Even aside from training datasets, as long as they keep hosting chats for people using these genAIs their storage will keep expanding over time as people upload/generate more pictures and videos.
I really hope all this urgency on getting as much hardware as possible is because the ai companies see the funding drying up soon and are trying to maximize what they get out of it, but it’s probably more likely they’re just focusing more on videos now and want to have all of youtube (+etc) in their training sets