Starting around June 1, there was a general strategy to upgrade hardware on the established Lemmy servers (that had been around for 6 months or longer, some going back several years).... but hardware upgrades didn't work. Lemmy had serious flaws with federation performance and the PostgreSQL database.
This landed us where we are in mid-July. over 1,300 servers with less than 66,000 active users!
A massive internal SQL usage performance problem was fixed in Lemmy 0.18.2, providing great relief to the most popular servers. It turned out that having an explosion of instances was making the performance far worse, as the code that distributed content to peer servers had extremely inefficient SQL queries.
the legacy is that we have a huge number of servers, but the problems with the logic in the code for federation not being complete/tested.