With distributed hash tables it is manageable. You do something like “store three copies on three peers” and as long as one of them is online the post is accessible. This is actually better than the way lemmy does it now. In principle each lemmy server stores the posts from its communities, and a copy of each post from communities its users are subscribed to. But since all instances are federated so well, in practice each of the 1000 lemmy instances stores a copy of almost every post ever made. That’s like 100GB x1000. With a DHT, the amount of space used on each user’s device is on average the amount of posts one user makes x3, no more.