“Google is currently deploying specialized hardware architectures and configurations specifically for inference workloads.”