Amazon founder Jeff Bezos' AI startup Prometheus just raised $12 billion at a $41 billion valuation. The startup co-founded by Bezos and Vik Bajaj, the former co-founder of Verily, Google's life sciences unit is building an "artificial general engineer" in the realm of physical AI. To do this Prometheus is essentially rewriting the playbook used by AI companies such as OpenAI, the maker of ChatGPT. Instead of feeding software digital text, the company is training AI on the laws of physics, material science, and real-world industrial constraints.
"If I read a thousand books on being a gymnast, I still wouldn't be a good gymnast. There are different kinds of knowledge that you need to train these models on the right kind of data," Bezos recently said during an interview.
The core idea of the startup is to improve the efficiency of companies that design and build computers, automobiles, spacecraft and other physical products.
The billionaire Amazon founder who stepped down as its CEO in 2021 to focus on next-generation ventures such as space exploration and AI said: "The LLMs (large language models) as impressive as they are and as useful as they are - they have trained on this giant corpus of humanity's knowledge that was already pre-existing on the internet."
Bezos' co-founder Bajaj told NYT: "You can't build something like a jet engine with words alone - not even the words of mathematical equations.
"It is about multidimensional forces and fields and how they are changing over time."
Prometheus is building software that is capable of automating the design and manufacturing of a wide range of complex physical systems, from jet engines to drug compounds. However, the company remains tight-lipped about specifics at this point.
Prometheus has 150 employees across offices in San Francisco, London, and Zurich. Bezos, who primarily lives in Miami, Florida has been increasingly spending time at the SF HQ of the startup where he's co-CEO along with Bajaj.
Decoding Physical AI
To understand what Prometheus is actually building, it helps to look at the current AI landscape. Think of today's popular generative AI tools such as ChatGPT or Google Gemini as incredibly well-read academics. They have devoured almost every book, essay, and website on the internet. They can write poetry, translate languages, write code, and generate images in seconds.
But if you can't hand them a wrench and ask them to fix a leaky pipe under your sink. They only understand the world through words and digital patterns.
Physical AI is the effort to change that. It's about giving AI an understanding of the universe, not just language. Instead of training a model on sentences, companies in this space train software on the laws of gravity, thermodynamics, fluid dynamics, and material strength.
If regular AI is a brain in a jar, Physical AI is a brain that understands texture, weight, resistance, and friction. For Prometheus, this doesn't necessarily mean building walking humanoid robots. Instead, it's about creating an "artificial general engineer" - software that inherently understands that you cannot build a rocket booster out of a material that will melt under a specific thermal load. It basically bridges the massive gap between digital code and real-world concrete, steel, and carbon fiber.
Executing this shift is vastly more complicated than training a chatbot to write an essay. Commenting on this technical hurdle, Dr. Srinivas Padmanabhuni, AI expert and CTO of AiEnsured, told NDTV that while physical AI is undoubtedly the next frontier, its complexity can't be underestimated.
"Learning a real-world, messy physical reality and navigating those bottlenecks is harder and way more complex than next-token prediction," Padmanabhuni said, arguing that the industry needs a major reallocation of resources to get there. "In fact, the bulk of investments currently going into LLMs should be redirected towards this kind of model development."
The Land Grab for the Real World
Bezos is far from alone in realizing that the next big opportunity in artificial intelligence is beyond chatbots. A massive, heavily funded ecosystem is rapidly forming around physical AI, with several key players taking distinct approaches to dominating the physical world.
Take Physical Intelligence (Pi), a San Francisco-based startup that is essentially trying to build a universal "brain" for robotics. Founded by prominent AI academics and ex-Google DeepMind researchers, Pi creates foundational software that can be plugged into any robot, whether it's an industrial arm in a car factory or a folding machine in a commercial laundry.
The market is moving incredibly fast here; after raising a $600 million Series B led by Alphabet's CapitalG late last year, Pi has been in talks to secure another $1 billion round, which would rocket its valuation past the $11 billion mark. Interestingly, Bezos himself was an early individual investor in the company.
Then there is the hardware-heavy approach championed by Figure AI. Valued at roughly $2.6 billion after drawing big-ticket investments from Microsoft, Nvidia, and Amazon, Figure is focused on building autonomous humanoid robots. They have already deployed their hardware in BMW manufacturing facilities to handle dangerous, repetitive warehouse tasks, betting that the future of physical AI looks explicitly human.
Elon Musk is steering Tesla down a similar path, aggressively funneling billions into the development of the Optimus Gen 3 humanoid robot with the goal of deploying them directly onto Tesla's own automotive production lines to solve labor shortages.
But before a robot can build a machine, an AI has to simulate how that machine will behave. That is where companies like PhysicsX come in. Founded by former Formula 1 engineers, the London-based startup builds the fundamental simulation infrastructure that sits beneath heavy manufacturing. Their AI allows industries like aerospace and defense to simulate stress, aerodynamics, and physics before a single piece of metal is cast. Highlighting the intense investor appetite for this underlying tech, PhysicsX secured a $300 million Series C round led by Temasek, more than doubling its valuation to $2.4 billion.
Prometheus' latest funding round came from Bezos himself, as well as from JPMorgan Chase, Goldman Sachs, and BlackRock, among others.
While the first wave of the AI boom was defined by software that could write and talk, the next frontier may be about technology that can make and move.
