AWS announced several enhancements to its S3 storage platform at its recent re:Invent 2025, strengthening its object storage capabilities for adjacent markets, including vector databases, enterprise file systems, and enterprise data lakes.

The announcements include the general availability of S3 Vectors with substantially increased scale limits, new S3 integration with FSx for NetApp ONTAP file systems, cost-optimization features for S3 Tables, and expanded performance monitoring through S3 Storage Lens.

S3 Vectors Now Generally Available

AWS S3 Vectors, whcih provides through native integration with S3 object storage, rather than as a standalone database service, is now generally available.

The GA release includes substantial scale increases that make S3 Vectors a viable infrastructure for production AI workloads:

• Index capacity: Each vector index now supports up to 2 billion vectors, a 40x increase from the 50 million vector limit during preview. At the bucket level, organizations can store up to 20 trillion vectors across multiple indexes within a single vector bucket. This scale eliminates the need for index sharding and complex query federation logic that’d otherwise be required to search across datasets exceeding previous limits.
• Query performance: AWS says that infrequent queries return results in under one second, while more frequent queries achieve latencies around 100ms or less.
  • The performance improvement for frequent queries makes the service suitable for interactive applications, including conversational AI and multi-agent workflows.
  • The service now returns up to 100 search results per query, up from 30 during the preview, providing more comprehensive context for RAG applications.
• Write throughput: The service now supports up to 1,000 PUT transactions per second when streaming single-vector updates into indexes. AWS positions this write performance as suitable for workloads requiring immediate searchability of new data, including scenarios with multiple concurrent writers updating the same index.

Organizations pay based on storage consumed and queries executed, with pricing that varies by AWS Region. AWS claims potential cost reductions of up to 90% compared to specialized vector database solutions, though this assertion requires careful analysis.

The cost comparison likely assumes high storage costs and low query volumes, where S3’s per-query pricing model provides advantages over the compute resources required to maintain dedicated vector database infrastructure.

Organizations with high query volumes or requirements for sub-10ms latencies may find that specialized vector databases offer better price-performance characteristics despite higher baseline costs.

S3 Vectors Integrations

S3 Vectors includes two key integrations that extend its utility within the AWS ecosystem:

• Amazon Bedrock Knowledge Bases integration: Organizations can use S3 Vectors as the vector storage engine for Bedrock Knowledge Bases, enabling RAG applications with production-grade scale and performance.
• Amazon OpenSearch integration: The OpenSearch integration enables organizations to use S3 Vectors as their vector storage layer while leveraging OpenSearch for search and analytics capabilities. This separates storage economics from compute requirements, potentially reducing costs for organizations that need both vector search and traditional search/analytics capabilities.

The service has also expanded from five AWS Regions during preview to 14 Regions at general availability, improving geographic coverage for organizations with data residency requirements or multi-region architectures.

S3 Integration with FSx for NetApp ONTAP

AWS announced the capability to access data stored in FSx for NetApp ONTAP file systems through S3 protocols. This integration addresses a specific enterprise challenge: organizations with significant data investments in NetApp ONTAP or other NAS systems often struggle to leverage cloud-native AI, ML, and analytics services that expect data in S3-compatible object storage.

The integration enables several use cases without requiring data migration from ONTAP file systems:

• AI application development: Organizations can augment generative AI applications with Amazon Bedrock Knowledge Bases for RAG using enterprise file data that remains in FSx for NetApp ONTAP. This eliminates the data movement overhead and potential synchronization issues that would otherwise be required to make file system data accessible to Bedrock.
• Machine learning workflows: The S3 access capability enables ML model training with Amazon SageMaker using data stored in ONTAP file systems. This is particularly relevant for organizations with existing ML data pipelines built around file system protocols who want to adopt cloud-native ML services without rearchitecting their data infrastructure.
• Analytics integration: The capability extends to Amazon QuickSight for AI-powered business intelligence and third-party services integrated with S3. Organizations can run analyses using S3-based cloud-native applications against file data that continues to reside in FSx for NetApp ONTAP.

S3 Tables: Intelligent Tiering & Replication

AWS announced two new capabilities for S3 Tables, its managed table storage service for the Apache Iceberg format: an Intelligent-Tiering storage class that automatically optimizes costs based on access patterns, and replication support to maintain consistent Iceberg table replicas across AWS Regions and accounts.

S3 Tables Intelligent-Tiering Storage Class

The Intelligent-Tiering storage class for S3 Tables automates cost optimization by moving data between three low-latency access tiers based on observed access patterns:

• Frequent Access tier: The default tier for actively accessed data with standard S3 Tables performance characteristics.
• Infrequent Access tier: AWS claims 40% lower cost than Frequent Access, with data automatically transitioning to this tier after 30 days without access.
• Archive Instant Access tier: AWS claims a 68% lower cost than Infrequent Access, with data transitioning after 90 days without access. Despite the “archive” designation, this tier maintains instant access capabilities without retrieval delays.

All three tiers provide low-latency access, distinguishing this approach from traditional archival storage that requires retrieval operations. The transitions occur automatically without application changes or performance impact, according to AWS.

Users should note that the cost reductions are relative to the Frequent Access tier pricing, not absolute cost figures, and the actual savings will depend on data access patterns and the percentage of data that ages into lower-cost tiers.

The Intelligent-Tiering implementation includes optimization for table maintenance operations. Compaction processes automatically operate only on data in the Frequent Access tier, optimizing performance for actively queried data while reducing maintenance costs by avoiding unnecessary processing in lower-cost tiers. Snapshot expiration and unreferenced file removal operations similarly operate without affecting data access tiers.

Users can specify Intelligent-Tiering as the storage class when creating new tables, or configure it as the default storage class at the table bucket level. Existing tables default to the Standard storage class and require explicit migration to Intelligent-Tiering.

S3 Tables Replication

The new replication capability addresses challenges organizations face when maintaining Iceberg table replicas across AWS Regions or accounts.

Previously, organizations had to build custom architectures to track updates, manage object replication, and handle metadata transformations when synchronizing Iceberg tables across regions or accounts.

The native replication support automates these operations, maintaining consistent table replicas without manual synchronization. AWS has not provided detailed specifications for replication lag, consistency guarantees, or conflict-resolution mechanisms.

The replication feature is particularly relevant for organizations implementing disaster recovery strategies, running multi-region analytics workloads, or sharing data across AWS accounts.

S3 Storage Lens: Enhanced Metrics & Analysis

AWS expanded S3 Storage Lens with three new capabilities: performance metrics, support for analyzing billions of prefixes, and direct export to Amazon S3 Tables. S3 Storage Lens provides visibility into storage usage patterns and optimization opportunities across S3 infrastructure.

Performance Metric Categories

S3 Storage Lens now includes 8 new performance metric categories to identify and resolve performance constraints. These metrics are available at organization, account, bucket, and prefix levels, enabling granular analysis of storage performance characteristics.

AWS provides one specific example of the performance insights available: the service identifies small objects in buckets or prefixes that can slow down application performance.

Amazon suggests two mitigation approaches: batching small objects or using the Amazon S3 Express One Zone storage class, which is optimized for high-performance small-object workloads.

The performance metrics require enabling the S3 Storage Lens advanced tier when creating new dashboards or editing existing configurations. The advanced tier includes additional costs beyond the standard Storage Lens capabilities, which users should factor into their monitoring and optimization budgets.

Prefix Analysis at Scale

The capability to analyze billions of prefixes addresses scalability limitations in previous Storage Lens versions. Users with deeply hierarchical prefix structures or a large number of distinct data paths can now gain visibility across their entire S3 namespace.

This is particularly relevant for organizations that use prefixes to implement multi-tenancy, data partitioning, or complex data organization schemes.

S3 Tables Export

The direct export capability to Amazon S3 Tables allows organizations to perform more sophisticated analysis of Storage Lens metrics using SQL query engines compatible with the Apache Iceberg format.

Analysis

AWS S3 has long been the de facto industry standard for object storage, fast becoming central to how AI workflows manage data. The enhancements announced at re:Invent acknowledge that reality, improving the capabilities of AWS-native S3 for AI and observability.

The highlights of these announcements are the S3 vector search and the enhanced integration with FSx for NetApp ONTAP.

By embedding vector search, enterprise file system integration, and automated optimization directly into S3 rather than requiring separate specialized systems, AWS reduces architectural complexity and TCO for organizations building on AWS infrastructure.

The FSx for NetApp ONTAP integration with S3 protocols addresses legitimate enterprise challenges around cloud migration and hybrid infrastructure, enabling organizations to leverage cloud-native AI and analytics services against enterprise file system data without forced migration timelines. It’s a strong update that also provides greater differentiation for NetApp’s cloud-based ONTAP offerings, which remain a competitive advantage for the storage company.

These enhancements ultimately strengthen AWS’s competitive position in the AI infrastructure space while extending S3’s addressable use cases beyond traditional object storage workloads.

Organizations building production AI applications, migrating enterprise file systems to the cloud, or operating large-scale analytics infrastructure should evaluate these capabilities as potentially significant improvements to AWS’s storage platform that may reduce costs, simplify architectures, and improve operational efficiency when properly matched to specific workload requirements.

Competitive Outlook & Advice to IT Buyers

These sections are only available to NAND Research clients and IT Advisory Members. Please reach out to [email protected] to learn more.