There are less questionable things they can do with all that data: I'm sure it's really useful to figure out how electrical and mechanical faults correlate with driving conditions / weather / region etc. The result from this is just a better car in the next iteration.
Then someone offers them $1 each for a fleet of 1M cars, who is going to turn down $1M for a few TB of data?
Then someone offers them $1 each for a fleet of 1M cars, who is going to turn down $1M for a few TB of data?