A problem is that the underlying technology and protocols scale mainly proportional to number of instances and not users.
In other words, more instances mean a lot more resources are needed to communicate messages between them, but twice as many users doesn't require that much more.
This is baked in to the engineering decisions that were already made, and it would be pretty hard to change those decisions now.
So, whether or not the future is large instances philosophically, the system was programmed to promote them technologically.
Yep! AP gives some room for flexibility for transport mechanism, but it generally specifies how different instances communicate with each other with inboxes and outboxes.
@volkris @jeff I am not sure that it is intrisic to activitypub that instances need to communicate with each other for every update. We can use relays or implement a message propagation method to use gossip protocols, which would avoid this issue and make the network more resilient.