You're correct about ease-of-use being key, I think. It was easier for "us" (the industry) to build hosted web servers, and so that's the paradigm that has won

You're correct about ease-of-use being key, I think. It was easier for "us" (the industry) to build hosted web servers, and so that's the paradigm that has won out. It's a direct evolution from client-server computing in the mainframe-and-terminal era. But the user doesn't need to care what a server is, or what running one involves; it's a bit of a red herring. A winning platform wouldn't communicate to people that they're running a server at all; they'd upload their messages/profile/etc, and the application user experience would be akin to that of any other application, with the difference that -- at an implementation level -- their data would be encrypted, replicated and hosted across multiple devices. The platform provider then goes on to win-in-competition because their hosting and bandwidth costs reduce to near-zero. That of course conflicts with the second point: evolving the protocols for that is hard. I'd wager that a winning platform will get 98%+ of the protocol design and implementation correct up-front, because it would have to be based on simple, iterable, secure and near-correct fundamentals that stand the test of time.