> In theory, after the protocol was widely adopted, Mozilla could shut down the centralized service. In practice, nobody adopted the protocol

Sounds a lot like the Matrix chat protocol. Who’s using a private Matrix server?

I feel like a lot of these companies think they’ll handle the bootstrapping stage with economies-of-scale from having one central node; but in practice, having that one central node enables extra network effects (e.g. super-low-latency interactions with people on the same node) that make people resistant to eventually becoming more distributed.

I think WordPress.com represents a better model: it’s a first-party hosting provider, but everyone that signs up is signing up for their very own instance. (Maybe some things are shared under the covers—that’d certainly be good engineering—but everyone gets full control of their WordPress “engine”, with their own plugins, scheduled tasks, etc.) This way, people are used to the concept of these blogs being a loosely-federated network with pingbacks et al from the start, rather than everything being one server with one dashboard. Very easy from there to “lift” your instance out into a separate enterprise deployment; very easy for other cloud providers to spring up and offer you an import tool, to move “your server” from service X to service Y.