S3 Standard and EBS gp3 list pricing held flat for another quarter even as the substrate got more expensive: nearline HDD trade prices rose ~10% QoQ in Q2 and enterprise NAND is booking out through 2027. AWS can absorb a storage-line squeeze inside a much larger compute bill, which is exactly why the independents that couldn’t (Wasabi, Backblaze) already raised prices. The read for buyers is unchanged but more urgent: treat Q4 enterprise-agreement renewals as the checkpoint where this cost pressure finally surfaces.

Amazon S3 Vectors — the first cloud object storage with native vector indexing — is generally available, letting teams store and query embeddings for search and RAG directly in S3 without standing up a separate vector database. It is the clearest expression yet of AWS’s pattern this year: fold more of the AI data stack into S3 so each new workload defaults to the bucket and the switching cost of leaving compounds.

S3 Annotations — up to 1 GB of mutable, queryable JSON/XML/YAML context attached to each object — is GA across regions, AWS’s answer to Google’s object-context push. Both are converting object stores into native knowledge bases for AI agents and eliminating the separate metadata database. Whoever owns the metadata owns the workflows; portability of annotations across clouds is now a genuine architecture decision.

Microsoft opened the public preview of Azure Storage Discovery, a fully managed service giving enterprises a single view of their Azure Blob data estate — capacity, activity, configuration, and cost across accounts. It is the governance-and-FinOps layer productized: as storage becomes a first-order AI cost, Microsoft is selling the tooling to see and control it, not just the capacity underneath. Expect classification and policy enforcement to follow.

Microsoft continues extending native Blob connectors into Microsoft Foundry, Ray, Anyscale, and LangChain, letting enterprise data feed grounding and fine-tuning without a separate ETL hop. The strategic logic mirrors AWS’s S3-as-AI-substrate push: the more frameworks read Blob natively, the more each new AI workload defaults to Azure storage rather than a neutral data lake.

Two quiet Blob updates point the same way: Entra ID-based SFTP replaces legacy local-user auth (centralizing access under conditional-access policy and audit), and client-side CRC64-NVME integrity validation is GA across the .NET, C++, and JavaScript SDKs. Neither is a headline, but both are the plumbing of regulated AI pipelines where shared keys and silent bit-rot are audit findings waiting to happen. Inventory what still authenticates with account keys.

Nearline multi-region storage went from $0.010 to $0.015/GB-month, while Archive multi-region fell from $0.0040 to $0.0024/GB-month (US/EU). Lifecycle rules written before 2026 that push aging data into multi-region Nearline now cost 50% more to run and may no longer beat leaving data in Standard; rules pushing straight to Archive got cheaper. The fix isn’t a new tool — it is re-running the TCO math on policies nobody has touched since they were set.

Google is embedding AI models directly in the storage layer: automated metadata annotation that stores data with full context about its content, plus native agent connectivity via MCP. The pitch is that raw objects arrive already enriched and agent-ready, collapsing the separate cataloging step. It is the same metadata-ownership contest AWS is running with Annotations — whoever enriches the object store owns the agent workflows that read it.

Cloud Storage Rapid — Google’s zonal class on Colossus, delivering 15+ TB/s aggregate throughput with sub-millisecond latency and ingest-on-write — is moving from Next ’26 showcase to production, with checkpoint restores reported multiples faster than standard object storage. It is Google’s clearest answer to the claim that object storage is too slow for GPU-bound training loops; the tradeoff is zonal-only placement and tighter compute/storage co-location.

The storage supply crunch deepened rather than eased: nearline hard-drive trade prices rose roughly 10% quarter-over-quarter in Q2 and are up about 46% since September 2025, with Western Digital and Seagate confirming essentially all of 2026 capacity is sold to AI data centers and lead times stretching toward two years. Bank of America and Morgan Stanley project pricing won’t normalize until early 2028. Cold cloud storage rests on this substrate — the pricing floor has structurally lifted.

SanDisk’s August 5 fiscal Q4 report disclosed roughly $93.9B of contracted business across eight long-term agreements, with about half of fiscal-2027 output already committed and analysts lifting estimates on tight NAND supply and improving AI-driven pricing. With HDDs sold out and NAND booking out, there is no cheap medium left to tier toward — the entire storage hierarchy is now allocated years ahead.

Wasabi lifted pay-as-you-go pricing to $7.99/TB-month (its second increase in 14 months) and Backblaze raised B2 to $6.95/TB while dropping API transaction fees. Both major independent object-storage vendors have now raised list prices within a year, removing the pricing gap that was their main differentiator. It is the clearest signal that the secular decline in cloud storage pricing ended in 2025 as AI reallocated the memory-and-storage supply chain.