There's been much talk about which memories might best support AI and ML. But just as important are the architectures, regardless of whether the workload is done in the cloud or at the edge.