# Oklo Inc: The Revenue-Less Nuclear Emperor Worth $20 Billion > Sam Altman's nuclear startup Oklo is worth $20 billion with zero revenue, zero licenses, zero contracts. A political bet that smells like WeWork 2.0. - Canonical: https://siliconvalleyconfidential.com/en/dossier/oklo-inc-the-revenue-less-nuclear-emperor-worth-20-billion/ - Site: Silicon Valley Confidential (https://siliconvalleyconfidential.com) — weekly executive intelligence on Silicon Valley and global tech - Author: Jose Luis Cases (https://es.linkedin.com/in/jose-luis-cases-lozano) - Language: en - Published: 2025-11-01 (original LinkedIn edition: https://www.linkedin.com/pulse/oklo-inc-el-emperador-nuclear-sin-ingresos-que-vale-20-cases-dhhuf/) --- THE BOMBSHELL OF THE WEEK --------------------- On October 30, Bloomberg published something troubling about Oklo Inc., Sam Altman's nuclear startup, which reached a $20 billion market capitalization with zero revenue, zero license to operate reactors, and zero binding supply contracts. The stock is up 500% since January 2025, making it the most extreme case of speculative valuation in the post-ZIRP (Zero Interest Rate Policy) era. Bloomberg didn't mince words: it titled its investigation "The Risky Movement to Make America Nuclear Again" and revealed that Oklo's backers hold "wealth and political connections that could undermine nuclear safety." In other words: they're betting on influencing regulators, not on complying with the current rules. The strategy looks easy on paper — they signed a $2 billion deal with newcleo (Europe) to build nuclear fuel infrastructure in the United States, but they can't legally operate until they get permits from the NRC (Nuclear Regulatory Commission). Those permits have been getting rejected for years. Their real bet isn't technological, it's political: that a Trump administration will tear down the regulatory barriers that nuclear safety science has built. Altman resigned as Chairman in April 2025 after a decade, right before OpenAI announced multiple energy deals. The conflict of interest was too obvious. This is WeWork 2.0: a stratospheric valuation built on political narrative plus a famous founder, with nonexistent business fundamentals. But unlike WeWork, criticizing nuclear now reads as anti-American in the current political climate, which has silenced the scrutiny this valuation deserves. POWER MOVES ------------------------ Sam Altman promises an AI that does research on its own by 2028 (if he survives the spending) On October 28, during a Q&A session, Altman promised that OpenAI will have a "legitimate AI researcher" by 2028. It sounds abstract, but what it means is: a machine that can do real science without humans. The problem? To get there, he plans to spend $1,400 billion. (That's 1.4 trillion in our terms.) The Apollo program that took us to the moon cost about $280 billion adjusted for inflation. Altman wants to spend five times that. Now here's the good part: to avoid going broke spending all that, OpenAI needs to bring in between $100 and $200 billion a year. Today they bring in about $13 billion. To put it in perspective: Google just reported $102 billion in a single quarter (Q3 2025). OpenAI would have to achieve in a year what Google does in three months. Oh, and Google has a 25-year head start, millions of customers, and proven products. That's why Altman wants to go public in 2026 asking for a $1 Trillion valuation. He needs that money before the party ends. If it works out, OpenAI will have changed humanity. If it goes wrong, it will be remembered as the stupidest spending spree in the history of technology. Qualcomm says it can beat Nvidia For years, Nvidia has dominated the AI chip market. They're so dominant it's almost a monopoly. On October 27, Qualcomm announced two new chips (AI200 and AI250) that work in a completely different way: instead of being super fast at processing, they have massive memory (like having a giant hard drive glued directly to the processor). Why does it matter? Because today's AI models are choking from lack of memory, not lack of speed. It's like having a Ferrari with a motorcycle's gas tank: you go incredibly fast, but you stop every 20 kilometers. Qualcomm is proposing a truck with a giant tank that maybe isn't as fast, but never stops. Qualcomm's shares jumped 20% in a day. Investors understand that if this works, Nvidia has its first real competitor in a decade. And the timing is perfect: Nvidia is starting to show signs that its dominance isn't invincible — though I think there's still a long way to go. I'll explain below. MONEY TALKS --------------- Legora: the startup replacing lawyers (and law firms are happily paying) You've probably never heard of Legora. That's normal — it's a European company founded in 2023 that makes AI software for lawyers. But on October 30 it raised $150 million at a $1,800 million valuation. Just five months ago it was worth $675 million. It tripled its value in half a year. What exactly does it do? It automates the legal work that normally costs $800 an hour: reviewing contracts, researching case law, preparing documents. The kind of thing a lawyer charges good money for that's actually tedious work they hate. The interesting part is who's paying them: the most expensive law firms in the world (Linklaters, Cleary Gottlieb, Goodwin). These are the firms billing their clients $1,500 an hour. And they're buying software that replaces their own employees. Why? Because if they don't, another firm will, and it will steal their clients by offering the same service at half the price. While everyone talks about ChatGPT, the companies actually making money with AI are the ones nobody knows. Legora went from 250 to 400 clients in five months. That's the kind of growth that makes investors lose their minds. PRODUCT SECRETS -------------------- Reducto: the company that feeds ChatGPT Here's an AI industry secret: ChatGPT, Claude and all those tools that talk to you can't read PDFs, contracts or invoices directly. They need someone to convert those documents into clean text first. That "someone" is Reducto. On October 27 they raised $75 million led by Andreessen Horowitz. The figure they buried in the fine print: they process almost a billion pages a month. To give you an idea, that's like digitizing the entire US Library of Congress every two days. Who uses Reducto? Harvey (AI for lawyers), Scale AI (the company that trains AI models for OpenAI), and several Fortune 10 companies that can't publicly say they use AI. They're the equivalent of selling picks and shovels during the gold rush: it doesn't matter who strikes gold (OpenAI, Anthropic, Google) — everyone needs Reducto's tools to dig. Andreessen Horowitz calls them "the magic ingredient modern AI companies need to build." If Reducto stopped working, half the AI ecosystem would collapse. That's why in just two years they've raised $108 million. When you're critical, invisible infrastructure, investors chase you. MY TAKE --------------- The energy-for-AI issue can't be ignored, and it worries me that political interests could steamroll barriers that have been well established until now. Sam Altman keeps looking like a supervillain to me with every move he makes, but he's already inside a snowball he can no longer stop. His agent builder is a clear example that he doesn't know how to build product. And that he doesn't know it, I get — but with so many smart people around, someone must!! Google does it much better. In the end, people will want to browse in an environment they trust, not hop from one tool to another, and I think Google has a BIG head start. I'm a fan of NotebookLM and nanobanana, of Veo 3, etc... The superhuman research agent sounds very interesting, but I think he's losing the fundamentals that any neighborhood product manager has mastered. Start small, with an MVP, build something that solves a REAL problem — like discovering an impossible cure, proving the Riemann hypothesis, or something that resonates. And then... bring in the artillery... The Qualcomm story is interesting, but there are a few details worth flagging, in case you're tempted to start buying shares like a maniac (don't!). The AI200 chip is designed for INFERENCE, not initial training. It has a technical parallel with what Apple does in its unified memory architecture with the M1, M3, M4 and the new M5 chips. This is VERY important, and the devil is always in the details. While in inference that unified architecture with up to 768GB of RAM can be very interesting — though Apple has been doing it for a long time — in TRAINING, raw compute power (TFLOPS), memory bandwidth and parallelization capacity are key, and there NVIDIA still has no rival. And for a model to run inference, it first has to be trained. So draw your own conclusions. Legora feels like a natural step. As I've said before, AI moves along the path of least friction.... toward the money.... First coding, now legal... then will come finance, consulting (there's already work on MAAS — "management as a service")... And that's it. I hope this got you up to speed and opens your mind to make good decisions. And you know... if you like it, share it — you'd be doing me a big favor. Have a great weekend