Even if that kind of client side compression went totally unnoticed, and petabyte level use of cell and satellite services also went unnoticed, and even supposing we are only talking about a billion calls a day (unlikely given the population of the US alone, before even adding the rest of North America, Europe, and wherever else they want to spy), what are they going to do with all that data? It’s not a training source and any AI they unleashed to sort it would be subject to hallucinations and attack from people who know their conversations are being recorded.

I find it much more likely that they will target individuals directly for that level of surveillance. Hacking one journalist’s cell is going to be far easier than trying to find their recordings in the vault, assuming the targets have a new Chinese car.

They are much more useful as scouts for mapping things just like Google has been doing with their street view cars for decades now and as the other commenter pointed out, with a kill switch just like OnStar uses for stolen cars, as a first attack wave gumming up transport and evacuation options.