But as the name already implies, that one preferentially tries to match your files to the ones already available in Apple's music catalogue, with file upload only serving as a fallback if it can't find any matches.
And from what I've heard, the fingerprinting algorithm is fuzzy enough that it will often match songs/song versions that are very similar, but not actually identical, like censored/uncensored lyrics, different masterings, differing fade ins/outs, alternate takes and live versions, etc. etc.
Oh interesting – somehow in my mind I've mainly associated this phenomenon with Apple, but of course in a way this make sense and saving storage space that way is a tempting target for that kind of service I guess.
It makes me curious though how the accuracy vs. deduplication efficiency trade-off looks like in practice: How much storage is Apple saving with their current settings, and how much deduplication would they lose if they made their audio fingerprinting more accurate, up to a level where even audio buffs would stop complaining
Amazon used to provide this service, but they shut it down.