Jeff Bezos is one of the richest people who has ever lived, spends his days on rockets and philanthropy, and had no obvious reason to go back to being a hands-on operator. He did it anyway. In November 2025 he took the title of co-CEO of a new AI company called Project Prometheus, and the scale of the thing is hard to overstate: it launched with $6.2bn, and by mid-2026 had pulled in a $12bn Series B at a $41bn post-money valuation, for roughly $18.2bn of committed capital before shipping a single product. That makes Prometheus one of the best-funded private companies in the history of technology, and it has, so far, sold nothing.
Continuing our AI Innovation series, this is the boldest and best-financed entry in the field, and the one where the gap between ambition and evidence is widest.
The founders: the billionaire and the physicist
Bezos is the headline, but he is not doing this alone, and his co-founder is the more telling hire. Vik Bajaj is a chemist and physicist who spent years at Google X, the moonshot factory, and went on to co-found and run Foresite Labs, an AI-driven incubator at the intersection of computation and the life sciences. He is co-founder and co-CEO alongside Bezos. That pairing is the whole thesis in miniature: the world’s most relentless operator of physical logistics, and a scientist whose career is about pointing computation at the messy real world rather than at screens.
Bezos, for his part, has pushed back on the idea that any of this is secretive. “We’re not being secretive,” he said in mid-2026, describing the company plainly as building tools to make it far easier for engineers to design physical objects. The team he has assembled backs the seriousness up: nearly 100 employees, with researchers hired away from OpenAI, DeepMind and Meta, operating out of San Francisco, London and Zurich.
What they are actually building
Prometheus is chasing what it calls an “artificial general engineer”. Strip the grandeur and the pitch is concrete: AI that can help invent, design and manufacture complex physical products. Bezos has described the system as a “very, very modern version” of CAD, the computer-aided design software that engineers have used for decades to draw parts and machines.
That analogy is more radical than it sounds. Today’s CAD is a drawing tool: it does what the engineer tells it. An artificial general engineer would do the engineering, proposing designs, simulating how they behave in the real world, and feeding that straight into manufacturing. If it works, it compresses the loop between “we need a part that does X” and “here is the part, and here is how to make it” from months to hours. For a man who built the most sophisticated physical logistics operation on earth, and who runs a rocket company, the appeal is obvious.
| Project Prometheus at a glance | Detail |
|---|---|
| Founders / co-CEOs | Jeff Bezos and Vik Bajaj (ex-Google X, Foresite Labs) |
| Founded | November 2025 |
| Capital committed | ~$18.2bn ($6.2bn launch + $12bn Series B) |
| Valuation | $41bn post-money (mid-2026) |
| Headcount | ~100, hired from OpenAI, DeepMind, Meta |
| Mission | An “artificial general engineer” for designing and making physical products |
Why the ambition is genuinely thrilling
Software has eaten software for two decades. The physical economy, the world of factories, machine parts, materials and supply chains, has barely been touched by AI, largely because it is so much harder: atoms do not behave like text, and a wrong answer can mean a broken machine rather than a typo. Prometheus is aimed squarely at that untouched, enormous prize. If AI can genuinely accelerate how humans design and build physical things, the upside is not another chatbot; it is faster drug-delivery devices, cheaper clean-energy hardware, better rockets. Bezos going back to the trenches for it is, at minimum, a strong signal that a serious person thinks the moment has arrived.
The friendly sceptic’s corner
Here is the discomfort, said with respect. Eighteen billion dollars and zero product is an extraordinary ratio, even for AI. It buys the best people and the most compute, but it also sets an expectation that only a genuinely category-defining result can satisfy. The history of “we raised a fortune before we had a product” is not uniformly happy.
Then there is the hardness of the actual problem. Language models got good because the internet handed them a near-infinite training set of text. There is no equivalent trove of high-quality data for “how every physical object behaves when you build it”. Engineering knowledge is fragmented, proprietary, and often not written down at all. An artificial general engineer needs to learn physics and manufacturing reality, and the data to teach it is far scarcer and messier than the data that taught ChatGPT to write.
And there is Bezos himself. His attention is a finite resource split across Blue Origin, his investments and his philanthropy. “Co-CEO” is a real title, but nobody expects him in the lab at 2am debugging a simulator. The day-to-day rests on Bajaj and the team, and the company will live or die on execution that a famous name cannot supply.
How to think about it: bull, bear, neutral
Prometheus is private, so this is about how to read it, not a trade. (None of this is investment advice.)
Bull: the largest untouched market in tech (the physical economy), attacked by an operator with a unique record in physical logistics, near-unlimited capital, and a team poached from every top lab. If anyone can brute-force the artificial general engineer into existence, it is the group with $18bn and Bezos’s obsessiveness.
Bear: a staggering valuation resting on a vision with no product, no proven data strategy for the physical world, and a celebrity co-CEO whose focus is divided. The gap between “modern CAD” and “AI that engineers for you” is enormous, and money does not shrink it.
Neutral: watch for the first real demonstration, an actual physical object designed and simulated end-to-end by the system, and watch the burn rate against milestones. With this much capital, the question is not survival, it is whether the output ever justifies the cheque.
What this means
Project Prometheus is the purest test of a simple question: can you buy a breakthrough? Bezos has assembled more money, talent and ambition around the physical-world AI problem than anyone in history, and pointed it at a prize that would genuinely matter if won. The optimism is earned by the people and the target. The scepticism is earned by the price and the silence where a product should be. Both are entirely rational right now, which is exactly what makes it the most fascinating company on this list to watch.
Related on Top Tool Stack: World Labs: spatial intelligence · Accelerated Understanding: the team that told Bezos no
Did you know: Prometheus committed roughly $18bn before releasing a product. That is more than the entire annual research budget of many of the world’s largest companies, spent on a bet that AI can learn to engineer the physical world.