Jeff Bezos, co-founder of Amazon, has returned to an operating role as co-CEO of Prometheus, a new startup focused on "physical AI." The company has secured significant funding, with a Series B round raising $12 billion, bringing its total to $18 billion ($6.2 billion initially), and valuing the company at approximately $41 billion. This funding comes from entities like JPMorgan Chase, Goldman Sachs, BlackRock, and Bezos's personal wealth, much of it allocated to acquiring the substantial computing power required for their compute-intensive endeavors.
Prometheus's core objective is to create an "artificial general engineer" by applying deep learning principles to real-world engineering and manufacturing challenges. Bezos aims to dramatically shorten the invention cycle, citing the example of jet engine design, which typically takes a team of 1,000 engineers over a decade. Prometheus hopes to compress this to five, two, or even one year by simulating all physics and manufacturing processes digitally. This approach will involve building proprietary physics datasets and domain-specific training methodologies, setting it apart from existing general-purpose AI models like those from OpenAI or Google, which primarily rely on internet data.
The startup, with a team of 150 based in San Francisco, London, and Zurich, sees its product as a set of tools that will empower engineers to model, simulate, test, and manufacture physical assets, from microchips to bridges. By moving the iterative and costly trial-and-error process of physical engineering entirely into the digital realm, Prometheus anticipates drastically cutting down costs and time in industries such as aerospace, automotive, advanced manufacturing, data center infrastructure, and drug discovery. The company's unique advantage lies in training its AI on physics-bound workflows, robotics interactions, and empirical laboratory tests to achieve high-fidelity simulations for physical AI applications.
Prometheus is not alone in the physical AI space. Other companies like Antioch are also working on simulation tools for robot developers, aiming to close the "sim-to-real gap" by making virtual environments realistic enough for robot training. Antioch, which recently raised an $8.5 million seed round at a $60 million valuation, focuses on sensor and perception systems for autonomous vehicles, farm machinery, and drones. Physicl, another startup, is developing a data infrastructure layer for physical AI, providing scalable, simulation-ready 3D data and integrating with NVIDIA's Omniverse and Isaac stack.
