Some things can go faster if you add more workers, some things can only go faster if you make the workers bigger or faster .
If you’re tidying a garden you can get it all done more quickly, and tackle bigger gardens, by getting your friends to help. That’s horizontal scaling.
If you need to get a parcel from your house to Burkina Faso the only way to do that more quickly is to use a bigger, faster machine. That’s vertical scaling.
The way Lemmy is designed right now (says the op, I don’t know the detail) you can only support more users by making the server bigger and more expensive, not by using lots of smaller servers.
Edit: note that Lemmy as a whole scales horizontally: more instances == more users, but each instance has to scale vertically.
Seems like storage is the cost prohibitive part. Search is going to be difficult, too. I wonder whether the model there might be community supported shared services. A bunch of instances could jointly run an elasticsearch cluster or Algolia instance and charge for API access to cover costs.