Is there even an argument that nostr does scale? I haven't had the time to examine it as closely as I should. But my understanding is there's no mechanism to spread the load on popular relays; even simple things like avoiding DoS attacks aren't clear yet beyond "paid relays". Which of course, isn't a very decentralized way of moving data around.

At small scale these problems are probably not a big deal as lots of people can step in to run new relays. But at Twitter scale, if big relays went down it'd take quite a lot for someone else to take up the slack and for users to switch over.