Quick Take
- Company: Project Prometheus — physical AI lab focused on engineering and manufacturing; headquartered San Francisco; offices in London and Zurich; recruiting engineers in India .
- Founded: November 2025 — launched publicly with $6.2 Bn seed funding at ~$30 Bn valuation; currently ~5 months old
- Latest round: $10 Billion — closed April 23, 2026 valuation $38 Billion; no single lead investor; targeted non-Silicon Valley capital: PE firms, sovereign wealth funds, JPMorgan, BlackRock
- Total raised: $16.2 Billion in total funding in under 5 months — among the largest total raises for a company this young in history
- Leadership: Co-CEO Jeff Bezos (Amazon founder); Co-CEO Vikram ‘Vik’ Bajaj (PhD physical chemistry MIT; co-founder Google X’s Wing + Waymo; co-founder Alphabet Verily; co-founder Xaira Therapeutics)
- What it’s building: ‘Physical AI’ — models trained on real-world experimental data, robotics interactions, and manufacturing workflows; NOT text-based LLMs; targets: aerospace, automotive, advanced manufacturing, pharmaceutical drug discovery
- The $100 Bn plan: Separate from this round — Bezos in talks to raise a $100 Bn manufacturing transformation fund to acquire industrial companies (chipmaking, aerospace, defence, automotive) and modernise them with Prometheus AI. These acquired firms become captive Prometheus customers.
- Key talent: 120+ employees from OpenAI, Google DeepMind, xAI, Meta; founding advisors include Ashish Vaswani and Jakob Uszkoreit — two of the eight original authors of ‘Attention Is All You Need’
Project Prometheus, the physical AI startup co-founded by Jeff Bezos and Vikram Bajaj, has closed a $10 billion funding round that values the company at approximately $38 billion, per Bloomberg and Investing.com reporting on April 23, 2026. JPMorgan Chase and BlackRock are among the investors in the round, which had no single lead investor — a deliberate structural choice by Bezos and Bajaj, who targeted institutional capital from private equity firms and sovereign wealth funds rather than conventional Silicon Valley venture capital.
The round, first reported by the Financial Times on April 21, brings Project Prometheus’s total funding to $16.2 billion in under five months — making it one of the most capitalised early-stage companies in history, and one of the fastest companies ever to reach a $38 billion valuation from a standing start.
“[Project Prometheus will be] one of the most important companies in the world.” — Robert Nelsen, ARCH Venture Partners founder, at JPMorgan’s healthcare summit
The Founding Story : A Dinner, an Acquisition, and a Midnight Filing
Project Prometheus was not built the traditional way. Its origin story reads more like a quiet corporate manoeuvre than a startup founding. In June 2025 — months before the company was publicly announced — Vik Bajaj hosted a private dinner at Saison, a Michelin-starred restaurant in San Francisco. Among the attendees: Sherjil Ozair, co-founder of General Agents, a startup that had built Ace — a real-time computer agent capable of executing complex tasks across any computer system in real time.
Corporate filings obtained by Wired show that Bajaj formed an acquisition entity the morning after the dinner. Four days later, General Agents had been absorbed into Project Prometheus. Ozair — formerly of Google DeepMind and Tesla — joined Prometheus. So did his co-founder William Guss, formerly of OpenAI. The acquisition brought General Agents’ video-language-action (VLA) architecture into Prometheus’s technical stack — technology that enables machines to interpret visual data and perform physical actions from language commands, a critical capability for factory floor automation.
| Milestone | Date | Detail |
| Private dinner at Saison | June 2025 | Bajaj meets Sherjil Ozair (General Agents co-founder, ex-DeepMind, ex-Tesla) |
| General Agents acquisition | June 2025 | Acquisition entity filed the morning after the dinner; General Agents absorbed four days later; brings VLA architecture + Ace computer agent into Prometheus |
| Public launch | November 17, 2025 | NYT reveals Project Prometheus; $6.2 Bn seed funding confirmed; $30 Bn initial valuation; 120+ employees |
| Recruiting India team | Late 2025-2026 | Engineers recruited near New Delhi to build accurate 3D models of engine components — a key input for physical AI training data |
| WSJ reports $100 Bn fund | March 2026 | Bezos in discussions to raise a separate $100 Bn manufacturing transformation vehicle to acquire industrial companies |
| FT reports $10 Bn round | April 21, 2026 | Round not yet finalised at time of FT publication; JPMorgan and BlackRock named as investors |
| Round confirmed closed | April 23, 2026 | $10 Bn round at $38 Bn valuation confirmed by Bloomberg/Investing.com; total raised: $16.2 Bn |
The Team Who Built Prometheus
| Person | Role | Background |
| Jeff Bezos | Co-Founder & Co-CEO | Founder of Amazon; net worth ~$200 Bn; previously backed biotech ventures with Bajaj (Grail, Xaira Therapeutics); leads fundraising alongside Bajaj |
| Vikram ‘Vik’ Bajaj | Co-Founder & Co-CEO | PhD physical chemistry, MIT; co-founder and CEO of Foresite Labs (biotech incubator); worked at Google X alongside Sergey Brin on early Wing (drone delivery) and Waymo (self-driving); co-founded Alphabet’s Verily (health sciences); co-founded Xaira Therapeutics (AI drug discovery) |
| Sherjil Ozair | Co-Founder (via General Agents acquisition) | Ex-Google DeepMind, ex-Tesla; co-founded General Agents (built Ace computer agent); VLA architecture specialist; joined Prometheus in June 2025 acquisition |
| William Guss | Co-Founder (via General Agents acquisition) | Ex-OpenAI research scientist; co-founded General Agents; VLA architecture specialist; joined Prometheus in June 2025 acquisition |
| Nal Kalchbrenner | Founding Member | AI researcher; part of the core technical founding team |
| Alex Blocker | Founding Member of Technical Staff | AI/ML engineer; core founding team |
| Stephen Merity | Founding Member of Technical Staff | AI researcher; known in ML community for work on attention mechanisms |
| Ashish Vaswani | Founding Advisor | One of the eight original co-authors of ‘Attention Is All You Need’ (2017) — the paper that introduced the transformer architecture underlying GPT, Claude, Gemini, and all modern LLMs |
| Jakob Uszkoreit | Founding Advisor | One of the eight original co-authors of ‘Attention Is All You Need’ (2017); formerly Google Brain |
| Robert Nelsen | Investor / Evangelist | Founder, ARCH Venture Partners; said publicly at JPMorgan healthcare summit that Prometheus would be ‘one of the most important companies in the world’ |
Having two of the eight authors of ‘Attention Is All You Need’ as founding advisors is the most significant talent signal in Prometheus’s entire team composition. The transformer paper is the foundational intellectual architecture of every major AI system deployed today. Prometheus is not just recruiting AI talent — it is recruiting the people who invented the underlying mathematics of the field.
What Physical AI Actually Is And Why It’s Different
The term ‘physical AI’ is central to understanding why Project Prometheus justifies a $38 billion valuation at five months old. It is not a marketing phrase — it represents a genuinely distinct technical and commercial bet from what the rest of the AI industry is building.
| Dimension | Standard LLMs (GPT, Claude, Gemini) | Physical AI (Prometheus) |
| Training data | Text, images, code scraped from the internet — abundant and cheap | Real-world experimental data, robotics interactions, engineering workflows, 3D component models — scarce and proprietary |
| Learning mechanism | Pattern recognition in digital data; predicts next token | Learns through interaction with physical systems; understands cause-effect in real-world processes |
| Target applications | Writing, coding, answering questions, generating images | Engineering, manufacturing, aerospace, automotive, pharmaceutical drug discovery — physical-world problem solving |
| Competitive moat | Model quality + distribution; incumbents are OpenAI, Anthropic, Google | Proprietary industrial data that cannot be replicated; manufactured by acquiring firms (the $100 Bn fund) |
| Data scarcity | Data is commoditised; diminishing returns on more web text | Every new factory acquired or partnered with generates unique, irreplaceable training data |
| Regulatory exposure | Content moderation, copyright, misinformation | Safety-critical systems; physical failure consequences; rigorous standards required |
| Commercial model | API access; SaaS subscriptions; consumer products | Deep integration into industrial value chains; long-term contracts; potentially embedded in acquired companies |
The VLA (video-language-action) architecture acquired from General Agents is a key technical enabler: it allows AI systems to interpret visual data from cameras and sensors, understand natural language instructions, and translate both into physical actions — making it the bridge between digital AI and factory-floor robots.
The $100 Billion Manufacturing Transformation Plan
Separate from the $10 billion raised by Project Prometheus itself is a plan that, if executed, would make this the most ambitious industrial strategy in modern history. In March 2026, the Wall Street Journal reported that Bezos was in discussions with asset managers and sovereign wealth funds to raise a $100 billion ‘manufacturing transformation vehicle’ — a holding company that would acquire industrial businesses and then use Prometheus’s AI to modernise their operations.
- Target acquisition sectors: Chipmaking, aerospace, defence, and automotive — the industries most exposed to AI-driven disruption and most in need of manufacturing modernisation.
- The closed-loop strategy: Acquired companies provide Prometheus with two things simultaneously: captive customers for its AI tools, and proprietary industrial data that feeds back into Prometheus’s model training. The more factories Bezos owns, the more unique training data Prometheus has, the better its models get, the more valuable the factories become — a compounding flywheel.
- Scale context: $100 billion would make this larger than most PE megafunds and comparable to sovereign wealth funds in deal capacity. Two people briefed on the plan told the FT it would ‘put a few other AI roll-up efforts in the shade, including vehicles launched by Thrive Capital and General Catalyst.’
- India connection: Prometheus is already recruiting engineers near New Delhi to build accurate 3D models of engine components — a precursor to the kind of industrial AI training data pipeline the $100 Bn fund would generate at scale.
StartupFeed Insight
The valuation math that defies convention: $38 billion for a five-month-old company with no publicly demonstrated products and no disclosed commercial revenue. To contextualise: Anthropic — which has actual deployed products, millions of users, and hundreds of millions in ARR — was valued at $61 billion at its most recent round. Prometheus at $38 Bn with nothing public yet is priced at roughly 60% of Anthropic’s valuation. The premium is entirely based on Bezos’s reputation, Bajaj’s scientific credibility, the transformer paper advisors, and the closed-loop industrial data strategy.
The non-Silicon Valley investor strategy is deliberate: Bezos and Bajaj explicitly targeted PE firms, sovereign wealth funds, JPMorgan, and BlackRock rather than traditional VC. This is not a preference — it is a strategy. PE firms and sovereign wealth funds have deep exposure to physical industries and are natural eventual customers for Prometheus technology. The investors are being chosen partly for their commercial relevance, not just their capital.
The transformer advisor signal: Having Ashish Vaswani and Jakob Uszkoreit as founding advisors is not just a credibility signal — it is a technical direction signal. The transformer architecture was originally designed for sequential data processing. Applying it to physical-world interactions (what Prometheus is attempting) is an active area of research. Having two of the architecture’s inventors on the founding advisory team suggests Prometheus is working on fundamental model architecture improvements, not just applying existing LLMs to industrial use cases.
Why India should pay attention: Prometheus is already recruiting engineers in India for 3D component modelling — one of the most technically specific recruiting signals possible. India’s engineering talent pool (aerospace, manufacturing, automotive) is directly relevant to Prometheus’s industrial AI training data pipeline. If the $100 Bn acquisition fund targets Asian manufacturing companies, Indian manufacturing engineers could find themselves on both sides of this wave — as employees building the AI and as workers in the factories the AI is transforming.
The key risk: Physical AI is genuinely harder than LLM development. The scarcity of training data is a feature (moat) but also a bottleneck (slow). Safety-critical applications (aerospace, pharmaceutical manufacturing) require regulatory approvals that can take years. The $38 billion valuation prices in success on a timeline that physical-world deployment cycles may not support. Unlike software, you cannot ship a factory automation system to a billion users overnight.
Our prediction: Project Prometheus will announce its first commercial pilot — likely in aerospace component manufacturing or pharmaceutical process optimisation — by Q4 2026. The $100 Bn acquisition fund will close its first deal (a mid-size US or European industrial manufacturer) by H1 2027. Prometheus will IPO by 2028-29 at a $150-200 Bn valuation, making it one of the most valuable AI companies in the world alongside OpenAI and Anthropic.
The Competitive Landscape What Prometheus Is Up Against
| Competitor | Focus | Backing | vs Prometheus |
| OpenAI | LLMs; agentic AI; GPT series | Microsoft ($13 Bn+); $157 Bn valuation | Generalist — strong on text/code; weaker on physical-world reasoning; primary competition for engineering AI talent |
| xAI (Musk / SpaceX) | Grok models; physical AI via Tesla and Neuralink data | $100 Bn+ in ecosystem | Physical AI competitor via different data sources (vehicle data, neural data) — different industrial targets |
| Google DeepMind | Gemini; robotics; AlphaFold (drug discovery) | Alphabet (internal) | Most direct scientific competitor in drug discovery and robotics; DeepMind has decade-long physical AI research lead |
| Figure AI | Humanoid robotics | VC-backed | Adjacent but narrower — humanoid robots rather than factory AI systems; potential partner or acquisition target |
| Periodic Labs | AI-run research facilities; robotic labs | $300 Mn raised | Drug discovery focus; smaller scale; potential Prometheus partnership or acquisition target |
| Anthropic | Claude; AI safety; enterprise AI | Amazon $4 Bn; Google $2 Bn; $61 Bn valuation | LLM-focused; not targeting physical AI directly; Amazon’s parallel Anthropic investment alongside Bezos’s Prometheus is one of the most fascinating capital allocation stories in AI |
The Bezos-Amazon-Anthropic triangle deserves special mention: Amazon committed up to $25 billion to Anthropic and secured a $100 billion cloud spending pledge in return. Days later, the same week, Bezos personally leads a $10 billion close for Prometheus — a physical AI company that is technically in competition with Anthropic for AI research talent and engineering AI applications. Bezos is betting on both the generalist LLM model (via Amazon/Anthropic) and the specialised physical AI model (via Prometheus) simultaneously. That is either brilliant portfolio diversification or a sign that even Bezos does not know which AI paradigm wins.
Project Prometheus is the most expensive bet on physical AI in history. Whether it justifies a $38 billion valuation at five months old will be answered not by fundraising ability — that is already proven — but by whether AI models trained on factory data can actually improve manufacturing in ways that justify the price of the factories it may need to acquire to generate that training data.
One thing is certain: Jeff Bezos has spent his career betting that the physical world is underserved by technology. He built warehouses before robots. He built rockets before Mars was a commercial proposition. Project Prometheus is the same instinct applied to AI — and this time, he has $16.2 billion and the authors of the transformer paper standing behind him.
