Volume Clones, launched in October 2025 for same-account copies within an Availability Zone, now extend across AWS accounts with optional re-encryption. The clone is usable within seconds, without the snapshot-and-restore cycle. For multi-account landing zones, which is how most regulated enterprises run AWS, this removes the main friction in refreshing dev, test and analytics environments from production. It also narrows one of Azure’s clearer block-storage advantages, Instant Access snapshots.

Lambda functions can now be configured to read either from S3 Files’ high-performance storage layer or straight from the underlying bucket. It is a small control, but it makes S3 Files (the NFS-semantics layer AWS went GA with in April) a first-class serverless data source. Teams can put latency-sensitive inference or ETL on the fast path and leave bulk reads on cheaper bucket access.

Microsoft’s retirement of general-purpose v1 and legacy blob-only accounts takes effect October 13. Accounts not migrated by then are converted to GPv2 automatically, and transaction-heavy workloads can see higher bills under GPv2 pricing. Databricks-managed DBFS accounts are being migrated by Microsoft ahead of the date. Storage Discovery, now GA, is the inventory tool Microsoft points to. The decision left for customers is whether they choose access tiers and redundancy or accept Microsoft’s defaults.

Elastic SAN, Azure’s managed cloud SAN, now has native, generally available protection through Azure Backup. That closes a gap that had kept some SAN-migration projects on third-party backup tooling. It makes Elastic SAN a more credible landing zone for on-premises SAN estates, especially Microsoft’s own database workloads, as VMware-exit and SAN-refresh programs keep moving to the cloud.

Google made automated end-to-end checksumming the default for object reads and writes across Cloud Storage client libraries, so integrity is checked from client to bucket without opt-in code. At AI scale, where training corpora and checkpoints run to petabytes and a silently corrupted shard can waste a run, this is a practical reliability feature. It also lets Google argue that its client stack is safer by default than hand-rolled S3 SDK usage.

Storage Intelligence advisor is now generally available for monitoring Cloud Storage across organizations, folders and projects, with curated metrics, automated anomaly detection and recommendations. It follows September 3’s batch-operations expansion (CEL object filters, up to 1,000 buckets per job, bulk storage-class transitions, dry runs). Together they are GCP’s answer to S3 Storage Lens plus Batch Operations, aimed at cost governance across large, multi-project estates.

TrendForce’s 4Q26 forecast has NAND contracts up 15–20% QoQ, with enterprise SSD the only category still accelerating, as U.S. cloud providers raise their purchase forecasts. Hyperscalers buy flash on long-term agreements, but this is still rising cost of goods for every SSD-backed tier (EBS gp3/io2, Premium SSD v2, Hyperdisk). Expect list prices to hold rather than fall, and more aggressive steering toward HDD-backed and archive classes.

The first public financials from a frontier AI lab put multi-year infrastructure obligations at about $518B. The largest disclosed piece is a five-year TPU capacity lease that Broadcom may help finance with up to $42B of convertible notes. Storage follows the accelerators: dataset, checkpoint and serving storage for clusters this size is a large, long-dated revenue stream. It concentrates in whichever cloud hosts the compute, so AI-lab hosting deals now matter for storage market share.