Microsoft beefs up SQL Server 2025 for AI-driven applications

Microsoft is moving SQL Server 2025 into public preview, adding new features to help enterprises develop AI-driven applications.

One of the key new capabilities is vector search, which will enable enterprises to perform semantic search on their data using a natural language interface. Here, SQL Server is catching up with competitors such as MongoDB, Google, Oracle, and AWS, which have all added vector search capabilities to their databases over the last 18 months. Microsoft itself added vector search support to Cosmos DB a year ago.

Although it’s late to the party, SQL Server 2025’s addition of vector search is still significant because the database is so widely used in the enterprise, said Matt Aslett, director at ISG Software Research.

SQL Server 2025 uses the same underlying code as Cosmos DB for this: a suite of scalable approximate nearest neighbor search algorithms with support for real-time changes called DiskANN. Both databases also support retrieval augmented generation (RAG) as well, Microsoft said, adding that SQL Server 2025 will support seamless integration with popular AI frameworks such as LangChain, Semantic Kernel, and Entity Framework Core for application development.

Playing catch-up

Microsoft is introducing other capabilities to SQL Server 2025 to facilitate the development of AI-based applications, including JSON support, a Python driver, and change event streaming.

The JSON support could help enterprises enrich data from external sources, especially semi-structured and unstructured data, in turn helping them build more accurate generative AI-based applications, said Aslett.

But these features are “very 2023”, said dbInsight chief analyst Tony Baer. “Microsoft is several years behind AWS, Google, Oracle, and others when it comes to multimodal data, streaming, and Python support,” he said.

But there is one area in which SQL Server 2025 is ahead of the competition, said Aslett: Model management.

SQL Server packages model definitions right inside T-SQL, the language used by the database. In a blog post, Priya Sathy, partner director of product management at Microsoft’s SQL division, said, “The model definition can connect to AI services such as Azure AI Foundry, Azure OpenAI, OpenAI, and Ollama among others: Models are all accessed through REST APIs allowing developers to deploy any model securely isolated from the SQL Server engine.”

Other updates to the database includes performance enhancements for query processing and security features, such as optimized locking to reduce lock memory consumption and transaction ID locking to minimize blocking for concurrent transactions.

In addition, Microsoft said it is introducing a new Standard Developer edition to allow developers to build and test against the Standard edition without incurring license costs.