To keep expensive AI accelerators like GPUs and TPUs busy, you need storage that’s integrated with the compute and networking ...