Author of *The AI Bubble Hater’s Guide*, Ed Zitron, shares 5 bearish takes on large models and generative AI: 1. Margins are permanently broken OpenAI is projected to burn $20.9 billion by 2025. These companies’ costs rise proportionally with revenue, and there’s no evidence they can improve profit margins. OpenAI might even delay its IPO to 2027 because it can’t hit a trillion-dollar valuation. 2. The business model inherently encourages waste and can’t charge based on results Anthropic and OpenAI never charge based on outcomes. Their business logic is to encourage you to spend more, burn tokens like crazy, and casually learn from your ideas/IP (Anthropic’s Claude Design for Figma is an example). They even say they can’t wait to see what you’ll build with it, which translates to: they don’t know what this stuff is good for either 3. Oracle and new cloud providers are the first dominoes to fall Oracle admitted in its annual report: some customers (like OpenAI) are highly leveraged and may face payment risks. If OpenAI can’t pay, Larry Ellison can’t balance the books, and Oracle’s stock is in danger. New cloud providers (CoreWeave, Nebius, IREN, Cipher Mining, TeraWulf) are essentially NVIDIA subsidiaries. NVIDIA even has to pay to rent back GPUs it sold to them. This kind of thing only happens in an industry that isn’t real. 4. Big tech’s AI revenue growth is smoke and mirrors; AI itself is just losing money Microsoft and Amazon only disclose annualized AI revenue but never reveal actual AI profits because it’s all bad news. 5. Large models are a dead end—when will the collapse happen? The future of LLMs is boring hardware—a real market worth only $10–30 billion, dressed up to look like a trillion-dollar industry. When the first company slashes CapEx, the rest will follow. Who will cut first? #AI #GenerativeAI #TechBubble