Reminder that Bluesky is not decentralized, and can be censored or bought out just like Twitter.
Bluesky is private but the underlying mechanism is OSS and accounts are portable.
Go build the replacement and people can port their accounts across.
Maybe you could theoretically have an AT "app view" that takes data from multiple relays, but nothing in the implementation does anything to support that, and as far as I know nothing in the protocol does anything to help it discover the relays... which in practice means that even if you extend the app views to use multiple relays, there will never be more than a handful of relays with meaningful reach.
The AT protocol is at best a really crappy excuse for decentralization. And frankly a pretty poor example of open source too, given the usability and organization of the code they release.
Compare with, say, Nostr, which is actually decently decentralized... but, in not-unrelated news, suffers from massive content discovery problems. Or compare with Briar, which is even more decentralized but has both discovery and scaling problems. Or for that matter Usenet.
I'm not sure there is one. But that's because I don't accept the idea that "likes" and "follows" are the best way to find content, or even a good way. If you do accept the idea that those should be your primary way of discovering content, which Bluesky does seem to accept, then decentralization becomes a more important criterion, and Nostr or even Mastodon is more effective that AT. Unfortunate about the culture on Nostr, though...
You could maybe build a system that I would think was better by, say, indexing Nostr using some kind of DHT. But you'd have to do some things to traditional DHTs to make them more attack-resistant. And maybe more things so they could scale to that size. Having "topics" like newsgroups or subreddits would be another approach, and could probably be grafted into pretty much any protocol.
Contrast this with Mastodon which already has a vibrant federated ecosystem.