It is five months old, has released no products, published no research and generated no revenue. Yet Jeff Bezos’s AI startup, code-named Project Prometheus, is close to finalising a $10 billion funding round that values the company at $38 billion. A figure that puts it in rarefied company among the world’s most highly-valued private technology ventures. The question worth asking is not whether that valuation is justified, but why this particular bet is attracting serious institutional capital at such speed.

The team behind the valuation

Project Prometheus is building AI systems that understand the laws of physics and can be applied across engineering, manufacturing and other physical industries. Launched in November 2025 with $6.2 billion in funding, it has quickly scaled to over 120 employees drawn from leading AI firms, including OpenAI, xAI, Meta and DeepMind.

Bezos serves as co-chief executive alongside Vikram Bajaj, a chemist and physicist formerly of Google X who is also the co-founder and CEO of AI incubator Foresite Labs. The company is headquartered in San Francisco with offices in London and Zurich.

Prometheus has recruited staff spanning large language models, computer vision, weather forecasting and chip optimisation for physics simulations. It also acquired the startup General Agents, the firm behind an autonomous web-browsing tool called Ace, folding its founders directly into the team.

The move back into an operational role is significant in itself. This marks the first time Bezos has held an operational role since leaving Amazon in 2021, a signal that he views this not as a passive investment but as the defining technology challenge of the next decade.

Robert Nelsen, co-founder of ARCH Venture Partners and a director at Prometheus, has been characteristically direct about the ambition: “Figuring out how to reinvent the physical world is a big challenge. The pace of innovation in AI right now is truly hard to understate.”

Physics, not prompts

What distinguishes Prometheus from the broader AI funding frenzy is its underlying technical thesis. Rather than building systems that learn purely from massive amounts of digital data, as large language models are designed to do, Prometheus intends to apply AI to physical tasks, requiring systems that can learn from real-world trial and error. Its target sectors span manufacturing, aerospace, robotics, drug discovery and logistics automation.

The company is targeting industries such as jet engines and semiconductor production, aiming to make them faster and less resource-intensive. These are not niche applications; they sit at the heart of the global supply chains that the industry depends on for hardware procurement, energy infrastructure and industrial components.

A manufacturing transformation vehicle

Beyond the AI lab itself, Prometheus is building something more ambitious still. The company is planning a separate holding structure described as a “manufacturing transformation vehicle”. An entity that would have tens of billions of dollars to acquire businesses that Bezos and his team believe will be reshaped by AI. The strategic logic is straightforward: the holding company would feed data from those portfolio businesses back into Prometheus’s AI models, creating a self-reinforcing competitive advantage as the AI system learns from real operational environments.

The fund would target companies struggling with labour costs, production backlogs, or the capital demands of retooling for new technologies, with specific focus areas including chipmaking, defence and aerospace.

The investor profile of the new funding round reflects the scale of that ambition. JPMorgan and BlackRock are among the investors in the new round, although it has not yet been finalised.

A $100 billion fund would place Bezos in territory previously occupied only by the world’s largest sovereign wealth funds and private equity giants. The comparison to SoftBank’s Vision Fund, the only private vehicle to have previously raised capital at that scale, is instructive, though the underlying thesis here is more specific and arguably more grounded in genuine technical differentiation.