Amazon S3 Vectors is generally available, making S3 the first cloud object store with native vector indexing and search. Each vector index supports up to 1 billion vectors with sub-100ms query latency. AWS claims 90% lower cost than specialized vector databases for RAG workloads by eliminating the compute layer required to run a separate vector DB service — queries run inside the S3 storage engine. In four months from the July 2025 preview, customers created 250,000+ vector indexes and ingested more than 40 billion vectors. AWS is positioning S3 Vectors as “complementary” to specialized vector databases, but the cost argument puts direct pressure on Pinecone, Weaviate, and Milvus. For enterprise AI architects building RAG pipelines on AWS, S3 Vectors warrants a direct cost comparison against any existing vector database contract before renewal.

AWS slashed S3 Express One Zone pricing: 31% storage reduction and 85% GET request cost reduction — the two most significant barriers to enterprise adoption of the high-performance single-AZ storage class. Separately, Express One Zone added S3 Inventory support (April 2026), enabling daily or weekly object reports with metadata and encryption status. Previously Express One Zone had no object-level reporting, making cost attribution for AI training workloads impossible. The combined effect closes both the cost and governance gaps. For workloads requiring sub-10ms GET latency — AI inference, model serving, real-time feature stores — Express One Zone’s updated economics should be re-evaluated against S3 Standard and any managed file system alternatives currently used for latency-sensitive access.

Amazon S3 Metadata expanded to support all existing objects in S3 buckets — not just objects written after enabling Metadata. Live inventory tables provide a continuously updated view of every object; journal tables capture object-level events and annotation changes in near real-time for event-driven workflows. All table types are queryable via SQL through Amazon Athena. For FinOps teams managing large S3 estates with manual Inventory exports and Athena queries, S3 Metadata provides a live, always-current alternative. The annotation table integration also sets up S3 Annotations — the AI metadata layer announced at AWS Summit New York (June 17) — to be queryable alongside standard object metadata from a single interface.

Azure Storage Actions reached general availability, providing a fully managed platform for automating data management tasks across Azure Blob Storage and Data Lake Storage Gen2. Define policy-based rules that trigger actions — tagging, tiering, deletion, compliance flagging — based on object metadata conditions without building or maintaining custom Azure Functions chains. Storage Actions completes Azure’s automated storage management trifecta: Smart Tier handles auto-tiering, Storage Discovery provides estate visibility, and Storage Actions handles policy execution — all GA in Q2 2026. For enterprises currently running custom lifecycle automation on Azure Blob, the migration path to Storage Actions eliminates infrastructure maintenance overhead while adding auditable, policy-defined governance. Combine with SDK service version 2026-06-06 for programmatic rule management in Terraform.

The Azure Storage .NET SDK shipped service API version 2026-06-06, enabling Blob Smart Tier management, Directory-level SAS, and User Delegation SAS for ADLS Gen2 through code. Smart Tier has been GA since April 2026 but previously required portal-only configuration — blocking IaC-governed deployment and Terraform-managed Smart Tier policies. Directory-level SAS is a meaningful security improvement for multi-tenant Blob designs: it allows scoped, delegated access without full-container permissions or object-level SAS token sprawl. For teams that have been waiting to deploy Smart Tier in production environments governed by IaC, the SDK update removes the final blocker. Important caveat: Smart Tier’s 128 KiB minimum object size monitoring fee, and the July 1 billing floor for cool/cold/archive tiers, makes object-size profiling a prerequisite before enabling Smart Tier at scale.

Azure Storage Discovery, a fully managed service providing enterprise-wide visibility into Azure Blob Storage data estates, is now available. Discovery automatically catalogs storage accounts, containers, and object metadata across an entire Azure subscription or management group — surfacing orphaned data, anomalous access patterns, and cost attribution gaps. For large enterprises with distributed Azure estates accumulated over years, Storage Discovery provides the inventory baseline needed to execute the GPv1 migration before the October 13 deadline. Most organizations lack an accurate count of their Azure storage accounts — Storage Discovery generates that inventory automatically and keeps it current, making it a prerequisite for any serious Azure storage governance program in H2 2026.

Google Cloud Storage launched a fully managed remote MCP server (GA, June 3, 2026), enabling AI agents to read, write, and analyze GCS data using the open Model Context Protocol without custom integration code. The remote server requires zero infrastructure deployment — agents point to the managed GCS MCP endpoint and authenticate through IAM. Every agent action logs to Cloud Audit Logs. Palo Alto Networks, Airwallex, and Snap are already running production workloads: Snap reduced job investigation time from 30 minutes to 30 seconds by giving agents direct GCS access to Flink and Spark job specs and metrics at scale. AWS has no native S3 MCP server. Azure Blob has no equivalent. For enterprise AI platform teams, GCS is now the only hyperscaler object store with a GA, managed, IAM-governed MCP interface — a significant architectural advantage in agentic deployment velocity.

Google Cloud Storage Storage Intelligence added Activity Insights to Insights Datasets — a feature that automatically delivers a query-ready BigQuery index of an organization’s entire GCS estate, updated daily for metadata and frequently for activity signals. Platform administrators can monitor how unstructured data is accessed, moved, and modified across all buckets without writing custom export pipelines. Storage Intelligence is used by 70% of GCP’s largest customers (each managing 50B+ objects). The Activity Insights addition addresses the specific operational gap of tracking real-time data movement — useful for compliance teams investigating access anomalies, FinOps teams optimizing egress patterns, and data engineers debugging pipeline behavior. No configuration required beyond enabling Storage Intelligence.

Google repriced two multi-region cold storage tiers: Nearline moved from $0.010 to $0.015/GB-month (+50%), Archive dropped from $0.004 to $0.0024/GB-month (-40%). The repricing signals a deliberate repositioning — Nearline is pushed upmarket as a higher-value tier for monthly-access data, while Archive becomes more competitive against AWS Glacier Deep Archive ($0.00099/GB) and Azure Archive ($0.00099/GB). Enterprises using Nearline as a default cold landing zone without profiling access frequency will see meaningful cost increases. The Archive cut makes GCS Archive the strongest option for large-object, truly cold data in multi-region configurations. For storage architects on GCP: audit Nearline bucket access logs, move rarely-touched objects to Archive, and recalculate TCO for any cold-tier commitments made against the prior pricing.

The 2026 Wasabi Global Cloud Storage Index finds that roughly half of public cloud storage spend flows to non-capacity fees — egress, API calls, retrieval charges, replication overhead — not the per-GB headline rate. Real-world spend runs 2–5× higher than the storage rate alone. 64% of organizations are now deploying hybrid storage architectures specifically for AI workflows, up sharply from prior years. The practical implication: the per-GB rate is a poor decision variable for cloud storage vendor selection. The vendors capturing enterprise confidence are those narrowing the gap between pricing page and actual bill — which explains why egress, minimum object sizes, retrieval fees, and auto-tiering costs are dominating competitive storage announcements. For FinOps teams: model fully-loaded costs (access fees, egress, replication, minimum-size penalties) before any storage architecture commitment, not after.

GCS’s GA MCP server, compatible with Anthropic’s Claude and Google’s ADK, establishes MCP as the preferred integration pattern for AI agent-to-object-storage connectivity. The standard matters because it decouples agent code from cloud-specific storage SDKs — an agent built against MCP can swap the underlying storage provider by updating the endpoint, not the agent. AWS and Azure have no native MCP server for their respective object stores as of late June 2026. The hyperscaler that leads on MCP-native storage integration gets embedded in agent deployment patterns that will be hard to unwind. GCS has a 6–12 month head start on managed MCP-to-object-storage. Monitoring whether AWS announces a native S3 MCP server — likely at re:Invent 2026 — is now a priority signal for enterprise cloud strategy teams.

Oracle and AWS are launching Oracle Database@AWS in 20 additional regions. Oracle Database@Azure has also expanded to West Europe (Netherlands). For enterprises running Oracle workloads on AWS, the storage decision shifts: 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 expand Oracle workload footprint on their infrastructure. For cloud storage architects supporting Oracle estates, 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 the differentiators. Evaluate Exadata storage cell proximity to your primary S3 or Blob endpoints before committing to a region.