Nvidia is also pushing for local inference IMHO. They want open models and competition in the model layer, not two big labs controlling all of it.