DynamoDB Now Does Vector Search
Originally published on Build With AWS . On August 5, 2026, AWS made vector search generally available in Amazon DynamoDB. The headline promises single-digit millisecond searches, 99%+ recall, and a design that reaches trillions of vectors. The reaction across engineering forums split in two within hours. One group read it as the end of a product category, with…
On August 5, 2026, Amazon announced that Amazon DynamoDB now supports vector search, allowing users to perform searches based on semantic similarity rather than keyword matching. This feature promises single-digit millisecond search times and 99%+ recall rates, making it suitable for trillions of vectors. However, the announcement sparked debate among engineering forums. Some saw it as the demise of other products, while others focused on the need to manage query scopes for performance and scalability.
Vector search works by converting text into a numerical representation, known as an embedding, using machine learning models. Similar pieces of text have embeddings that are close together in a high-dimensional space. DynamoDB stores these embeddings as lists of numbers, each representing a dimension. The number of dimensions can range up to 4,096.
To find similar items, users convert their search queries into embeddings and then compare these embeddings to those stored in DynamoDB. This comparison is done using various distance functions, such as cosine distance, Euclidean distance, and dot product. Cosine distance focuses on the angle between the vectors, while Euclidean distance measures the straight-line distance between them.
Dot product, on the other hand, considers both the direction and magnitude of the vectors. AWS recommends using cosine distance as the default, but dot product can be useful when the magnitude of the vectors carries meaning.
It is essential to normalize the embeddings to unit length if using dot product, as this ensures that only the direction of the vectors matters in the comparison. Failure to normalize can lead to incorrect results, especially in multi-tenant applications where security is a concern. DynamoDB does not introduce new data types or change the item schema, but users may still need to adjust settings to optimize performance.
Written by urgent.news from Dev.to's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
