AWS Introduces Native Vector Search for DynamoDB

AWS has announced native vector search support for Amazon DynamoDB, enabling developers to store vector embeddings directly in their existing tables and run approximate nearest-neighbor (ANN) queries without a separate vector database. The new capability, introduced this week, allows applications to perform semantic search, retrieval-augmented generation, and agentic memory tasks using the same fully managed NoSQL service. This marks a significant step for DynamoDB, which previously required data duplication into purpose-built vector stores. Until now, teams building AI-powered features on DynamoDB had to maintain complex data pipelines to sync data between their primary database and a separate vector engine, increasing both architectural complexity and operational overhead. With native vector search, embeddings and application data now live in the same DynamoDB table, eliminating the extra synchronization layer. The feature is built on a new DynamoDB index type that supports up to 4,096 dimensions, with distance functions including Euclidean, Cosine, and Dot product. Developers can choose any embedding model—such as Amazon Bedrock Titan Text Embeddings, Cohere Embed, or OpenAI text embedding models—and create a vector index with the required dimensions and distance function, then query it using the new SearchVectors API. The indexes scale horizontally with no storage limits, and AWS claims single-digit millisecond latency even at trillions of vectors. Inline filtering lets users combine semantic and scalar queries in a single request, making the tool flexible for production workloads. Vector indexes are billed on three dimensions: data written, data processed during search, and data stored, with all metering calculated per byte and charged per GB, in addition to standard DynamoDB table charges. Cost optimization strategies, such as using lower dimensions, minimal index projections, and selective partitioning, can significantly reduce expenses, according to AWS solution architects. The launch has sparked strong discussion in the developer community. Some practitioners argue AWS is 'too late to the party,' since many other databases have already shipped vector support, while others welcome the simplicity of keeping application data and vectors together. One user noted that while DynamoDB offers very low latency, it may be considerably more expensive than S3 for large-scale vector workloads. AWS plans to extend the feature through ExtendDB, a DynamoDB-compatible adapter for local development and self-managed deployments. Vector search is available in all AWS regions where DynamoDB operates, across both Standard and Standard-IA table classes. As vector search becomes a standard database feature, this move positions DynamoDB to compete directly with specialized vector databases and reinforces the broader industry trend toward unifying operational and AI workloads.

Turn any website into clean JSON.

Ready-made scrapers & AI agents

Try Apify