AWS S3 Unleashed: Native Vector Storage & 50 TB Max Object Size for AI/Big Data

According to re:Invent 2025, AWS announced a broad set of enhancements to Amazon S3, including the general availability of Amazon S3 Vectors with native vector storage and querying and a 50 TB maximum object size. The production S3 Vectors release enables a single index to support up to 2 billion vectors and a bucket to store up to 20 trillion vectors, while delivering 2–3× faster performance for high-frequency queries and potential infrastructure cost reductions of up to 90%.

The update also integrates with Bedrock Knowledge Bases and OpenSearch Service, facilitating the building of AI agents and retrieval-augmented-generation systems, with early adopters listed as BMW Group, mixi, and Twilio. AWS raised the maximum S3 object size from 5 TB to 50 TB to allow massive individual files, including ultra-high-resolution video and large AI training datasets, to be stored without splitting.

Amazon S3 Batch Operations received a major upgrade, with large-scale workflows able to run up to 10× faster and each job supporting up to 20 billion objects. Additional updates cover Iceberg workloads, new Intelligent-Tiering and Automatic Replication capabilities, and extended access to FSx for NetApp ONTAP via S3-based tools.