Azure SQL Database vector search and DiskANN indexes reach general availability

Microsoft moved vector search in Azure SQL Database to general availability. The feature set includes:

- a native vector data type;

- VECTOR_DISTANCE functions for exact similarity;

- approximate nearest-neighbor search through new CREATE VECTOR INDEX and VECTOR_SEARCH syntax, backed by DiskANN indexes.

The practical upshot is that teams can build retrieval-augmented generation and semantic search in plain T-SQL. The embeddings live next to the relational data they describe.

DiskANN is a graph-based ANN algorithm originally developed at Microsoft Research. It is designed to keep most of the index on SSD rather than in RAM, which lets vector indexes scale beyond memory limits with good recall and latency. Inside SQL Database, developers get transactional consistency, existing security models, row-level permissions and backups for vector data. That spares them the synchronization pipelines needed to keep an external vector store in step with the source of truth. WindowsForum's GA write-up also details production limits developers should check before migrating.

Competitively, this continues the absorption of vector search into general-purpose databases:

- Postgres has pgvector.

- Oracle, MongoDB and Elastic all ship vector indexes.

- AWS offers vector support in Aurora and OpenSearch.

That pressure falls on standalone vector database vendors such as Pinecone and Weaviate, whose pitch increasingly depends on scale and specialized features.

Community reaction was positive, with developers praising RAG in plain T-SQL. The caveats are performance at very large scale, index build times, and how limits compare with dedicated engines. For the huge installed base of SQL Server shops, though, GA removes a major reason to add another system to the stack.