Jeff Bezos is on the cusp of sealing one of the most eye-watering early-stage fundraisings the artificial intelligence sector has yet produced, with his nascent physical AI laboratory, Project Prometheus, reportedly closing in on a $10bn (£7.9bn) round that would value the venture at $38bn.
The Financial Times, citing people familiar with the matter, reported on Monday that BlackRock and JPMorgan are among the institutional heavyweights that have signed up to the round, though the transaction has yet to be finalised. BlackRock declined to comment. The fundraising, if completed at the mooted terms, would place Prometheus among the most richly valued early-stage AI businesses on the planet, less than six months after it emerged from stealth.
Launched quietly in November 2025 with $6.2bn of initial backing, Prometheus is chasing a very different thesis to the generative AI giants that have dominated the investment cycle since ChatGPT arrived in late 2022. Rather than training ever-larger language models on the internet’s text and imagery, it is building systems that can reason about the physical world itself, materials, tolerances, processes and the immutable laws of physics. The stated target markets are engineering, manufacturing, aerospace, robotics, drug discovery and logistics automation, sectors where large language models have, so far, made only glancing contact.
Running the show on a day-to-day basis is chief executive Vikram Bajaj, a former Google X scientist and co-founder of Foresite Labs. The lab has swelled to more than 120 staff, poached from the likes of OpenAI, xAI, Meta and DeepMind. Bezos, described as one of the initial backers, has been leading the fundraising alongside Bajaj, and, notably, has taken an operational role in the business. It is the first time the Amazon founder has rolled up his sleeves at a technology company since stepping down from the chief executive’s chair at the group he built in 2021.
The timing is striking. Prometheus’s raise is landing only days after Amazon itself committed up to $25bn of fresh investment in Anthropic, securing in return a $100bn cloud-spending pledge from the Claude-maker, a transaction that underlined quite how dramatically the scale of AI infrastructure deals has shifted. A $10bn round for a six-month-old laboratory would, for perspective, exceed the lifetime fundraising of most AI companies in existence.
Why are institutions the size of BlackRock and JPMorgan prepared to write cheques of that magnitude into an unproven venture? The answer lies in the peculiar economics of physical AI. Unlike the vast quantities of cheap, publicly available text and code that power today’s language models, the data needed to teach a machine how steel fatigues, how a drug molecule binds or how a robotic arm should pick a part is proprietary, scarce and devilishly expensive to gather at scale. That scarcity is itself a moat, and accumulating it early may confer a durable advantage on whichever laboratories manage it first.
For Britain’s small and mid-sized manufacturers, aerospace suppliers and life sciences specialists, many of whom already sit on decades of unique operational data, the emergence of a well-capitalised Bezos-backed laboratory is a development worth watching. If Prometheus delivers on its ambitions, the model for applying AI to the industrial economy will not be built on the back of scraped web pages but on partnerships with the firms that actually make, mend and move things.
That, of course, is a sizeable “if”. Prometheus has yet to publicly demonstrate a product, let alone a commercial deployment, and the lab remains firmly in its early phase. Plenty of sceptics will also point out that the broader AI market is wearing increasingly frothy valuations. Peter Fedoročko, chief technology officer at analytics firm GoodData, takes a measured view. “Yes, AI has a bubble, but the technology is real,” he argues. “When dot-com crashed, the internet didn’t disappear, it became infrastructure. The same thing happens here. The dot-com crash took a decade to recover financially, but the internet reshaped everything during that time. It didn’t wipe out jobs; it transformed them. AI follows the same pattern. Once the hype burns off, the real builders get back to work.”
For Bezos, the calculation is simpler. Having built the world’s largest logistics and cloud empire on the back of an earlier technological wave, he is now betting, in person and in size, that the next one will be written not in pixels and prose, but in physics.
Read more:
Bezos’s physical AI lab Prometheus nears $10bn raise at $38bn valuation