Bezos's Project Prometheus Nears $10 Billion Raise at $38 Billion Valuation as Physical AI Lab Recruits From OpenAI, xAI, Meta and DeepMind
JPMorgan and BlackRock are backing Jeff Bezos's physical AI startup Project Prometheus in a $10 billion round at a $38 billion valuation, his first operational role since leaving Amazon in 2021.
Overview
Jeff Bezos’s stealth artificial intelligence venture, code-named Project Prometheus, is close to finalizing a $10 billion funding round at a $38 billion valuation, with JPMorgan and BlackRock among the backers, according to Bloomberg, citing a Financial Times report. The round, which has not yet closed, would roughly triple the amount of capital the San Francisco company has raised since it emerged in November 2025 with $6.2 billion in initial backing.
The venture is Bezos’s first operational technology role since he stepped down as Amazon chief executive in 2021, as noted by The Decoder. He co-leads the company with Vikram Bajaj, a former Google X scientist and co-founder of Foresite Labs.
What We Know
Project Prometheus is pursuing what the industry has begun calling “physical AI” — models trained to understand the behavior of materials, engineering tolerances and industrial processes, rather than the text and image corpora that underpin large language models. According to The Next Web, the company is targeting applications across engineering, manufacturing, aerospace, robotics, drug discovery and logistics automation.
The startup has scaled rapidly since its launch. The Next Web reports that Prometheus now employs more than 120 people, with hires drawn from OpenAI, xAI, Meta and Google DeepMind. The company operates out of San Francisco with additional offices in London and Zurich, The Decoder reports.
Bloomberg notes that the $38 billion valuation places Prometheus among the most richly capitalized AI startups outside the established frontier-model developers, despite the fact that it has not yet publicly demonstrated a product or disclosed a commercial deployment. The Next Web describes the company as focused on systems that “learn through interaction with the real world” — a positioning that sets it apart from the text-centric roadmaps of OpenAI, Anthropic and Google DeepMind.
The fundraise sits alongside a separate, larger ambition. The Decoder reports that Bezos is discussing a holding company that could raise up to $100 billion to acquire industrial businesses, whose operational data would in turn feed Prometheus’s models. That structure, if realized, would give the AI lab a proprietary training corpus of real-world industrial telemetry that competitors would struggle to replicate from public data alone.
What We Don’t Know
The investor list beyond JPMorgan and BlackRock has not been disclosed, and neither the company nor the lead investors have publicly confirmed the round — all figures currently trace back to reporting by the Financial Times, relayed by Bloomberg and other outlets. The precise closing date, final deal terms, and whether the round is primary capital or includes secondary sales remain unreported.
Project Prometheus has not described the architecture of its models, whether they are built on transformer foundations, diffusion methods, or a bespoke stack for physical simulation. No technical papers, benchmarks or customer pilots have been made public. It is also unclear when the company intends to begin generating revenue, or how it plans to navigate the regulatory landscape for AI systems deployed in safety-critical industrial settings such as aerospace and drug discovery.
The scope of Bezos’s personal operational involvement — whether full-time, part-time, or confined to strategy and fundraising — has not been quantified by any of the reporting outlets.
Context
Prometheus arrives during a period in which the largest AI funding rounds have concentrated in a handful of model developers, as previously reported by The Machine Herald, with three companies absorbing the majority of venture capital inflows. A $38 billion valuation for a pre-product startup, anchored by two of the world’s largest asset managers, suggests that investors are willing to extend that concentration into adjacent categories — in this case, the still-speculative bet that AI systems can generalize from language to the physics of factories, drug pipelines and aerospace manufacturing.
Bezos’s industrial-holding thesis, if funded at the scale reported, would represent a departure from the conventional venture model of betting on software alone. It assumes that the durable moat in physical AI is not algorithmic but data-driven, and that the shortest path to that data runs through ownership of the factories, logistics networks and engineering firms that generate it.