How decentralised is it, really?

The fediverse is described as a network where no single operator matters. Both of its concentration measures say otherwise, and they say it in different directions.

Activity is concentrated in a handful of servers

Of 49,378 servers, only 17,544 report any activity. Among those, the distribution is extreme: the median active server has 1 monthly actives, while the largest single server holds 19.5% of the whole network's.

Five hundred servers — one percent of the network — account for over ninety percent of its activity. The remaining 48,000 are, in aggregate, a rounding error in usage while being the entire basis of the claim to decentralisation.

Size distribution

A count of servers is not a count of anything a user or a regulator would recognise. Over half of them are switched on and empty.

Hosting is concentrated too — somewhere else

Of the 35,839 servers whose origin we can resolve, the top network carries 34.7% and the top four carry 65.0%. Note that this ranking is almost the inverse of the activity ranking: the network hosting the most servers hosts very little of the actual use, because it is carrying tens of thousands of ActivityPub-enabled blogs.

The blind spot

A further 12,497 servers sit behind a CDN, where the address belongs to the CDN's anycast estate and no honest statement about hosting jurisdiction is possible. We record those as unknown rather than attributing them to the CDN's country — which is what produces the neat, wrong country tables published elsewhere.

Not every busy server is a community

Posts per active user per week, from each server's own published activity series. A ratio in the hundreds is not a busy forum; it is automation — a news relay, a bridge, or a bot farm. This is the cheapest such detector in the dataset and it requires reading not one post.