That user number is not reproducible last time I tried (~8 months ago). I was looking at the code the other day and saw what I believe is one of the reasons, but I still couldn't find several million accounts (>10%), which is pretty hard to lose. (8+ bsky run pds equivalents)
This also does not account for (1) people with multiple accounts (labellers, feeds, bots, intent) or actual activity (significant % are likely churned, didn't delete)
Even if the stats are off by a factor of 10 it wouldn't matter. You could remove the numbers from the user axis entirely and it would paint the same story: there were massive user influxes after the 2024 US election and the inauguration in January, but user retention has been on a steep decline for the year+ since.
Again, this is not a reflection of anything bad about Bluesky as a user. IMO a smaller and more focused is a good thing for the actual community, hence why I read/post on HN and not Reddit or Twitter. However as an investor there's basically no way to interpret those statistics as anything but bad.
Strong agreement, though I would say it looks to have reached a stable level for now. I've found several subcommunities that I can get good info from. I'm curious how the '26 election cycle will affect things, already seeing increase political discourse.