Jeff Bezos's Project Prometheus Nears $10B Raise at $38B Valuation for Physical AI

Five months after its quiet launch, Jeff Bezos’s AI venture is making its most audacious move yet. Project Prometheus, the San Francisco-based startup co-founded and co-led by Bezos himself, is closing in on a $10 billion funding round at a $38 billion valuation — a staggering leap that would make it one of the most valuable AI companies in the world before shipping a single commercial product.

According to a report first published by the Financial Times and confirmed by multiple sources, JPMorgan and BlackRock are anchoring the round, joined by a constellation of institutional investors drawn to a bet that the next frontier of AI is not language or images, but the physical world.

Bezos Back in the Operator’s Seat

Project Prometheus, launched in November 2025, is notable not only for its scope but for what it represents personally for Bezos. It marks the first time the Amazon founder has taken an active operational role at a technology company since stepping down as Amazon’s CEO in 2021. After years of backing ventures from a distance — funding Blue Origin, backing Anthropic, and investing in dozens of climate and robotics startups — Bezos is once again in the building.

His co-CEO is Vikram Bajaj, a former Google X scientist with a PhD in physical chemistry from MIT. Bajaj’s track record at Alphabet is extraordinary: he led early work on both Wing (the drone delivery program) and Waymo (the autonomous vehicle unit), before co-founding Verily, Alphabet’s life sciences arm, and Xaira Therapeutics, an AI drug discovery company. Together, Bezos and Bajaj form an unusually credentialed founding team that has had little difficulty attracting talent.

In just five months, Prometheus has grown to over 120 employees, recruited from leading AI labs including OpenAI, xAI, Meta AI, and Google DeepMind. The company launched with $6.2 billion in its initial capitalization — itself a historic seed-stage investment — and is now nearly doubling that with the new round.

What Is “Physical AI”?

The term “physical AI” is central to Prometheus’s pitch, and it’s worth unpacking precisely what the company means. Traditional large language models and multimodal systems learn from vast corpora of text, images, and video scraped from the internet. They are extraordinarily capable at reasoning, language, and visual recognition. But they have a fundamental limitation: they don’t understand the physical world.

A language model cannot intuit that a material will fracture under stress, or that a drug molecule will bind to a protein receptor, or that a jet engine component will fail after a certain number of thermal cycles — unless that information is explicitly present in its training data. Physical AI, as Prometheus defines it, closes this gap by training models on real-world experimental data, sensor readings, robotics interactions, and engineering workflows. The goal is systems that understand causality in the physical domain, not merely statistical correlations in text.

This distinction matters enormously for industrial applications. Aerospace engineers spend thousands of hours running physical and computational experiments to validate designs. Pharmaceutical researchers conduct wet-lab experiments that cost millions of dollars and take years. Advanced manufacturers rely on tacit knowledge embedded in human machinists who understand materials by touch and experience.

Prometheus is building AI that can absorb this knowledge and accelerate it by orders of magnitude.

Target Industries and Go-to-Market Strategy

The company has identified four primary verticals where physical AI can deliver transformative ROI: aerospace, automotive and advanced manufacturing, drug discovery, and logistics automation. These are not coincidentally the four industries most defined by expensive, slow, hard-to-scale physical experimentation.

In aerospace, the company is targeting everything from airframe design validation to engine materials analysis — processes that currently require multi-year testing programs. In pharma, it sees an opportunity to replace early-stage wet-lab experiments with AI-simulated results accurate enough to guide clinical development decisions. In manufacturing, it aims to automate the kind of engineering judgment that currently requires a senior machinist with decades of experience.

The go-to-market strategy is partially revealed by a secondary initiative Prometheus is reportedly pursuing: a separate investment holding company seeking to raise up to $100 billion to acquire majority or minority stakes in architecture, engineering, and construction firms. The strategic logic is data flywheel: acquire companies that generate physical-world engineering data, feed that data back into Prometheus’s AI models, and use the improved models to make the portfolio companies more productive. It’s a vertically integrated AI play that would make OpenAI’s enterprise strategy look modest by comparison.

Valuation in Context

At $38 billion, Prometheus’s valuation demands scrutiny. The company has no revenue, no commercial product, and has been operating for less than six months. Even by the exuberant standards of 2026’s AI funding landscape — where Anthropic is approaching a $200 billion valuation and OpenAI is preparing an IPO at over $1 trillion — the Prometheus number is striking.

Investors appear to be pricing in three things: the Bezos brand and his operational involvement, Bajaj’s unique combination of scientific depth and operational track record, and the sheer scale of the industrial markets they are targeting. Global aerospace manufacturing alone represents a $350 billion annual industry. Advanced drug discovery, from early research to clinical trials, represents hundreds of billions more.

The concentration of risk is real. Physical AI is significantly harder to build than text AI — the training data is expensive and proprietary, benchmarks are nascent, and the path from research prototype to production-grade industrial tool is long. But the upside, if Prometheus can deliver, is correspondingly enormous.

The Bezos Effect on AI

Bezos’s entry into the AI race as an operator rather than an investor changes the competitive dynamics in ways that are only beginning to be understood. OpenAI has Sam Altman. Anthropic has Dario Amodei. Google DeepMind has Demis Hassabis. xAI has Elon Musk. Now the AI lab landscape has Jeff Bezos — a figure with unmatched experience scaling technology companies and building physical-world logistics infrastructure.

The $10 billion round, if it closes as reported, will not merely validate Prometheus as a company. It will mark the arrival of physical AI as an investable category in its own right, distinct from the language model wars that have dominated the past three years. For industries that still depend fundamentally on human expertise interacting with the real world, that distinction could be the most consequential development in enterprise AI since the transformer architecture itself.

The round is expected to close before the end of Q2 2026.