A2) Not trying to compete in #AI model building, primary training, etc. gives more options for where to run the resulting inference workloads. The hardware and software options for that are all over the map from GPUs to baked in accelerators to #IBM #Spyre and more. #ciochat