Nvidia can start putting weights in silicon if model development slows down.