Cloud Storage Briefing

Announced at AWS Summit New York on June 17, Amazon S3 Annotations lets you attach up to 1 GB of rich, mutable, and queryable context directly to S3 objects — designed for AI agents and autonomous workflows that need to discover, understand, and act on data at scale. Annotations are stored inline with the object, IAM-governed, and queryable via S3’s existing APIs without standing up a separate vector database or metadata catalog. This is a direct competitive counter to Google Cloud’s Smart Storage Object Context (GA at Next ’26). For enterprise data engineering teams, this changes the architecture question: instead of building retrieval pipelines and annotation services, you annotate at write time and let S3 surface the context on demand. The implication for AI agent infrastructure is significant — S3 begins to function as a queryable knowledge layer, not just a dumb object store.

Amazon S3 Express One Zone — AWS’s high-performance, single-AZ storage class with 85% lower GET costs and 10× lower latency than S3 Standard — now supports S3 Inventory. Scheduled daily or weekly reports list all objects in directory buckets along with metadata and encryption status. The practical gap this closes: Express One Zone was previously a black box for compliance and governance teams because it lacked the object-level reporting tools available in standard S3. For FinOps teams tracking AI training workloads on Express One Zone, Inventory enables cost attribution at the object level for the first time. Combined with CloudWatch request metrics (added in March), Express One Zone now has a complete operational instrumentation layer comparable to S3 Standard.

AWS announced a new Local Zone in Athens, Greece (GA July 2026) — one of the first globally to offer both Amazon S3 and Amazon EBS Local Snapshots in-zone. Local Zones historically offered compute and EBS but required data to traverse to the parent region for object storage. Athens changes that pattern: enterprises with Greek data residency requirements can now run full storage architectures — object and block with local snapshots — without data leaving the jurisdiction. For regulated industries (financial services, healthcare, government) expanding in Southern Europe, this is the first Local Zone that qualifies for strict data localization mandates. The signal for the broader AWS roadmap: local-residency object storage is now a repeatable deployment pattern, not a one-off exception.

Microsoft has blocked creation of new GPv1 and legacy Blob storage accounts via the Azure Resource Manager API as of June 2026 — the enforcement layer used by every IaC toolchain (Terraform, Bicep, ARM templates). The Azure portal had already blocked GPv1 creation; the ARM block closes the last programmatic path. Full retirement is October 13, 2026: remaining GPv1 accounts auto-migrate to GPv2, which may result in higher billing costs and loss of legacy pricing commitments. The practical risk for large enterprises: IaC pipelines that were never updated to specify GPv2 will now fail at deployment rather than silently provisioning GPv1. Remediation: (1) inventory all GPv1 accounts, (2) update IaC templates to explicitly target GPv2, (3) test migrations in non-production, (4) migrate production accounts before October 13. Teams skipping this are not avoiding the migration — they are accepting a Microsoft-controlled forced migration on an uncontrolled timeline.

The Azure Storage .NET SDK has shipped service API version 2026-06-06, adding Blob Smart Tier support, Directory-level SAS, and User Delegation SAS for ADLS Gen2. Smart Tier SDK support means enterprises can now manage auto-tiering policies through Terraform and CI/CD pipelines rather than portal-only configuration — closing the gap between Smart Tier’s operational promise and IaC governance requirements. Directory-level SAS is a meaningful security improvement for multi-tenant Blob estate designs: it allows delegated access patterns that don’t require full-container permissions or blob-level SAS token sprawl. For teams that have been waiting on programmatic Smart Tier management before deploying it in production, this release removes the final blocker.

Current standard-tier pricing: Azure Blob Hot $0.018/GB-month, GCS Standard $0.020/GB, AWS S3 Standard $0.023/GB. Azure’s 22% advantage over S3 Standard on pure storage looks compelling, but the economics flip under real workloads. The 128 KiB minimum billable object size (July 1, 2026 for new accounts) multiplies per-object billing by up to 32× for small-object workloads in cool/cold/archive tiers. Smart Tier, designed to automate cost optimization, amplifies this risk if object size distributions are not profiled first. Additionally, Wasabi’s 2026 Storage Index confirms that roughly half of public cloud storage spend flows to non-capacity fees. Buyers using the $0.018 headline as the decision variable should model fully-loaded costs — access fees, retrieval charges, minimum size penalties, and tier-transition costs — before committing to Azure at scale.

Cloud Storage Rapid Bucket is generally available. Powered by Colossus — the same distributed storage system underlying Gemini and YouTube — it delivers more than 15 TB/s of bandwidth, 20 million requests per second, and sub-millisecond latency in a single zonal bucket. Checkpoint restores are 5× faster and writes 3.2× faster than traditional object storage; GPU blocked time drops 50% during multi-modal training. Rapid Cache (formerly Anywhere Cache) pairs with it to provide 2.5 TB/s aggregate read throughput for existing buckets with no code changes and 2.2× faster checkpoint restores via ingest-on-write. GCP is now the only hyperscaler with a GA managed object storage tier that rivals specialized parallel file systems on performance while retaining object storage economics. For enterprises paying for EFS, FSx, or third-party solutions to feed GPU clusters, Rapid Bucket warrants a direct benchmark comparison before the next infrastructure renewal.

Google Cloud Storage Smart Storage’s Object Context is generally available, adding structured, mutable, IAM-governed metadata to every object — automatically annotated by Google’s AI pipelines with labels, extracted entities, compliance signals, and image annotations at write time. A new Cloud Storage MCP Server lets AI agents read, write, and analyze GCS data using the standard Model Context Protocol, enabling agent-native data access without custom retrieval pipelines. Storage Intelligence, now used by 70% of GCP’s largest customers (each managing 50B+ objects), adds zero-configuration dashboards that instantly surface cost anomalies and integrate Security Command Center’s DSPM for security vulnerability detection. AWS responded directly with S3 Annotations on June 17. The race to make object storage semantically intelligent is now the defining storage competition of 2026.

Google Cloud NetApp Volumes now supports Flex Unified, provisioning both block (iSCSI, NVMe/TCP) and file (NFS/SMB) protocols on the same storage pool. New ONTAP-mode support lets enterprises bring existing automation — Terraform, Ansible, ONTAP APIs — directly to NetApp Volumes without rewriting toolchains. Google Cloud Backup and DR has also added agentic AI capabilities that autonomously audit backup estates and remediate coverage gaps, with GA integrations for AlloyDB and Filestore. For enterprises running SAP HANA, Oracle, or mixed-protocol enterprise applications on GCP, Flex Unified eliminates the need for separate volume types — reducing account sprawl and simplifying DR. The ONTAP-mode is the key migration enabler: it lowers friction for on-premises NetApp shops evaluating GCP and reduces migration complexity from the dominant enterprise storage platform.

Google Cloud Storage Intelligence — the unified management layer for Cloud Storage estates — has reached 70% adoption among GCP’s largest customers, each managing over 50 billion objects. New zero-configuration dashboards instantly surface cost anomalies and integrate Security Command Center’s Data Security Posture Management (DSPM) to detect critical security vulnerabilities across all Cloud Storage — no setup required. New object events and bucket activity tables in Insights Datasets enable cost analysis previously requiring custom BigQuery pipelines: optimizing bucket placement based on egress patterns and troubleshooting 429 errors. For FinOps and security teams managing large GCS estates, Storage Intelligence is consolidating the management control plane — and the 70% adoption rate suggests Google is winning on data management tooling, not just on raw storage performance.

The 2026 Wasabi Global Cloud Storage Index finds 64% of organizations deploying hybrid storage architectures specifically for AI workflows — up sharply from prior years. The more structurally important finding: roughly half of public cloud storage spend flows to non-capacity fees — egress, API calls, retrieval charges, and replication overhead — not to the per-GB rate that dominates pricing-page comparisons. For enterprise FinOps teams, this reinforces that the per-GB rate is a poor proxy for total storage cost. The vendors winning on fully-loaded TCO are those narrowing the gap between the pricing page and the actual bill — which explains why egress pricing, minimum object sizes, and retrieval fees are increasingly the focus of competitive storage announcements from all three hyperscalers.

Oracle and AWS are launching Oracle Database@AWS in 20 additional regions over the coming year. Oracle Database@Azure has also expanded to West Europe (Netherlands). For enterprises running Oracle workloads on AWS, this changes the storage decision: Oracle Exadata’s high-performance storage architecture runs within AWS’s network fabric, creating a hybrid storage pattern where Exadata storage cells coexist with S3-backed data lakes. The competitive implication: Oracle is commoditizing multicloud database deployment while hyperscalers benefit from expanded Oracle workload footprint on their infrastructure. For cloud storage architects supporting Oracle workloads, the question is no longer “Oracle on-prem vs. AWS native” — it’s “Oracle@AWS vs. Oracle@Azure,” with storage integration quality and network proximity as key differentiators.