it means they can serve more with what they have if they implement models with deepseek's optimizations. More usage doesn't mean Nvidia will get the same margins when cloud providers scale out with this innovation.