[socksy]

He's almost certainly talking about the AppView. Yes, people run their own PDSes, but afaik no-one is running their own AppViews. It seems prohibitively expensive. Even running a relay requires at a minimum terabytes of storage space (and presumably more when the network gets bigger?).

The AppView - controls the way you can search posts - has the ability to show or hide posts from everyone using that AV - has per-user data such as notifications, last skeet seen etc - is a CDN for media, and as such can filter there - is where the algorithms of what posts to show are implemented

In other words, someone who controls the AppView holds a lot of power in the AT Protocol world.

At the moment, it seems that Bluesky are doing a great job with their AppView, and they've rather magnificently scaled with the extreme load increases in the last months. However, for someone like Doctorow who wants independence from being on someone else's platform, there has to be a viable alternative AppView, or at the very least a documented process for how to host it.

However, it's my personal belief that it's only a matter of time until people do this. I don't know how practical it will be (and what level of resources would be needed to divulge if Bluesky go bad compared to a Mastodon server), but it seems like this is a priority for them, and everything else in the stack has been made easy to self host (see https://alice.bsky.sh/post/3laega7icmi2q).