Azure Cosmos DB Now Does Vector Search Natively
Azure Cosmos DB just made a significant move in the AI infrastructure space: vector indexing and search is now generally available. That means you can store embeddings alongside your operational data and run similarity searches directly in Cosmos DB, without exporting to a separate vector database like Pinecone or Weaviate.
The feature is designed for large vector datasets paired with ultra-high throughput inserts and searches. According to Microsoft Learn documentation, the implementation can accommodate millions of queries per second with predictable, low latency. If you are already running Cosmos DB for your NoSQL workloads, this removes a major reason to add a separate vector store to your architecture.
Why This Matters
The vector database space has been heating up fast. Pinecone, Weaviate, Qdrant, and pgvector all have their fans. But they add operational overhead: another database to provision, monitor, and pay for. Cosmos DB vector search means one database for both your structured data and your embeddings.
For developers building RAG (Retrieval Augmented Generation) systems, this is particularly useful. You can keep your documents in Cosmos DB, generate embeddings, and query them in the same place. No ETL to a separate vector store, no synchronization issues between your document database and your embedding index.
The use case Microsoft is targeting is clear: AI applications that need to search across millions of vectors with low latency. Think semantic search, recommendation engines, and anomaly detection built on top of existing Cosmos DB workloads.
What You Need to Know
To use vector search in Cosmos DB for NoSQL, you will need to enable it on your container and define a vector index policy. The feature supports common embedding models and provides approximate nearest neighbor (ANN) search capabilities. Check the Microsoft Learn documentation for setup details.
One thing to note: the feature is for NoSQL (SQL API) mode. If you are running MongoDB API or other APIs, you will need to check compatibility. This is a NoSQL-first feature.
For teams already using Azure OpenAI or other AI services, bringing vector search into Cosmos DB simplifies the data layer. You can now have your documents, your embeddings, and your queries all in one place, backed by Cosmos DB’s global distribution and SLAs.