Part 2 made the distinction between utility and capture. This chapter applies that distinction to the strongest current ownership claim in AI: that a few centralized model providers will remain the permanent home of intelligence. I think that story is much weaker than it looks, and it sets up the SaaS consequences explored in Part 4 .
Intelligence will flow through a small number of centralized model providers because they have the best models and the infrastructure to run them.
That sounds reasonable today. It may even be true today, but I do not think it is the final form.
OpenAI, Anthropic and the other frontier labs are betting on a world where intelligence remains scarce, expensive, centralized and hard to reproduce. They have the models. They have the talent. They have the infrastructure relationships. They have the APIs. They have the mindshare. They have the brand.
That is a powerful position, but it is not automatically a durable moat.
A model lead is not a moat. It is a lead, and leads decay.