Jeff Bezos Raises $12 Billion at a $41 Billion Valuation to Build Prometheus

Jeff Bezos emerged from stealth on June 11, 2026 to announce that Prometheus, the physical AI startup he co‑leads as chief executive, had raised $12 billion at a valuation of $41 billion. This is Bezos's first CEO role since handing the reins of Amazon to Andy Jassy in 2021. The round included personal investment from Bezos alongside institutional capital from JPMorgan Chase, Goldman Sachs, BlackRock, DST Global, and Arch Venture Partners, the deep technology fund that co‑founded the company from seed formation. It is Arch's largest investment in its history.

The raise is the second for Prometheus. The company launched in late 2024 with an initial $6.2 billion raise, and the new $12 billion follows roughly seven months later. At $41 billion, Prometheus is one of the most richly valued AI startups ever funded, and one of the largest single bets made on what the industry has begun calling physical AI, a category defined by the application of AI to the design, engineering, and manufacturing of real‑world objects rather than the generation of text, code, or images.

What Prometheus Is Actually Building

The company's goal, in Bezos's own words, is an artificial general engineer. The phrase is deliberate. Large language models like GPT and Claude are sometimes described as moving toward artificial general intelligence, systems capable of reasoning across arbitrary domains of text and knowledge. Prometheus is attempting something analogous but for the physical world: a system capable of reasoning across arbitrary domains of engineering, from the aerodynamic design of a jet engine to the formulation of a drug compound to the structural analysis of a skyscraper.

Bezos has described the core problem as one of cycle time. The gap between an engineer's dream and a manufactured product at scale is currently measured in years. There are simulation steps, prototyping steps, failure analysis steps, supplier qualification steps, regulatory approval steps, and manufacturing ramp steps, each of which is slow, expensive, and dependent on domain‑specific expertise that is concentrated in relatively few people. The ambition of Prometheus is to compress that cycle by ten times or more, using AI that can reason through the design and manufacturing process with the breadth of a generalist and the depth of a specialist.

Co‑CEO Vik Bajaj, who co‑founded and ran Verily, Alphabet's life sciences company, before joining Prometheus, has described the goal as facilitating the engineering process end to end. That means AI that can design products, predict their performance under real‑world conditions, identify failure modes before physical prototyping begins, and guide the manufacturing process itself. Bajaj and Bezos have been careful to position this as amplification rather than replacement of human engineers. The pace of physical creation, Bajaj has argued, is nowhere near the pace of human imagination. Prometheus, in their framing, exists to close that gap.

The Data Problem That Makes This Hard

Bezos has acknowledged what makes Prometheus fundamentally different from language model development: there is no internet of physical engineering data to scrape. When OpenAI and Anthropic built their foundational models, they trained on a vast corpus of text that the internet had been generating for decades. Every book, article, paper, code repository, and forum post was available as training data. For physical engineering, no equivalent corpus exists in digitised, accessible form. The specifications for a jet engine, the failure modes of a complex mechanical system, the manufacturing tolerances of an industrial component, the trade‑offs in a drug formulation process, all of this knowledge is largely locked inside proprietary company databases, paper files, and the heads of senior engineers.

This is why, earlier this year, reports emerged that Bezos was exploring a $100 billion fund to acquire or invest in legacy industrial and manufacturing companies. The thesis, if confirmed, would be strategically coherent. You do not find physical engineering data. You acquire the factories, companies, and research organisations that generate it. That data, accumulated across decades of real production, would be the training foundation that Prometheus cannot build by scraping the public web.

Bezos's Argument on Jobs

The question of what Prometheus does to employment is unavoidable when the stated goal is to automate large swaths of engineering work. Bezos has addressed it directly, in what he calls a labour scarcity thesis. His argument is that AI‑driven productivity gains will increase the total output of the global economy so significantly that demand for human workers will ultimately outpace supply. The economy will produce so much more, and create so many new categories of work, that the net effect will be a shortage of labour rather than an excess of it.

The argument is contested by economists who have studied previous automation waves, and the short‑term distributional effects of productivity‑enhancing technology are well documented. Bezos knows something about this at firsthand. Amazon, where he remains executive chairman and largest individual shareholder, employs more than 1.5 million people and has cut tens of thousands of roles over the past year as CEO Andy Jassy has accelerated automation across the company's warehouse and logistics operations.

Both Bezos and Bajaj have been consistent in saying Prometheus is building tools to help smaller teams of engineers accomplish more, not software to eliminate engineering departments. Whether the two things are distinguishable in practice over a ten‑year horizon is the kind of question that will be asked a great deal as Prometheus moves toward commercial deployment.

The Company Today

Prometheus currently employs approximately 150 people across offices in San Francisco, London, and Zurich. Bezos confirmed that a substantial portion of the $12 billion will go toward compute, specifically GPU clusters, which are the core infrastructure requirement for training and running models at the scale that physical engineering simulation demands. The company also operates its own large GPU cluster for internal research, supplemented by external compute providers.

At 150 employees and $12 billion raised, the capital‑to‑headcount ratio is extraordinary, and reflects the compute‑intensive nature of the research programme rather than a large commercial sales or deployment organisation. Prometheus is in the research and early development phase. It has no commercial product currently available.

Key facts about the round and company:

• Series B raise: $12 billion
• Post‑money valuation: $41 billion
• Investors: Jeff Bezos (personal), JPMorgan Chase, Goldman Sachs, BlackRock, DST Global, Arch Venture Partners
• Previous raise: $6.2 billion (late 2024)
• Co‑CEOs: Jeff Bezos and Vik Bajaj
• Headcount: approximately 150 employees
• Offices: San Francisco, London, Zurich
• Arch Venture Partners: largest investment in the firm's history

The physical world has been the last frontier for AI precisely because it is so hard to digitise. Language, images, and code exist natively in forms that computers can ingest and learn from. Steel, chemistry, aerodynamics, and mechanical systems do not. If Bezos and Bajaj can solve the data problem and build models that reason reliably about physical engineering at scale, the commercial and social consequences would be profound. The $12 billion is Bezos's bet that this is the most important unsolved problem in artificial intelligence, and that Prometheus has the team, the compute, and now the capital to be the company that solves it.