The claim that a dataset with billions of users has only dozens of people able/interested in doing data processing on it is a damning admission that the format is too hard to deal with
I maintain some OSM data-mangling code - moderately popular perhaps, but certainly not core - and even that has 840 github stars. I'd take the "dozens" as poetic licence really.
Dozens of open source volunteers who are interested in volunteering their free time to do software development using the format.
In addition to the innumerable developers in Facebook, Apple, and other corporations who are paid to do the data processing and actually bring the data to those billions of users.