Azure Blob Storage

Use it when

• The platform runs on Azure; it is the substrate under Synapse, Databricks on Azure, and Microsoft Fabric's OneLake.
• Directory semantics matter: enabling the hierarchical namespace (which is all Data Lake Storage is) makes directory rename and delete atomic metadata operations, with POSIX ACLs per directory.
• Engines arrive through Hadoop tooling; the ABFS driver lets Spark and Presto treat it as a filesystem.
• More than objects are needed: NFS 3.0, SFTP, and WORM retention to SEC 17a-4 come from the same account.

Think twice when

• Tooling is written against the S3 API; there is no S3 compatibility, so clients must speak the Blob REST API or ABFS.
• Data leaves Azure regularly; egress past the free allowance is charged by zone, the same shape as S3 and GCS.
• Cold data churns: the Cool, Cold, and Archive tiers carry 30-, 90-, and 180-day minimums.

How it runs

Fully managed, with durability from eleven nines locally to sixteen nines geo-zone-redundant, and asynchronous geo-replication to a paired region with readable secondaries. Hot through Archive tiers with automatic movement via Smart tiering; events flow through Event Grid into Functions or webhooks. The hierarchical namespace is a setting at ordinary Blob prices, so it is a capability decision, not a cost one.