What is Prometheus building?

Prometheus is building an artificial general engineer to run the path from idea to manufactured product, from jet engines to chips.

How much has Prometheus raised?

Prometheus has raised $18.2 billion in under eight months, including a $12 billion Series B.

Who leads Prometheus?

Jeff Bezos co-founded Prometheus and now runs it as co-CEO.

Prometheus Pre-IPO: Bezos' AI Startup

Large language models got good by reading. Decades of text, code and images sat on the open internet, and the models trained on all of it. Nobody ever put the knowledge behind a jet engine online in the same way. Prometheus, the company Jeff Bezos now runs as co-CEO, was built around that gap.

The goal is what Bezos calls an "artificial general engineer": AI that works across the whole path from an idea to a manufactured object, through design, simulation, prototype, test and production. Bezos has said he wants that loop to run about 10 times faster than it does today. That is an aim, and the company is still early. It has raised $18.2 billion in under eight months to chase it.

What Prometheus Is Actually Building

Bezos co-founded Prometheus in November 2025 with Vik Bajaj, a chemist and physicist who worked at Google X, then co-founded Verily, Alphabet's life sciences company, and later Foresite Labs. Bezos took the co-CEO title himself. It is his first day-to-day operating role since he left the Amazon CEO chair in 2021, and he chose an industrial problem to go back to work on.

The pairing matters. Bezos brings the operating record and the capital. Bajaj brings a scientist's view of how physical systems behave, and a career spent at the edge of research and industry. Drug design sits on the company's list of targets, and Verily is where Bajaj learned what it takes to turn lab science into a product.

Bajaj explains the problem with a jet engine. A new one takes an engineering team a decade or more to design, prototype and bring into production. Thousands of engineers, long calendars, and years of proof before anything flies. The same pattern shows up in chip manufacturing, automotive production, batteries and drug design. Each runs on a loop of design, test and rework, and every turn of that loop is slow and expensive. The decade is not spent on one brilliant idea. It is spent going round the loop, again and again, until the part works and a factory can make it.

Prometheus wants to run the loop with AI, end to end. Bezos has described it as a modern version of CAD software, then said the comparison undersells it. CAD helps an engineer draw a part. Prometheus is aimed at the whole invention process, from the first idea to a product a factory can make. It is not a chatbot, and Bezos has said the company has nothing to do with robotics. It has no formal ties to Amazon or Blue Origin either, though Bezos has floated Blue Origin as a possible future customer. A rocket company is a fair picture of who Prometheus is building for: teams whose products are hard to design, costly to test and unforgiving when they fail.

The Data Problem Explains Everything

The hard part is not the model. It is what the model learns from.

A chatbot can learn language from the web because language is already there, written down by billions of people. Engineering knowledge is different. How a turbine blade fails under heat, how a semiconductor process drifts, what a production line does when one input changes: that knowledge lives inside factories, labs, simulations and test rigs. Most of it has never been published, and much of it has never been written down at all.

So Prometheus has to make its own. It generates engineering data through its own experiments and brings in real manufacturing data, so the system learns physics, design and production together. Bezos has said most of the company's capital goes to two things: compute, and building that training data from scratch.

That sets up a loop. Get inside real engineering work, generate data from the physical world, train on it, put the model back to work on the next problem, and learn from that too. Every design that runs through the system leaves it with more to learn from. Language models never had to build a loop like that. Prometheus cannot work without one, and it is the part of the company a competitor would find hardest to copy.

The Amazon Pattern, Pointed at Invention

Bezos has made this kind of move before. Amazon took the path from wanting a product to holding it and made it shorter at every step: the search, the checkout, the warehouse, the truck. Prometheus points the same instinct at a different path, from idea to manufacture.

His stated thesis is that invention drives productivity. If the time it takes to invent a physical thing falls from years to months, everything built on that invention arrives sooner. That is why the company talks about engineering in general rather than a single industry. The customer it is designing for is anyone who has to turn an idea into a physical product and currently waits years to do it.

A Startup, and an Industrial Move

On one level, Prometheus is a startup. It has about 150 people across San Francisco, London and Zurich, many of them researchers hired from OpenAI, DeepMind, Meta, Nvidia and xAI. Shortly after launch it acquired General Agents, a company building agentic AI that turns visual input into real-world actions through video-language-action models, and folded that team in. That is the step between a model that can describe a part and one that can see what happens on a test rig and act on it.

On another level, it is an industrial move. Elon Musk built his name on physical industry: rockets at SpaceX, cars at Tesla, and AI at xAI, all under one founder. Bezos now has a company pointed at the same ground, where AI meets manufacturing and heavy engineering. The two most famous founders of this era are both putting AI to work on the physical world. That is where the next industrial revolution gets decided, and it will be built out of engines, chips and production lines, not chat windows.

Prometheus has company in the category. Fei-Fei Li's World Labs, Yann LeCun's AMI Labs and Physical Intelligence, which Bezos backed in 2024, are all working on AI that understands the physical world. Prometheus is the one aimed squarely at engineering and manufacturing, and one of the best funded.

The Funding Pace Is the Real Story

Prometheus raised a $6.2 billion Series A at launch in November 2025, with Bezos as the largest individual backer. Seven months later, in June 2026, it raised a $12 billion Series B that valued the company at about $41 billion. JPMorgan Chase, Goldman Sachs, BlackRock, DST Global and Arch Venture Partners took part.

That is $18.2 billion across two rounds, less than eight months apart. The roster reads like a late-stage round, not a Series B. Banks and asset managers of that size rarely write cheques into a company this young, and they did it before any public product. Bezos has said it is premature to share what the company has built, and Prometheus has disclosed no product, customer or revenue. The backers are paying for the team, the thesis and the data engine. With most of the money going into compute and new training data, the size of the rounds is the cost of building something that did not exist before.

On 11 June 2026, the day the Series B was announced, Bezos and Bajaj went on CNBC to explain it. Bezos said an artificial general engineer is a long-standing idea that was not practical until recently, and that the team had been working on it since late 2024. Bajaj went back to the jet engine: the size of the teams, the length of the calendar, and the years of proof before anything ships.