AWS Launches Native Vector Search for DynamoDB, Scaling to Trillions of Embeddings Without a Separate Database

Overview

Amazon Web Services has added native vector search to Amazon DynamoDB, letting applications store vector embeddings alongside their existing operational data and query them directly rather than replicating that data into a separate vector database, according to an AWS blog post published August 5, 2026 and credited to AWS’s Esra Kayabali. The new SearchVectors API queries across up to 4,096 dimensions and supports Euclidean, Cosine, and Dot product distance functions, InfoQ reported, confirming the dimension limit and distance functions AWS listed in its own announcement.

What We Know

• AWS says the feature lets developers “store vector embeddings alongside your operational data in DynamoDB and run similarity searches directly against that data, without replicating it to a separate vector store,” according to the AWS announcement.
• AWS says the feature delivers single-digit millisecond latency with 99%+ recall and scales to trillions of vectors without requiring server provisioning or maintenance, per the same post.
• Vector indexes support inline filtering on non-vector attributes and return up to 100 top-ranked results per query, according to AWS.
• Vector search runs on DynamoDB’s serverless, pay-per-request pricing model, AWS said. InfoQ reports that vector indexes are billed across three separate dimensions — data written, data processed during searches, and data stored — all metered per byte and billed per gigabyte, on top of standard DynamoDB table charges.
• AWS lists supported applications for the feature as “semantic retrieval on agentic memory, retrieval augmented generation, recommendation engines, personalized experiences, anomaly detection, and more,” according to the AWS blog post.
• The capability works with embeddings from Amazon Bedrock Titan Text Embeddings, Cohere Embed, and OpenAI’s embedding models, according to AWS and independently confirmed by InfoQ.
• AWS says vector search is available in all commercial AWS Regions and AWS GovCloud (US) Regions, according to the AWS announcement; InfoQ reports the feature supports both DynamoDB’s Standard and Standard-IA table classes.
• InfoQ reports AWS plans to extend vector search support to ExtendDB, the open-source, DynamoDB-compatible adapter AWS previously open-sourced in May, for local development and self-managed deployments.

What We Don’t Know

• Neither source discloses specific dollar figures for the per-byte, per-gigabyte charges tied to vector index writes, searches, or storage — only the three-part billing structure itself, per InfoQ.
• Neither source gives a timeline for when ExtendDB will gain support for the new vector search API.

Analysis

The change targets a specific architectural pain point AWS itself describes: teams building retrieval-augmented generation pipelines or AI agent memory on DynamoDB previously had to replicate embeddings into a dedicated vector database and keep the two systems in sync. Folding vector indexes into DynamoDB’s existing serverless, pay-per-request model — with the same table classes, region footprint, and billing mechanics as the rest of the service — removes that second system from the architecture rather than adding a new one alongside it.