The last few weeks have been rough for anyone holding AI names. The question everyone is asking is simple: is this where the whole thing cracks? I didn’t want to answer that from my gut, so I took the 12 bear arguments I kept seeing and checked each one against the filings, Anthropic’s S-1 and the sell-side numbers. A lot of the research out there gives you a target or a number without showing the reasoning behind it. Selloffs make the bear case louder, so I wanted to work through each concern myself and show the math. Here are the 12 I cover: 1. Capex isn't earning a return 2. Free cash flow went negative 3. GPUs become obsolete too soon 4. We're overbuilding 5. Token buyers aren't getting value 6. AI labs don't make money 7. OpenAI RPO concentration 8. The payback math is too good to be true 9. Credit is cracking 10. Memory is crowding out the rest of tech 11. There isn't enough capital 12. A slowdown in the data center buildout Some of these didn’t hold up once I looked closely. Some are real, and I explain who they actually hurt. And one I hadn’t given enough weight until I ran the numbers myself. I also built a stress test model of what one gigawatt of compute earns when it’s funded entirely with debt, and stress-tested it at higher rates and lower revenue to see where it breaks. That model, along with what Anthropic’s S-1 shows about AI lab economics, may change how people think about the funding risk. If you’re trying to make sense of your own AI positions right now, I think this one is worth your time. Read it here: