Why? What’s the benefit of centralizing in a handful of large relays?
As long as we figure out a sustainable model (a matter of time I think) and keep relays dumb, there’s no need to centralize.
Discussion
Whether we will centralize on few relays is really up to client developers. If most clients still query from relays and don’t treat local storage the primary storage, we will end up centralized. It has nothing to do with smart or dumb a relay is. We will centralize on relay implementations. But operations won’t