Isn't it obvious? Web scraping and even scanning written books is at record levels because the data is so useful for training. They are using every byte of user data.