Jeff Bezos may have missed the first wave of AI, dominated by large linguistic models. But with his startup Prometheus reaching a $41 billion valuation this month, he's betting on the concept of physical AI—by building what he calls the “general artificial engineer” (AGE).
“This is an old dream, the idea that you could build a set of tools that can actually do engineering,” Bezos told CNBC earlier this month, adding that his team has been working on the concept since late 2024. Prometheus, officially founded in November 2025, is led by Bezos as CEO and Vikram Bajaj, former director of Google X.
Details about the company remain scarce. Prometheus operates from offices in London, Zurich and San Francisco and has a team of approximately 150 employees, according to Bezos. Although he said the startup is not “secret,” little has been shared publicly about its work. The Observer reached out to multiple employees but did not receive a response.
What is known is that Prometheus is developing AI-driven tools to accelerate manufacturing and construction models designed to perform physical tasks throughout the engineering workflow. This emphasis on physical AI can differentiate a company in a crowded market.
“Everyone is debating which AI model is winning. Bezos is asking a different question: what if AI could compress the time between an idea and a manufactured product?” Shashi Bellamkonda, chief research director at Info-Tech Research Group, told the Observer.
“The 'common engineer' framework is ambitious, but the basic bet that global data creates a drain that Internet-trained models can't replicate is the most logical scenario I've seen in this funding round,” added Bellamkonda.
Bezos' key is hiring to power his AI dream
Prometheus has assembled a top-notch team to pursue that vision. In addition to Bajaj, who co-founded Google Life Sciences (now Verily), the company hired Kyle Kosic, an xAI co-founder and former OpenAI employee who led XAI's Colossus supercomputer infrastructure and now works on the AI infrastructure at Prometheus.
The startup also acquired General Agents last year, bringing in founders Sherjil Ozair and William Guss, who have previous experience at DeepMind, Tesla and OpenAI. Additional hires came from Microsoft, Meta, Anthropic, Nvidia and Grammarly.
A recent LinkedIn activity suggests continued hiring. Drew Jaegle, former DeepMind researcher and Head of AI at Mirage, joined Prometheus this year as a member of the technical staff, noting on his website that he is working on virtual world AI. In one post, founding member Cristian Bodnar, formerly of Microsoft and leading the Aurora Earth System Forecasting Model, wrote, “We're putting together a world-class team at the intersection of AI and material science.”
Should we buy the hype?
Still, the company's rapid rise raises questions. A valuation of $41 billion is staggering for a startup that has yet to ship a product. While PWC estimates that the global AI market could reach nearly $500 billion, translating that potential into real-world impact remains a challenge.
“They skipped a step,” Nicholas Nadeau, founder and CTO of Onix, told the Observer. “Calling it a generic synthetic engineer avoids the real question: are they building a world model that encodes physics, or is the engineer agent thinking through the design loop using the solvers we already have? Different bets, different data requirements, and AGE allows him to avoid committing to both.”
Nadeau added that Prometheus “must be looked at, especially who is involved,” noting that few founders can integrate production chains to the extent of Bezos, given his ties to Amazon and Blue Origin.
Currently, the startup joins a growing group of highly rated, AI startups, including Mira Murati's Think Lab ($12 billion), Yann LeCun's AMI Labs ($3.5 billion) and Fei-Fei Li's World Labs ($5 billion).
