doing is so difficult, the last thing I’m worried about is moats.”

He’s also not worried about job losses. Manufacturing employs about 8% of US nonfarm workers, but Bezos said the economy could see a labor shortage as AI makes it easier and cheaper to build new things.

Prometheus employs about 150 people and operates a large GPU cluster for its internal use, as well as buying from other providers. The former Amazon CEO said compute is “absolutely” in short supply and “probably will continue to be for some time.”

The pair wouldn’t talk about the second part of their vision: a Berkshire Hathaway-esque portfolio of companies that uses Prometheus’ AI models to manufacture things faster and more cheaply. They have been raising funds to acquire existing industrial giants or build competitors from scratch to reshape the physical economy much the way LLMs are now challenging the knowledge economy.

Prometheus offers an answer — a hazy one, given the few specifics on offer — of something that’s been hard to see two years into this hype cycle.

We now know how AI is helping companies cut costs by replacing humans with agents and robots, and Bezos and Bajaj repeatedly cited the advances in productivity we’re getting by automating the long, iterative process of refining prototypes.

But PwC has estimated that most of AI’s $15 trillion contribution to global economic growth by 2030 won’t come from gains in productivity but from gains in consumption, by enabling the creation of more goods that people want to buy. It will make more things, not just cheaper things. By making it cheaper to tinker, we’ll get more tinkering, and some of those ideas will be entirely new goods that will take their place next to the bicycle of the 1890s or the desktop computer of the 1990s as mass-market goods we didn’t know we wanted.

The limited details about what Bezos’ AI lab is trying to do opens him up to critics who say the billionaire’s endeavor may amount to him using his clout to get institutional investors to throw gobs of money to chase after AI FOMO. Last year, Elon Musk called Bezos a “copycat” in a post on X after news of the venture leaked.