“everybody has say problems retention which is the turn is very high um and defensibility like you make something and immediately you have a lot of clones because everybody's working with the same stuff everybody's using gp4 uh for Tech stuff almost or llama... within I would say if you're smart within three months with if you're less smart maybe six months to a year so it's very difficult to make a startup AI startup now”

AI startups have structural moat problems — assume profit margins collapse

All AI startups share the same underlying tech (GPT, Llama, Stable Diffusion), so clones arrive within 3-6 months and price competition pushes margins toward zero. Model the AI business as a high-revenue, low-margin, short-window play — not a long-term moat. Pieter explicitly tracks AI startup profit margin separately and accepts it will keep falling; the strategy is harvest then sell, not compound.