Note Bluesky is architected to be downstream of PDS (personal data servers) which any user can switch to another provider, and the Bluesky app server acts as an aggregator (but anyone else can build their own aggregators — and people already have). So as Natalie notes in the post, you don’t even “have to” use the Bluesky app API to access the posts. You can get them from a third-party app server (“app view”) or even have your own. They’re all aggregating from the same data source.
Some of this is already outdated. With the switch to non-indexing relays (“Sync 1.1”), people are already running independent relays quite cheaply. There are also actual independent AppViews coming up.
Just to clarify, do you understand that this isn’t the same model as in Mastodon/ActivityPub? It isn’t a bunch of application servers “talking to each other” that different people are “on”. It’s not the kind of situation where your post goes viral and “your server” “goes offline” as a result. There’s a separation.
Data is hosted on PDSs (which are already super cheap to host and are not resource-intensive because they’re just storing data and not doing much computation). They’re not “talking to each other” or serving the web app like Mastodon servers do — they just store your commits like Git and let you subscribe to that stream. I don’t know what “relatively large” means to you as most independent PDS’s are individual people hosting their own data (so they’re actually small) but this sector is growing.
Then, there’s relays (not a lot now but again, they’re now much cheaper to run since Sync 1.1 updates).
And then there’s AppViews (which are more expensive to run because it’s like running an actual app backend with database and all that). Which isn’t too bad if you serve a smaller userbase but is harder if you want to serve millions of people like Bluesky’s own AppView does. Normal considerations of “running a web app for millions of users” apply here. But keep in mind that people on all backends see exactly the same data — it’s not being sent back and forth like messages between “federated servers” like on Mastodon. It’s being aggregated from the entire network via relay(s). So if I want to run a Bluesky backend (aka AppView) for a small group of people, I can do that relatively cheap and see the same content as the rest of the network sees. If my content goes viral, my servers won’t go down because going viral has nothing to do with the load on AppView. The load on AppView is just about how many people are hitting my own backend from their browsers (apps) — nothing to do with which server I have chosen to store data on (PDS) or to host my custom web app on (AppView).
So when you say “list of servers”, what kind of servers are you referring to? They all have different incentives to run (I’ve enumerated them above) and different relation to the user (PDS is something you choose, akin to hosting; AppView is determined by your client; relay is kind of behind the scenes and determined by AppView). And if you’re concerned about the load, I want to better understand which scenario you’re describing.
