Object storage
Cheap, very large storage that keeps data as objects reached over the web, used for datasets, models and backups.
Also: S3-compatible storage
Object storage holds data as named objects in buckets, reached over HTTP rather than as a mounted drive. Amazon S3, launched in 2006, set the pattern, and most clouds offer a compatible service. It scales to exabytes at low cost but is slower than a parallel file system, so AI clusters usually keep training data and older checkpoints in object storage and copy what they need onto faster tiers. Model weights are often loaded from object storage when inference servers start.