What are the differences between DFS, Object Storage, and local NVMe?

Choose based on access pattern, durability, and locality:

CoreWeave AI Object Storage is an S3-compatible object store, accessible from any region and shared across an organization. Use it for datasets, model artifacts, and anything read many times by many workloads. See About Object Storage.
Distributed File Storage (DFS) is a POSIX shared filesystem exposed as PVCs with ReadWriteMany access. Use it for shared scratch space, home directories, and checkpoints during a training run within a cluster. See About Distributed File Storage.
Local NVMe is per-Node ephemeral storage with the lowest latency and highest IOPS, but data is lost when the Pod or Node terminates. Use it for shuffle buffers, intermediate compute artifacts, and scratch. See Local storage.