AI in crypto has so many issues: 1. The most badass models are definitely out of reach. 2. Can't even handle the infrastructure for the models. 3. So, they can only focus on the most downstream applications. The problem is, the really awesome applications won't come to the crypto space. If they're good, big companies will acquire them for billions of dollars (like Manus, OpenClaw). The crappy, patchwork ones are the only ones that end up in crypto. 4. But here's the thing: the overall development model of AI is 'fat models, thin applications.' No matter how fancy the applications are, the next model upgrade might just render them obsolete. So, it's not just traditional SaaS that's struggling—AI applications, in the context of models iterating every 4 months, might also be a false proposition with no moat.