Amazon S3 Vectors: AWS adds native vector storage to S3

2025-07-15 · Publisher : AWS · Domain : Cloud & DevOps

AWS launches Amazon S3 Vectors in preview, an S3 bucket type that stores and queries vectors with sub-second responses. AWS claims up to 90% lower cost for uploading, storing and querying vectors.

### Background On 15 July 2025, AWS announces the preview of Amazon S3 Vectors, presented as the first cloud object store with native support for large vector datasets. The service adds "vector buckets" to S3, with dedicated APIs to store, access and query vectors, with no infrastructure to provision. ### Key facts - AWS says it can reduce the total cost of uploading, storing and querying vectors by up to 90%, with sub-second response times (vendor figures). - Up to 10,000 vector indexes per bucket, each index able to hold tens of millions of vectors. - Metadata (dates, categories, preferences) can be attached to vectors to filter queries. - Native integration with Amazon Bedrock Knowledge Bases and SageMaker Unified Studio for retrieval-augmented generation (RAG) applications. - Integration with Amazon OpenSearch Service: rarely queried vectors stay in S3 and move to OpenSearch as demand grows. - The service automatically optimizes vector data as it is written, updated and deleted. ### Why it matters Generative AI applications accumulate millions of embeddings (numerical representations of documents, images or sounds). A dedicated vector database can be expensive when data is rarely queried. By putting vector storage in S3, AWS offers a cost level close to object storage for cold data. For an IT services firm, this changes the trade-off between a specialized vector database and object storage. ### Takeaway Vector storage becomes a native S3 function: generative AI reshapes basic storage services.

Tags : AWS, Données, IA, Hyperscaler