Comment by veggieroll

Indeed, I forsee a "training dataset consortium" arising out of this, whereby a bunch of companies team up to buy one copy of everything and then share it for training amongst themselves (ex. by reselling the entire library to each other for $1).