Sure. But there are alot more use cases for GPUs than machine learning, many of which are latency sensitive and must be done locally.