Everything above measures registry presence. None of it measures use, and the gap is probably large.
Publishing to the registry is close to free and permanent. There is no liveness requirement in these records, no usage signal, and — as the 231 with no connection details show — not even a guarantee that the entry describes something connectable. A count of registered servers is a count of acts of publishing, which was also true of every package registry in its growth phase, and in those cases the eventual gap between published and used was one to two orders of magnitude.
So the honest framing of "17,100 servers" is: at least 17,100 people or organisations went to the trouble of publishing one. That is a real and meaningful signal about developer interest. It is not a statement about how many are running, maintained, or called by anyone.
Treat any adoption claim built on a registry count — including the ones in this note — as measuring supply. Nothing here measures demand.