Semafor · Nov 17, 2026 · 32m · 23m saved
With Jeff Bezos, Founder of Amazon and Owner of Blue Origin · Vik Bajaj, Co-Founder and Co-CEO of Prometheus
Photo via Wikipedia
Jeff Bezos, founder of Amazon and owner of Blue Origin, said on Semafor's Silicon Valley & The World that Prometheus, the engineering AI startup he co-leads, now has 250 people and has raised about $14.5 billion, a lot of it already committed to computing power.
Most of the AI race runs on language models trained by reading. Jeff argued those models cannot do engineering, because the knowledge it takes is not something a model can learn from text.
"If you don't bring Jensen back, I would not hire an outsider. Nvidia is too complicated. I think outside CEOs in a complex company have a vanishingly small chance of succeeding." — Jeff Bezos
Jeff invested in Prometheus at the seed stage in late 2024 before joining it, and he has advised every US president since Bill Clinton.
The full interview is covered here so you can skip it. 32 minutes of audio, 9 minutes of reading.
Here are the 10 lessons that matter.
Key Takeaways
Language models cannot do engineering, Jeff argued, because engineering knowledge is not learned by reading
Prometheus has 250 people and about $14.5 billion raised, much of it already committed to compute
Most companies have no engineering data worth guarding, but those that do should not share it
Housing costs rise because policy subsidizes demand and restricts supply, and only more building lowers them
If Nvidia's board does not bring Huang back, an outside CEO has "a vanishingly small chance"
AI investors are becoming more discerning, which Jeff called a normal part of the cycle
1. LLMs Can't Do Engineering
Jeff described Prometheus as an attempt to do for engineering and manufacturing what large language models did for text. He argued the existing models cannot get there, and gave a thought experiment to show why.
Jeff Bezos says knowing physics does not teach anyone to build a product
"You pick Isaac Newton and Einstein and Maxwell. Take those six brilliant physicists, put them in a room together and say, 'Build me a toaster oven.' They would have no idea how to do it." — Jeff Bezos
Prometheus instead builds neural surrogates, AI models that stand in for slow numerical simulations, along with models trained on specific engineering fields. Vik Bajaj, Prometheus's co-founder and co-CEO and a former Google X executive, said specialization is required because engineering is so varied. He pointed out that even Anthropic's coding-focused models beat generalist ones by specializing.
"Engineering is very diverse and very heterogeneous and is unlike any of those other things. So you have to really specialize in it to perform well." — Vik Bajaj



