Jeff Bezos is not building another chatbot. He is raising $100 billion to buy old factories and rebuild them with AI.
The Wall Street Journal broke the story on March 19, 2026, reporting that Bezos was in early discussions with the world's largest asset managers and sovereign wealth funds to create what investor documents describe as a "manufacturing transformation vehicle." The fund would acquire traditional industrial companies in sectors including chipmaking, aerospace, and defense, then use artificial intelligence to rebuild how they operate from the inside.
This is not a standalone investment thesis. It is the acquisition engine for Project Prometheus, a stealth AI startup Bezos co-founded in November 2025 that is building what it calls AI for the physical economy. Bezos has been traveling to Singapore and the Middle East to pitch the capital raise personally, and when the world's second-richest person courts sovereign wealth funds in person, the opportunity tends to be structural rather than speculative.
What Project Prometheus Actually Is
Project Prometheus launched in November 2025 with $6.2 billion in initial funding, valuing the company at $30 billion at inception. A portion of that capital came from Bezos personally. This marks the first time Bezos has taken an official operational role in a company since leaving Amazon.
Bezos serves as co-founder and co-CEO alongside Vikram "Vik" Bajaj, a physicist and chemist who spent years at Google X working alongside Sergey Brin on unconventional research programs, including the early work that became Wing (drone delivery) and Waymo (self-driving vehicles). Bajaj also co-founded Alphabet's health sciences company Verily. The two founders were connected through Bezos's investments in Bajaj's biotech ventures, including Grail and Xaira Therapeutics.
As of December 2025, the company had hired over 120 employees, including researchers recruited from OpenAI, Google DeepMind, and Meta. Among the founding advisors are Ashish Vaswani and Jakob Uszkoreit, two of the original co-authors of "Attention Is All You Need," the 2017 paper that introduced the transformer architecture underlying modern AI.
The company is headquartered in San Francisco, with offices in London and Zurich. It operates with no public website, a bare-bones LinkedIn page that says only "AI for the physical economy," and has given no public statements about its technology.
The General Agents Acquisition
In June 2025, before Prometheus's existence was publicly known, Bajaj hosted a private dinner at Saison, a Michelin-starred restaurant in San Francisco. Among the attendees was Sherjil Ozair, co-founder of General Agents, a startup that had built a computer agent called Ace 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, who had previously worked at Google DeepMind and Tesla, joined Prometheus alongside several colleagues. His co-founder, William Guss, had previously worked at OpenAI.
The acquisition brought General Agents' video-language-action (VLA) architecture into Prometheus's stack. While Ace applied that architecture to digital tasks, VLA models are equally critical for physical robotics, enabling machines to interpret visual data and perform physical actions based on language commands. This alignment with Prometheus's manufacturing focus was evidently the point.
The $100 Billion Fund: Structure and Strategy
The $100 billion fundraising target is separate from Project Prometheus's own capital raise, though the two are designed to work together.
Prometheus is building the AI technology. The fund is the acquisition vehicle that will own the companies where that technology gets deployed. Instead of selling software to skeptical factory owners, Bezos would own the factories outright and apply Prometheus's models directly. This removes the sales cycle, eliminates the integration negotiation, and provides a controlled environment in which the AI can actually be trained on real operational data at scale.
Investor presentations cited by media describe the fund's goal as transforming "legacy manufacturers into AI-first companies that can design, simulate and operate complex systems far more efficiently than today."
At $100 billion, the fund would rival SoftBank's Vision Fund as the largest private capital vehicle ever assembled. SoftBank's fund relied heavily on Middle Eastern sovereign capital. Bezos is making the same pitch to the same pools of money, this time in person during recent trips to Saudi Arabia, the UAE, and Singapore.
Target Sectors
The fund is not focused on consumer goods or light manufacturing. Investor documents identify three primary target sectors:
- Chipmaking and semiconductors: where design cycle times, yield optimization, and supply chain complexity create exactly the kind of multi-variable engineering problem that AI can compress
- Aerospace and defense: where production timelines for complex components are notoriously long and simulation-based design has been the state of the art for decades
- Automotive manufacturing: where established players face compressing margins, electrification pressures, and mounting competition from AI-native Chinese manufacturers
These are not sectors with thin operating margins that private equity has avoided. They are sectors with strategic importance to national governments, which makes them interesting to sovereign wealth funds for reasons that extend beyond financial return.
What "Physical AI" Means at Prometheus
The term physical AI is doing a specific amount of work in how Project Prometheus describes itself, and it is worth unpacking.
Most AI systems that dominate headlines, including large language models, image generators, and coding assistants, are trained on digital data and operate in digital environments. Their outputs are text, images, and code, and they do not interact with the material world.
Physical AI is designed to model and interact with real-world systems: factory floors, supply chains, machines under mechanical stress, aerospace components in operation.
The core technology Prometheus is reportedly developing centers on digital twins, AI systems that build accurate simulations of how a physical environment operates and predict how changes will play out before any physical change is made.
The Axios report on the fund was careful to distinguish what this is and is not: Prometheus is focused on AI that optimizes pre-production machinery and processes, including prototyping and design simulation. It is not primarily focused on assembly robotics or displacing workers on a production line, a distinction that matters when pitching to governments simultaneously interested in industrial efficiency and anxious about manufacturing employment.
Physical AI also requires something that LLM training does not: real-world operational data from physical systems, which cannot be scraped from the internet. By acquiring the companies themselves, Prometheus gains direct access to the sensor data, production logs, materials science data, and failure records these manufacturers have been accumulating for decades. The manufacturers own the data pipelines, and the fund is buying those pipelines along with the factories.
Why This Is a Different Bet Than the Rest of AI
The AI industry in 2026 is running on a few dominant investment theses. Frontier model labs believe the value accumulates in whoever builds the most capable model. The hyperscalers believe it accumulates in whoever owns the compute. Consumer AI companies believe it accumulates in whoever captures attention and distribution.
Bezos is making a different argument: that the largest untouched opportunity in AI is the physical economy, specifically the $16 trillion global manufacturing sector, most of which runs on processes designed in the 1990s or earlier.
The hyperscalers, Amazon, Microsoft, Google, and Meta, are collectively projected to spend close to $700 billion on AI infrastructure in 2026. Amazon alone is committing $200 billion. Virtually all of that capital flows into data centers, GPUs, and cloud infrastructure. Almost none of it flows into factories.
Venture capital follows a similar pattern, funding chatbots, coding assistants, and software companies. Until now, no one was writing $100 billion checks to transform how things actually get made.
Robert Nelsen, co-founder of ARCH Venture Partners and a director at Project Prometheus, put it directly at an event in January 2026: "Figuring out how to reinvent the physical world is a big challenge. But the pace of innovation in AI right now is truly hard to understate."
The Bezos Credential
The skeptical read on Project Prometheus is that it is another well-funded AI startup with an ambitious thesis and no proven technology. The credentialed read is more compelling.
Bezos built Amazon's fulfillment network into one of the most sophisticated AI-augmented logistics operations on the planet. Kiva robots, computer vision inventory systems, route optimization algorithms, and predictive demand models are all running at scale inside Amazon warehouses right now. If anyone has demonstrated that AI can be pushed into physical operations at scale, it is the executive who built that system from a blank sheet of paper.
The structural parallel between what Prometheus is attempting and what Amazon did to retail and logistics is not accidental. Bezos identified a massive, inefficient industry, built the infrastructure to serve it, and then expanded relentlessly.
He did it with retail. He did it with cloud computing through AWS. He did it with logistics. The pattern is the same. The industry this time is manufacturing.
The Risks
Raising $100 billion and transforming an entire industrial sector carries specific risks that the thesis does not automatically overcome.
- Capital raise difficulty. SoftBank's Vision Fund is the only private vehicle to ever assemble capital at this scale, and it was heavily reliant on Middle Eastern sovereign funds that may be more cautious with their deployment right now. Bezos's fundraising trips to the region are a direct acknowledgment of who the critical limited partners would need to be.
- Operational complexity. Acquiring and transforming legacy manufacturers is not a software deployment. It requires multi-year, high-touch operational engagement with companies whose cultures, workforces, and physical infrastructure were built for a different era. Private equity firms have attempted industrial transformation plays at much smaller scale and failed.
- Technology maturity. Prometheus's AI models are still in development. The fund's success depends on that technology maturing at a pace that outpaces the depreciation of the acquired assets and the patience of limited partners expecting returns.
- Competition. Blackstone, KKR, and other large private equity firms have been circling manufacturing assets for years and are increasingly touting their own operational improvement capabilities. Andreessen Horowitz and other tech-forward investors have funded industrial AI startups that promise modernization without requiring full company acquisitions.
- The data advantage is not guaranteed. Buying factories buys access to operational data, but extracting value from that data requires building AI systems that can actually use it, which is the hard part that the fund's capital does not automatically solve.
Conclusion
Project Prometheus is a direct bet that the most important AI opportunity of the next decade is not in the model layer, not in the infrastructure layer, and not in the consumer application layer. It is in the physical economy, specifically in the gap between what AI can do and what manufacturing companies have been able to adopt.
The $100 billion fund is the mechanism for closing that gap by acquisition rather than persuasion: own the factories, deploy the AI, and capture the efficiency gains directly rather than waiting for skeptical factory owners to buy software subscriptions.
Whether Bezos can raise the capital, build the technology, and execute the transformation at the scale he is describing is an open question. The ambition is undeniable. The structural logic is sound. And the person making the bet has a credible track record of identifying physical industries that looked resistant to technology disruption and proving that assessment wrong.
Frequently Asked Questions
What is Project Prometheus?
Project Prometheus is an AI startup co-founded by Jeff Bezos and Vikram Bajaj in November 2025. It launched with $6.2 billion in funding, valuing it at $30 billion at inception, and is building AI systems for manufacturing and engineering. The company focuses on what it calls "physical AI," meaning systems that model and interact with real-world industrial processes rather than generating text or images. It has over 120 employees recruited from OpenAI, Google DeepMind, and Meta.
What is the $100 billion fund Bezos is raising?
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." This fund would acquire companies in chipmaking, aerospace, defense, and automotive sectors, then use Project Prometheus's AI technology to modernize how they operate. The fund is separate from Prometheus's own capital but is designed to provide Prometheus with captive customers for its technology.
Why is Bezos targeting manufacturing specifically?
Manufacturing is a $16 trillion sector of the global economy that has been largely bypassed by the AI investment wave of 2024 to 2026. Most AI capital flows into data centers, software, and consumer applications. Bezos's thesis is that physical manufacturing represents the largest untapped AI opportunity, and that the only way to deploy AI effectively into legacy manufacturers is to own them outright rather than try to sell software to resistant incumbents.
What is "physical AI" and how is it different from regular AI?
Physical AI refers to models trained on and designed to interact with real-world physical systems rather than digital data. Project Prometheus focuses on digital twins, which are AI systems that simulate factory floors, machinery, supply chains, and engineering processes to predict how physical changes will perform before they are made. This requires operational data from physical systems that cannot be scraped from the internet, which is part of why owning the manufacturers is central to the strategy.
What is the General Agents acquisition and why does it matter?
In June 2025, Project Prometheus acquired General Agents, a startup that built Ace, a computer agent capable of executing complex tasks across computer systems in real time. The founders, Sherjil Ozair from Google DeepMind and William Guss from OpenAI, joined Prometheus. The strategic value was General Agents' video-language-action architecture, which enables machines to interpret visual data and perform physical actions based on language commands, a technology that applies directly to factory floor automation.
What are the main risks of the Project Prometheus plan?
Key risks include the difficulty of raising $100 billion in capital at a time when sovereign wealth funds are more cautious, the operational complexity of transforming legacy manufacturers over multi-year timelines, the fact that Prometheus's AI technology is still in development, competition from established private equity firms with industrial relationships, and the challenge of actually extracting value from the operational data that comes with acquired factories.
Related Articles
Dani
Dani
Dani
