TL;DR

The Financial Times reported today that Project Prometheus, the physical-AI lab Jeff Bezos co-runs with Vikram Bajaj, is finalizing a $10 billion round at a $38 billion post-money valuation. JPMorgan and BlackRock are anchoring it. The round began as a $6.2 billion seed last November and got expanded after investor demand outran the original target. The deal is reported as “near close,” not closed. Treat the $38B as a working number.

What the FT Actually Reported

The story broke in the Financial Times this morning, then got picked up across Bloomberg, TechCrunch, The Next Web, and The Decoder. The numbers everyone is reporting:

  • New round size: $10 billion
  • Post-money valuation: $38 billion
  • Lead investors: JPMorgan, BlackRock
  • Original target: $6.2 billion seed (November 2025)
  • Why it grew: investor demand exceeded supply

BlackRock declined to comment when asked. The FT’s framing is that the round is “nearing close”: paperwork essentially done, no direct announcement from Prometheus or the lead investors. We’ve seen mega-rounds at this scale get re-priced in the final week (the Cursor $50B round had similar last-minute chatter), so wait for a press release before booking the number.

What is confirmed by multiple outlets is the lead investor pair. JPMorgan and BlackRock together is a meaningful signal. These are not VC firms running on TVPI math. JPMorgan’s involvement points to either its private bank (placing money for ultra-high-net-worth clients) or its Onyx-adjacent strategic investment desk; BlackRock has been quietly building a private-AI exposure book through its Long-Term Private Capital fund. Both are patient money.

The Valuation Math

Five months from a $6.2B seed to a $38B valuation. To put that on a graph:

DateEventValuationCapital raised
Nov 2025Seed announced~$30B (per launch coverage)$6.2B
Apr 2026Round near close$38B post-money$10B
Implied 5-month change~+27%+$3.8B incremental

The valuation jump is real but smaller than the dollar figure makes it sound. The reported step from $30B to $38B is the kind of mark-up you’d expect for a hot AI lab that has hired aggressively and announced no setbacks. It’s not a Cursor-style 4x or a Cohere-style 2x; it’s roughly a 27% step.

What’s striking isn’t the per-share appreciation. A five-month-old company with no shipped product is now in the same market-cap bracket as Stripe, Databricks, and SpaceX’s most recent secondary tranches. Almost nothing operational sits behind it. What the investors are buying is Bezos, the talent he’s pulled in (more on that below), and a thesis that physical-world data is the next AI frontier.

What the Money Buys

None of the reporting includes a use-of-proceeds breakdown, which is typical for a round that hasn’t formally closed. Reading between the lines of what Prometheus has been doing publicly:

120+
Reported headcount
3
Office locations
$38B
Post-money

The 120+ number comes from FT reporting earlier this year. Three offices: San Francisco (HQ), London, Zurich. Compute spend at this scale is the obvious sink. Physical-world simulation training is data-hungry and GPU-hungry in roughly equal measure, and Prometheus reportedly hired Kyle Kosic (xAI co-founder, ex-OpenAI) specifically to lead infrastructure. That signals build-out, not a cloud-only posture.

The other plausible sink: data acquisition. Bezos has separately floated a $100B holding-company vehicle for buying industrial assets to feed Prometheus’s training data, but that’s a separate fund (or so the reporting goes). The $10B operating round is more likely going into compute, talent, and partnership/licensing deals to get hands on jet-engine telemetry, semiconductor process logs, and similar proprietary streams.

Why a Patient-Capital Stack Is the Tell

The lead investors aren’t VCs, and that’s the most interesting structural feature of the round.

For comparison: the Cursor $50B raise is led by a16z and Thrive Capital, with Nvidia participating. OpenAI’s mega-rounds get led by SoftBank and Microsoft. xAI’s runs are stacked with Andreessen, Sequoia, and Gulf sovereign capital. These are institutional players, but the cap-table math expects a 5-10 year exit horizon.

JPMorgan private bank money and BlackRock LTPC money have a different shape. They’re patient. They’re benchmarked against long-duration bonds and infrastructure assets, not against IRR hurdles. They can sit on a position for 15 years without a partner meeting going sideways.

That matches the Prometheus thesis: physical AI is slow. A model that learns to simulate manufacturing flow doesn’t get useful in 18 months the way a chatbot does. Training requires real-world data that takes years to acquire. Iterating requires deploying models into actual production environments, watching them fail, and revising. Bezos is signaling that he wants investors who won’t pull the ripcord at year three when there’s still no commercial product.

It’s also a signal to the talent pool. Engineers debating whether to leave OpenAI for Prometheus care about whether the company gets squeezed by impatient investors three years in. Anchor investors who don’t need a quick exit reduce that risk.

How This Fits Into Bezos’s Broader Bet

Prometheus is one piece of a larger structure. The other piece, still mostly speculative, is the $100 billion Berkshire-style holding company Bezos has reportedly been raising for. The thesis there: buy old industrial businesses (chip fabs, aerospace suppliers, defense contractors), use their decades of proprietary process data to train Prometheus models, deploy those models back into the businesses to widen margins, recycle the cash flow into more acquisitions.

If both vehicles fund out, Bezos ends up with something nobody else in AI has: a closed loop of capital, data, and operational deployment. OpenAI, Anthropic, and xAI all monetize the same way, by selling API access. Prometheus, in this structure, would own the entire value chain from raw industrial data to AI optimization to the resulting profit stream.

Today’s $10B round doesn’t change that thesis. It just funds the AI-lab leg of it. The harder leg, convincing LPs to back a $100B industrial-acquisition fund tied to an AI lab that hasn’t shipped a product, is still ahead of him.

Risks That Still Apply

Three things to keep in mind before treating this as a done deal.

The round isn’t closed. “Nearing close” is a useful piece of information, but it’s not the same as a press release. The number could move. The investor list could move. We’ve seen rounds at this scale get re-cut in the final 48 hours.

Physical AI hasn’t been validated commercially yet. Prometheus has hired well and raised aggressively, but no benchmark exists for whether their models actually work on industrial workflows. Until there’s a customer reference (even a private one), the $38B valuation is essentially a bet on Bezos and the talent pool.

AI hiring is a leaky bucket. The current AI talent market churns at roughly 18-month average tenure. Prometheus needs people to stick for 5-7 years to build something that works in physical domains. The compensation packages are reportedly top-of-market, but a Meta or OpenAI counter-offer is always one quarter away.

How to Read This If You’re Watching the Sector

A few takeaways for builders, investors, and operators in adjacent spaces:

  • Physical-AI is now a tier-1 fundraising thesis. A year ago, the pitch was “we do robotics with LLMs.” Today, JPMorgan and BlackRock are leading $10B rounds for it. Expect more capital to chase the category in the next 6-12 months.
  • Specialist data is the new moat. Internet text is fully scraped and broadly commoditized. The next valuation premium is going to companies that can credibly source data nobody else has: manufacturing telemetry, drug-discovery wet-lab outputs, defense-grade simulation datasets.
  • Bezos is now actively in the AI-founder pool. He’s not just a Series-A check writer. The co-CEO title is real, and he’s reportedly running this hands-on. If you’re hiring AI talent, you’re now competing with him directly.

FAQ

How much is Project Prometheus raising?

The FT reports the new round is sized at $10 billion at a $38 billion post-money valuation. The round began as a $6.2 billion seed in November 2025 and was expanded after investor demand exceeded the original target.

Who is leading the Prometheus round?

JPMorgan and BlackRock are the two named lead investors. BlackRock declined to comment to the FT, and Prometheus has not issued an official announcement at the time of writing.

What is Project Prometheus building?

Project Prometheus is developing physical-AI models: systems trained on real-world experimental data, robotics interactions, and engineering workflows rather than internet text. The company targets manufacturing, aerospace, drug discovery, and logistics automation.

When did Project Prometheus launch?

Prometheus launched in November 2025 with a reported $6.2 billion seed round, co-founded by Jeff Bezos and physicist Vikram Bajaj. Both hold the co-CEO title; this is Bezos’s first hands-on operational role since stepping down as Amazon CEO in 2021.

Is the $10B round officially closed?

No. As of April 21, 2026, the round is reported as “near close.” Paperwork is largely complete, but no formal announcement has come from Prometheus or its lead investors. Treat the $38B figure as a working number until confirmed.

Sources

Bottom Line

A $10B raise at $38B for a five-month-old company with no shipped product would be absurd if the founder were anyone else. With Bezos, it’s the second-most-interesting AI fundraising story of the year, behind only OpenAI’s $122B run. The patient-capital stack is what tells you what the deal is actually about. JPMorgan and BlackRock are placing a 10-year-plus bet on Bezos repeating, in industrial AI, what he did for online retail. The decade-long wait is the price of admission.