If you have 10 million people all spread out across small servers, how do you suggest I find everyone posting a particular hashtag?
Social graphs can offer a similar discoverability without search. You see what those you follow post, comment on, like, share, etc. Following tags just like you follow a person also works well. Before assuming we must have to have massive indexing and searching services, ask yourself whether you really need to be able to discover anything that anyone around the world posts, regardless of whether you have any sort of connection to that person, their context, their society, or their culture.
Anyway, I agree that a social graph has the potential to offer human-scale and $5-VPS-scale discoverability. The problem is probably with user expectation and network effect. How do you get people to use the thing when no one is using it yet, to bootstrap to the point where you might actually bump into someone with the same niche interest as you. Still, it could start with small communities same way telegram group chats and discord servers do. You just have to bring the people to you instead of just trying to index the firehose of global content.
The best I can come up with is a historical record. Beyond that I'm at a loss for a use case that is noticeably better with search compared to social graph-based discovery.
My point was that something like BlueSky can make content more discoverable by letting you build your own social graph and view content from those you're connected to.
Search is most useful for discovering content that you otherwise have no connection to at all. There is also the historical aspect and there really isn't a way around that, but in the original context of finding people actively talking about a specific topic it isn't a necessarily must-have feature.
