The way lemmy (and federation) works, it needs to do a bunch of operations that can't happen simultaneously, so there's a job queue. The queue needs to do some database operations and a bunch of communication operations and each of the jobs needs to reach out to distant servers that may or may not be overwhelmed themselves.

You start with one server it costs almost nothing to host. Sooner or later you want to split out the job servers, then you end up needing to split out the database, when you start getting that many people on your server now you want to consider fault tolerance, Even after tuning you can only fit so many simultaneous users on a web server, you end up needing to do some load balancing. The next step would be trying to split it up geography-wise.

That's scaling up and it's what big companies do and it's very expensive but easy for a small team to manage.

Lemmy on the other hand is designed to be scaled out, running smaller individual user bases on lighter hardware with a bunch of individual administrators instead of a organized team.

If people want to be on a large single cluster application Reddit is still there.

I like what we have a lot better.