IMPLICATOR
Project Prometheus, the secretive artificial intelligence startup co-founded by Jeff Bezos, is closing on a $10 billion funding round at a $38 billion post-money valuation, according to a Financial Times report published Monday. JPMorgan and BlackRock anchor the round, which would push the company's total capital raised above $16 billion only five months after it launched with $6.2 billion in seed funding. BlackRock declined to comment. Nothing had closed when the FT hit publish. A $38 billion price tag five months out of the gate is a number the AI industry has written maybe a handful of times.
Key Takeaways
- Project Prometheus, Bezos's physical-AI startup, is closing a $10 billion round at a $38 billion valuation per the Financial Times.
- JPMorgan and BlackRock anchor the round; total capital raised would cross $16 billion five months after launch.
- Bezos is separately pitching up to $100 billion for a holding company to buy and feed data from industrial firms.
- The lab has 120+ employees poached from OpenAI, xAI, DeepMind, Meta, Anthropic, and Nvidia, with no shipped product yet.
AI-generated summary, reviewed by an editor. More on our AI guidelines.
Physical AI, not chatbots
Prometheus is not chasing the chatbot. Bezos co-founded the lab in November 2025 alongside Vikram Bajaj, a physical chemist out of MIT. Before Prometheus, Bajaj spent years running experimental programs inside Google X. He later launched Foresite Labs to put AI onto drug molecules. The target at Prometheus is a different category of problem altogether. Industry people call this physical AI. Models that reason about engines, tolerances, airflow, crack propagation. Things large language models cannot learn by reading the internet. Per Wall Street Journal reporting cited by the Los Angeles Times, Prometheus is building systems to simulate how metal fatigues under pressure and how air flows around a wing, letting engineers test designs without bending sheet metal.
That thesis is expensive. Text data is close to free. Industrial data is not. It sits inside factories, CAD pipelines, defense contractors, and chip fabs, and it has taken incumbents decades to accumulate. The implication for BlackRock and JPMorgan is that the scarcity itself is the moat, and whoever assembles the dataset first owns a category LLMs cannot touch.
The talent raid
The money is spending quickly. Prometheus employs more than 120 people, according to coverage of its scale, with offices in San Francisco, London, and Zurich. Hires have come from OpenAI, xAI, Google DeepMind, Meta, Anthropic, and Nvidia. One April recruitment turned heads: Kyle Kosic, a founding member of Elon Musk's xAI, left OpenAI to run AI infrastructure at Prometheus, Observer reported, citing the Financial Times. Losing a founding engineer to a five-month-old startup is the kind of signal most labs notice.
Prometheus also absorbed General Agents, an agentic AI firm co-founded by DeepMind alumni Sherjil Ozair and William Guss. Both now sit on the Prometheus technical bench. The recruiting pattern is consistent. Infrastructure specialists, simulation engineers, researchers with bench experience in physics and robotics.
The holding company play
The $10 billion round is not the biggest number in the Prometheus story. The Wall Street Journal reported in March that Bezos is separately raising up to $100 billion for a holding company that would buy manufacturers, chipmakers, defense suppliers, and architecture-engineering-construction firms outright, then route their operational data back into Prometheus's models. Bezos has been traveling to the Middle East and Southeast Asia to pitch sovereign funds on the vehicle, per reporting by Business Insider.
Get the AI signal, not the noise
Strategic AI news from San Francisco. No hype, no "AI will change everything" throat clearing. Just what moved, who won, and why it matters. Daily at 6am PST.
No spam. Unsubscribe anytime.
The structure resembles Berkshire Hathaway crossed with a data moat. Buy the company. Harvest the data. Train the model. Deploy it back into the company. Then use the returns to buy the next one. Nothing about that loop is new. The scale is. If even half the $100 billion target closes, it would rank among the largest private capital vehicles ever raised, and it would give Prometheus a proprietary industrial dataset no public-internet scraper can match.
What still could break
But the numbers are a claim, not a close. The FT's sources said the round is expected to finalize soon but had not been signed; the $100 billion vehicle is, for now, a pitch. Physical AI is also slower to validate than text generation. You cannot A/B test a jet engine overnight. Partners have to hand over proprietary process data and trust that the model improves their margins rather than leaking their IP.
Prometheus enters a crowded field. World Labs closed a $1 billion round in February on world-model research. Yann LeCun's Paris-based AMI Labs raised $1 billion in Europe's largest seed. Periodic Labs, which Bezos himself backed, is chasing the same physics-aware target with $300 million. None has demonstrated a commercial deployment yet.
What Prometheus has is $16 billion, a poached founding engineer, and an empty factory floor.
Frequently Asked Questions
What is Project Prometheus and who runs it?
Project Prometheus is a secretive artificial intelligence startup focused on physical AI, co-founded in November 2025 by Jeff Bezos and Vikram Bajaj, an MIT-trained physical chemist and former Google X scientist. Both serve as co-CEOs. The company is headquartered in San Francisco with additional offices in London and Zurich. It aims to build AI systems that reason about real-world physics for engineering, manufacturing, aerospace, robotics, drug discovery, and logistics.
How much has Project Prometheus raised, and who are the investors?
The lab launched with $6.2 billion in seed funding in November 2025 and is now closing on a $10 billion round at a $38 billion post-money valuation, per the Financial Times. JPMorgan and BlackRock anchor the new round. If completed, total capital raised would cross $16 billion. BlackRock declined to comment and the round had not formally closed at publication time.
What is physical AI, and how is it different from ChatGPT-style models?
Physical AI refers to models trained to reason about real-world physics, materials, engineering tolerances, and industrial processes, rather than text and images. Large language models like ChatGPT learn from internet-scale text. Physical AI systems need proprietary data from factories, CAD pipelines, and engineering simulations. The data is scarce and expensive, which Prometheus is betting creates a long-term moat over generalist AI labs.
What is the $100 billion holding company plan?
The Wall Street Journal reported in March that Bezos is separately raising up to $100 billion for a Berkshire-style holding company that would buy stakes in manufacturers, chipmakers, defense suppliers, and architecture-engineering-construction firms. Those businesses would feed operational data back into Prometheus's models. Bezos has been meeting with sovereign investors in the Middle East and Southeast Asia to pitch the vehicle.
Who are Prometheus's main competitors in physical AI?
The field includes World Labs, the Fei-Fei Li venture that raised $1 billion in February on world-model research; AMI Labs, Yann LeCun's Paris-based startup that closed Europe's largest-ever seed at $1 billion; and Periodic Labs, which Bezos himself backed, raising $300 million for physics-aware AI. None has publicly demonstrated a commercial deployment yet. Prometheus enters a crowded but still pre-product field.
AI-generated summary, reviewed by an editor. More on our AI guidelines.
Maria Garcia
Tech culture and generative AI reporter covering the intersection of AI with digital culture, consumer behavior, and content creation platforms. Focusing on technology's beneficiaries and those left behind by AI adoption. Based in California.
