Jeff Bezos Raises $10B for Project Prometheus — His Physical AI Lab Valued at $38B with BlackRock and JPMorgan
On April 21, 2026, Jeff Bezos nears a $10 billion funding round for Project Prometheus, his secretive Physical AI lab, at a $38 billion valuation. BlackRock and JPMorgan lead. Launched November 2025 with Vikram Bajaj, Prometheus trains foundation models on real robotics data for aerospace, automotive, pharma and advanced manufacturing. The bet: the next AI frontier isn't text, it's the physical world.
On April 21, 2026, the Financial Times reported that Jeff Bezos is finalizing a $10 billion funding round for Project Prometheus, the Physical AI lab he co-founded in November 2025 with Vikram "Vik" Bajaj. The post-money valuation hits $38 billion. The round is led by BlackRock and JPMorgan — an unusual lineup that signals frontier AI's shift from classic venture capital to large institutional asset managers. Adding the $6.2 billion raised at launch and a planned compounding push toward a $100 billion industrial acquisition holding company, Prometheus becomes the best-capitalized AI entity on the planet after OpenAI and Anthropic — without having shipped a single consumer product.
Physical AI: The Thesis Bezos Is Paying $38 Billion to Defend
The keyword is Physical AI. Prometheus isn't trying to build a better ChatGPT. Its stated ambition, championed by co-CEO Vikram Bajaj (formerly at Verb Surgical and Google X), is to train foundation models on real-world experimental data — robotic interactions, engineering workflows, physics simulations, industrial sensor data. The thesis is straightforward: text-first LLMs have exhausted the public internet, and the next 10x capability jump will come from corpora that Google, OpenAI and Anthropic cannot scrape from the web.
The target domains were specified this week: aerospace (launcher design and propulsion simulation), automotive (EVs and autonomy), advanced manufacturing (reconfigurable factories, collaborative robotics), pharma and drug discovery (molecular simulation and laboratory robots). It's the same thesis Nvidia has been pushing with its Physical AI vision since GTC 2026 — but this time carried by an industrial operator (Bezos, ex-Amazon Robotics, ex-Blue Origin) rather than a compute vendor.
| Axis | Frontier LLMs (OpenAI, Anthropic, Google) | Physical AI (Prometheus, Nvidia, Skild) |
|---|---|---|
| Training data | Web text + code + synthetic | Robotics + sensors + simulations |
| Primary market | Knowledge work, SaaS, coding | Industry, automotive, pharma, defense |
| Dominant modality | Natural language | Multimodal + proprioception + 3D vision |
| Success metric | SWE-bench, MMLU, AIME | Real-world manipulation success rate, MTBF |
| R&D cycles | 6-12 months per release | 2-4 years (hardware + data pipeline) |
Why BlackRock and JPMorgan Lead — Not Sequoia or a16z
The most telling signal of the round isn't the amount but the lineup. BlackRock manages $11.5 trillion in assets; JPMorgan runs the largest U.S. bank balance sheet. Neither is a traditional venture investor. Their presence at the table says three things.
First, $10 billion at a $38 billion valuation is a multiple that Sequoia or a16z can no longer carry alone — even Andreessen Horowitz co-leading the $2 billion Cursor round operates with fund sizes well below this. Frontier lab capitalization has shifted onto the rail of institutional asset managers, exactly like highways, energy infrastructure and telecoms at other eras.
Second, Physical AI requires capital-intensive timelines incompatible with the classic 7-year venture cycle. We're talking robotics labs, simulation chains, industrial acquisitions. BlackRock and JPMorgan can lock up capital over 15-20 years through infrastructure vehicles.
Third, Bezos is explicitly aiming at an industrial acquisition holding company capitalized at $100 billion to absorb factories and feed their operational data back into Prometheus. It's a structure that looks less like a startup and more like an Industrial Conglomerate 2.0 — Berkshire Hathaway on AI steroids.
The Team: OpenAI, xAI, Meta, DeepMind — and the General Agents Acquisition
The Prometheus team is the second pillar of the pitch. The strategic acquisition of General Agents, founded by Sherjil Ozair (former DeepMind researcher), brought a key brick: agents capable of operating simulation and robotics tools without task-by-task fine-tuning. Around that nucleus, Prometheus has recruited senior researchers departing from OpenAI, xAI, Meta Superintelligence Labs and DeepMind. Offers are estimated at $3-8 million annual packages for senior profiles — meaning in the upper range of what Meta Superintelligence Labs has been paying since the Muse launch or what Anthropic offers its top researchers.
Offices are in San Francisco (HQ), London and Zurich. The Zurich choice is strategic: ETH Zurich is one of Europe's top academic centers for robotics and control, and the Swiss ecosystem supplies mechatronics talent hard to poach elsewhere.
What's at Stake on the Frontier AI Map
Project Prometheus enters a landscape where the competition isn't a binary OpenAI vs Anthropic. We're watching a fragmentation into large families of models:
- Text-first frontier: OpenAI with GPT-6/Spud and Anthropic with Claude Mythos
- Open-weight dense: Meta Llama 4 and Google Gemma 4
- Open-weight MoE: Alibaba Qwen 3.6-A3B and DeepSeek V4
- Physical AI / embodied: Project Prometheus, Skild, Physical Intelligence, Covariant
- Specialized vertical: Harvey (legal), Glean (enterprise search), Hippocratic (healthcare)
Prometheus is betting that each family will have its own SOTA, and that the Physical AI family will outpace the others in industrial value-add within 5 years. If the thesis is right, the $38 billion 2026 valuation will look like a low-risk entry in 2030. If it misses, Bezos will have committed $16 billion of personal and institutional cash to a bet that Nvidia and DeepMind were better placed to lead with their compute and data stack.
The Open Questions
No commercial product has been announced. No revenue. No named customer partnerships. Prometheus is raising $38 billion on a thesis — probably the purest "roadmap-over-traction" bet of 2026. It echoes xAI's or Inflection's first year, with two major differences: Bezos is a proven industrial operator (Amazon, AWS, Blue Origin), and the technical team aggregates profiles who have shipped at least one breakthrough before.
Talent risk remains high. Between Meta Superintelligence Labs, Anthropic, OpenAI, xAI, DeepMind and now Prometheus, senior Physical AI researchers number in the hundreds globally. Cost per head exceeds $5 million annually. Financial discipline will be watched closely by institutional investors who don't tolerate classic venture burn rates.
The timeline is uncertain. Bajaj has mentioned "demonstrable progress within 24 months" on generalist robotic manipulation. Without public milestones or shared benchmarks (there's no SWE-bench equivalent for physical manipulation), tracking the progression curve before 2028-2029 will be difficult.
What Prometheus Changes for the Ecosystem
Beyond the round itself, Prometheus's emergence has three structural consequences.
1. Capital available for Physical AI has just doubled in six months. Physical Intelligence raised at $5.6 billion, Skild is near $4 billion, Covariant was acquired by Amazon, and Prometheus parks at $38. European and Israeli robotics teams will accelerate their rounds.
2. The industrial acquisitions market becomes strategic. If Bezos closes the $100 billion holding play, manufacturing mid-caps in Germany, Japan and the US will pass under the control of an AI operator. It's a vertical integration pattern we haven't seen since the cloud consolidation of 2010-2015.
3. Top researcher recruitment becomes brutal. A senior robotics researcher at Google DeepMind now receives weekly offers. 2026 packages will keep rising, which will pressure public labs and universities — a phenomenon already documented in the US and starting to hit Europe via the UK's Callosum sovereign AI fund.
TL;DR:
- Project Prometheus closes $10B at $38B post-money valuation
- Lead investors: BlackRock, JPMorgan (signal: venture → asset managers shift)
- Total since launch (Nov 2025): more than $16B
- Focus: Physical AI via robotics, simulation, industrial data
- Target verticals: aerospace, automotive, pharma, manufacturing
- Team: ex-DeepMind, OpenAI, xAI, Meta + General Agents acquisition
- Holding play: up to $100B to acquire integrable industrial assets
Project Prometheus embodies a sharp thesis: AI over the next 10 years won't be won in text-first datacenters but in laboratories, factories and robotic fleets. Bezos is methodically assembling the capital, talent and data access to push this bet at planetary scale. What remains to be proven is whether a post-Amazon operator can still innovate faster than a scaling-maxed frontier lab — or whether he's right to bet on a different category altogether.
For teams building on top of foundation models, the practical lesson is simple: the multiplication of model families (frontier text, open MoE, Physical AI) accelerates the need for agnostic orchestration layers that route each task to the right model. That's exactly the direction Factory is taking on the coding side, the direction Idlen's chat-ad SDK is taking on the publisher monetization side, and the direction every serious AI product will need to take before end of 2026.
Sources: Bloomberg — Jeff Bezos Nears $10 Billion Funding Round for AI Lab, Tech Funding News — BlackRock and JPMorgan back Bezos' AI lab, The Next Web — Bezos Project Prometheus 10 Billion, PYMNTS — Bezos Targets $38 Billion Valuation for Physical AI Lab.
