Bezos’s $100B Plan for Project Prometheus is Finally Clear

0:00 Jeff Bezos is back. After a near 5-year hiatus, he's trying to tackle one single problem. The global manufacturing market is $16.8 trillion, but has less than 1% AI penetration. This gap is why Bezos is stepping back into the CEO role for his new company, Project Prometheus, and raising a reported $100 billion to support it. His thesis is that the next big AI opportunity won't be coming from chatbots, but will come from the physical economy. Products, manufacturing, supply chains rebuilt with AI. Despite all the hype around this company, I didn't find any clear coverage into what they're actually building. So, in today's video, I'm going to help you understand Prometheus, what they're building, and why the biggest names in finance are investing billions into what they believe will be the next major AI breakthrough. Before we start, know that this video is for educational purposes only. I'm not a financial advisor, investing is risky, and you must do your own due diligence.

0:48 >> >> So, what pulled one of the most successful founders in history back into the CEO role? According to Jeff, the more he saw what Prometheus was building, the less he could stay on the sidelines. In his own words, he became so impressed by the potential that he had to jump in with both feet. So, in 2025, Bezos launched Prometheus with his co-founder, Vic Bajaj, an MIT-trained physicist and chemist who co-founded Google's Verily and built his career at the intersection of hard science and data. Bezos also reportedly helped fund the company's $6.2 billion launch. Around them, they have an all-star roster, growing the company to 150 people, >> >> and hiring researchers away from the biggest names in tech, such as OpenAI, Google's DeepMind, and Meta. Bezos also put David Limp, the CEO of Blue Origin, on board, adding serious aerospace and hardware credibility. This all goes to show that this isn't a lean experiment run out of curiosity, it's an all-in bet assembled by someone who has built world-changing companies before and clearly believes he's about to do it again. So, what did Bezos see that was so promising? Bajaj put it plainly, saying the pace of our physical creation right now is nowhere near the pace of human imagination. Every product that never got made because the engineering was too hard, too slow, or too expensive, that's the backlog. Better products, cheaper chips, new medicines that were once considered too complex to design. That gap between what we can imagine and what we can actually build is exactly what Prometheus is built to close.

2:05 Now, let's get into what Prometheus is actually doing. Everyone assumed Bezos plus AI meant robots, but Bezos said flatly that the company has nothing to do with robotics. What Prometheus is building is something Bezos calls an artificial general engineer. Here's what that actually means. Think about how any physical product gets made. A car part, a phone, new drugs. Before a factory can build anything, an engineer has to design it on a computer. That entire process, design, test, prototype, manufacture is slow, expensive, and takes an enormous amount of skill. Jet engines are a perfect example of this. Building a jet engine isn't one engineering challenge, it's thousands all connected. A single turbine blade must spin thousands of times per minute inside an engine that's temperature would normally melt it for tens of thousands of hours without breaking. You can't simply design that and ship it. You need to build it, then test it until it breaks, then figure out the reason why it broke, redesign it, and run the whole process again. Every pass through that cycle can cost millions of dollars and months of time.

2:56 This is the example that Bezos keeps coming back to. He's pointed out that even an incremental change to a jet engine, a customer who just wants to add 10% more thrust can turn into a decade of work. Not because the engineers aren't smart, but because a jet engine is so complex that changing one thing changes a thousand other things. Bezos says they want to treat the entire process from the idea to finished product as one single AI problem. You describe what you want and the AI designs it and figures out how to manufacture it. In other words, they're building tools to make that entire process 10 times faster. And if it works, think about what that unlocks.

3:26 Right now, every idea that requires engineering carries a hidden price tag. How many times are you willing to fail, test, and retry before your product is ready? If the answer is 50 times of dollars a pass, that's a price almost no one is willing to pay for. This means that most ideas never get attempted at all. They die in a meeting or in someone's mind just because the economics don't make sense. But if Bezos is right and Prometheus can actually speed up the entire process by 10x, then suddenly every idea that was considered economically unviable now becomes a possibility. Multiply that across every stalled project, and the gain isn't just that we start building things faster, is that we start building the solutions that we thought were too difficult and unrealistic to solve in the first place.

4:07 So, now that we understand the problem that Prometheus is aiming to solve, we have to talk about what they're building because it's not a chatbot. It's what AI researchers call a world model. A large language model predicts words, but a world model tries to understand physics, how objects move, how materials behave under stress, how the real world actually works. And if you could feed a model enough real-world data, then in theory, it can start designing physical objects the way a large language model writes sentences. You describe what you need, and it gives you a design that works. The theory makes sense, but the problem is getting enough real-world data. Prometheus wants to be the ChatGPT of design and manufacturing, but ChatGPT, Claude, Gemini, those models all had one enormous advantage. They were trained on the internet. Decades of human writing, code, and images sitting on public servers essentially free for the taking. Nobody had to go out and create the training data. It was already there. But physical engineering has no internet. There's no giant public library of how titanium behaves at 3,000°, no open data set of how a turbine blade cracks under pressure.

5:02 That knowledge is locked away inside private company databases and expensive testing labs that guard their results carefully. It's some of the most valuable proprietary information on Earth, and none of it is on public servers. So, right now, the core challenge for Prometheus isn't their AI, it's collecting the right data. Prometheus hasn't fully spelled out how they're going to solve this, so some of what follows is informed reasoning, but the logic points in three directions. Option one is simulation. Engineers already have software that simulates physics on a computer. So, Prometheus could run millions of these simulations and generate practice data out of nothing. Think of it like a flight simulator. A pilot can log thousands of hours without ever leaving the ground, but there is still a reality gap. An AI model can work perfectly in a simulation and still fail in a real factory because simulations miss things like humidity and worn-out machines and inconsistent materials. So, simulations can act as a starting point but not as a final source. That brings us to option two, partnerships. So, the next move is to partner with real manufacturers, data in exchange for early access to software. For now, that works. Live results a simulation can't fake. The failed batch, the machine that dies for a reason that nobody saw coming.

6:08 But, it's a rental, not a foundation. A partner only shares what it wants to, never its worst failures, which is the data that's actually worth the most. This is a good place to start but not something Prometheus can build its empire on, which leads us to one last option, ownership. And this is where things get very interesting for Prometheus and also where that hundred billion-dollar fund comes into play. Think about what a factory is from an AI company's perspective. It's a data set, decades of information, every part that came off the line, every batch that failed inspection, every material that behaved differently in the real world than the spec sheet promised. That data is not for sale, but the company that owns it is. According to the Financial Times and later the Wall Street Journal, Bezos has been raising money for a separate vehicle, a holding company whose entire purpose is to buy manufacturing businesses outright. The Journal reported the target raised at roughly 100 billion dollars, but nothing's been confirmed around that number. To be clear, that isn't money for Prometheus, they already have their funding. This is a second pool, reportedly far larger than the company itself. An investor document seen by both outlets describe it as a manufacturing transformation vehicle.

7:08 Now, Bezos declined to discuss the fund directly, but what he did say is that Prometheus may buy parts of companies that could benefit from its technology and help them improve how they manufacture. Now, put the two halves together. Prometheus builds the AI, the fund buys old industrial companies, specifically the ones most exposed to being disrupted by that AI. Those companies come with decades of proprietary manufacturing data attached. And now, the loop closes. The AI trains on that data and gets better. A better AI makes those same factories faster and cheaper to run, which makes the factories worth far more than Bezos originally paid for them. That's the business model.

7:43 You've heard the plan, now look at who's funding it. Prometheus just raised $12 billion in a single round, backed by names like JP Morgan, Goldman Sachs, BlackRock, DST Global, and Arch Venture Partners. Plus, Bezos himself. For a company that hasn't shipped a single product, raising this much capital is highly unusual. But what's even stranger is who wrote the checks. Arch Ventures makes sense. Its co-founder already sits on the Prometheus board. But take a look at the others. The largest money manager on Earth and some of the biggest banks in the world, all backing a startup that doesn't sell anything yet at a $41 billion valuation. Firms like these usually want to see revenue, customers, real numbers to justify that price. Prometheus has none of that. So why are these companies lining up behind them? Start with BlackRock. In 2024, they struck a deal worth about $25 billion combined, buying an owner of physical infrastructure and one of the largest private lenders in the world. So they are now a giant lender, too. And you don't buy $100 billion of factories with cash, you buy them with debt. So BlackRock is exactly the kind of firm you'd want if your plan is to buy an industry, but it's not the one that opens up the door to decision makers. That's JP Morgan. In late 2025, they launched something called the Security and Resiliency Initiative, a one and a half trillion-dollar 10-year plan to rebuild American industry with a $10 billion fund to take direct stakes in companies. When their CEO laid out their priorities, he named advanced manufacturing, defense, aerospace, semiconductors, and life-saving medicines. If that sounds familiar, it's because those are the exact industries Prometheus is going after. This is where Prometheus and JP Morgan's incentives start to align. JP Morgan says their advantage comes from their relationships. The bank serves 34,000 mid-size companies and more than 90% of the Fortune 500. Its clients are regional part makers, the machine shops, the aerospace suppliers. If you're building AI for factories, that's your customer list. And if you're raising $100 billion to buy factories, that's your shopping list. BlackRock can lend, but JP Morgan is the one that can walk Prometheus straight to the companies it wants to partner with or to buy. But this goes even further. To steer the whole initiative, JP Morgan built a 12-person advisory council.

9:35 Jamie Dimon chairs it alongside Michael Dell, Ford CEO, two former defense secretaries, and Jeff Bezos. That's the room that Bezos is in, alongside the people who run American industry inside a program built to rebuild American manufacturing. And 6 months after he joined it, Dimon's bank wrote a check into his AI company. So, step back and take a look at the shape of it. BlackRock, a lender ready to finance takeovers, and J.P. Morgan, the bank leading a one and a half trillion-dollar initiative to rebuild American manufacturing with Bezos advising its plan to rebuild industries Prometheus is targeting. Once you look at all these pieces coming together, you see the AI, the data, the money, and the relationships all start pointing towards the same direction, towards taking over an industry and rebuilding it with the most capable and powerful partners in the world.

10:22 Everything I've described so far is the vision for Prometheus. Now, let's argue against it. Bezos has convinced investors to value his company at over $41 billion before it shipped a single product. That leaves room for a lot of skepticism, starting with the obvious. This company still hasn't disclosed a product. No one knows what their go-to-market will look like because there's nothing to sell yet. And when they are ready to sell, the approval process could be brutal.

10:45 Dimon can put Bezos in the rooms of decision-makers, but he can't make the industries themselves move any faster. Aerospace, semiconductors, pharmaceuticals, these are among the most regulated industries on Earth. Nothing gets built or sold in any of them without clearing proper testing, certification, and regulators whose entire job is to move slowly on purpose. Take Bezos' jet example, adding 10% more thrust takes 10 years to deliver. Design is only part of that decade. The rest is testing and passing certifications. A turbine blade won't just get certified faster because an AI designed it. Pilot programs and safety sign-ons often takes years before anyone approves anything because if the AI is wrong, failure isn't a bad answer on a screen, it could be a fatal mistake that can put companies and human lives in danger. And there's another factor that could derail Prometheus from achieving their goals. The path towards building their world model or artificial general engineer is still unclear. Large language models prove that more data and more compute reliably produce better models. That's why OpenAI and Anthropic can keep raising capital while remaining unprofitable because investors can see each round buying a measurably better model. Whether that translates into returns is still an open argument, but at least this scaling relationship is observable and clear. But the commercial value of world models has yet to be proven at scale. Obtaining the data required to improve these models has proven difficult. And even if Prometheus acquires all of the data from these factories, understanding how the physical world works is a completely different learning process than written words any large language model can learn from. In other words, world models are still early, promising, but commercially unproven. So, the $41 billion valuation and the reported $100 billion fund aren't priced off a proven business model, they're priced off a belief that chatbots were the first chapter and that language models are a stepping stone towards AI that creates value in the physical world. But Prometheus isn't the only company betting on that. Fei-Fei Li's World Labs raised about a billion dollars in February at a reported $5 billion valuation building AI that understands and creates realistic 3D worlds for physical AI. Dekart is building AI that creates realistic virtual worlds where physical AI and robots can learn and train. Their company was recently valued at nearly $4 billion with Nvidia and Toyota Ventures backing them. These aren't identical products, but they're all competing to establish the foundational model that lets AI understand and eventually operate in the physical world. And if one of these other companies gets there first, Prometheus could be seen as the overvalued company whose tech is behind competitors worth a fraction.

13:01 when Prometheus could actually go public, Bezos was asked directly about an IPO. His answer was that it's way too early to think about it. So, to wrap it all up, Prometheus is a company with no product, no disclosed revenue valued at $41 billion reportedly attached to a plan to raise $100 billion more to buy the factories that will teach its AI how the physical world behaves. If it were anyone else in the world making this pitch, they would be laughed out of the room. But this is Jeff Bezos, the man that built Amazon into the world's most advanced supply chain and is sending rockets into space with Blue Origin. So, yes, Prometheus is one of the most ambitious industrial bets of our lifetime. But they have the biggest financial institutions in the world already moving in to support them. And it's worth being clear about what the prize actually looks like. Because if this works, it's a civilization-changing technology. Bezos and Bajaj argue that AI won't destroy engineering jobs, it'll create more of them. Bezos even predicts labor scarcity, more demand for engineers than there are people to fill the roles. This is Jevons paradox. When something becomes far cheaper to produce, we don't end up wanting less of it, we want more. Make invention cheap and you don't get fewer engineers, you get more invention from more people. If you enjoyed this video, leave a like, comment, and if you're interested in learning about the most important private companies in the world before they go public, subscribe. And I'll see you in the next video.