# What's new in SQL Server 2025 and Azure SQL Databases - Christian Peeters - NDC Oslo 2026

## Executive summary

SQL Server 2025 and Azure SQL Databases are evolving beyond simple storage engines to become active components in AI-driven architectures. Key innovations include native support for vector data types, enabling semantic search and Retrieval-Augmented Generation (RAG) directly within the database. Furthermore, developer productivity is significantly boosted by enhanced T-SQL capabilities, such as calling external REST endpoints, advanced JSON indexing, and robust pattern matching using Regular Expressions (RegEx).

## Key takeaways

- Native Vector Database Capabilities: The introduction of the `vector` data type allows SQL Server to store embeddings (floating-point representations of meaning) for advanced AI workloads. This enables semantic search and RAG by calculating similarity (e.g., using cosine distance) directly within SQL, drastically simplifying complex data pipelines.
- Enhanced Developer Productivity via T-SQL: Developers can now perform complex operations directly in SQL, including calling external REST endpoints (HTTP calls) to interact with external services, improving system integration without leaving the database layer.
- Advanced Data Handling and Pattern Matching: Improvements include native `JSON` data type support (allowing indexing on semi-structured data properties), enhanced Regular Expressions (RegEx) for data sanitization and masking, and Fuzzy Search algorithms (like Jaro-Winkler) to handle user spelling errors.
- Real-Time Event Streaming: The addition of Change Event Streaming allows every `INSERT`, `UPDATE`, or `DELETE` operation to stream an event (including the old and new object versions) to external services like Azure Event Hubs, facilitating real-time auditing and event-driven architectures.

## Technical details

- Vector Search and RAG: SQL Server 2025 introduces the `vector` data type and functions like `vector distance` and `vector search`. This enables RAG by storing embeddings and calculating similarity metrics (cosine, Euclidean, dot product) to find semantically related data, moving beyond keyword matching. Indexing is available for performance optimization.
- JSON Data Type and Indexing: The native `JSON` data type allows storing semi-structured data as binary objects. Crucially, it enables creating indexes on specific properties within the JSON structure (e.g., using `JSON_CONTAINS`), providing massive performance gains over previous methods.
- T-SQL REST API Integration: SQL Server can now execute HTTP calls and REST API calls to external endpoints. This allows the database to interact with external services, such as fetching data or performing actions, directly within T-SQL statements.
- Pattern Matching and Data Quality: New capabilities include advanced Regular Expressions (RegEx) for data extraction and sanitization (e.g., masking emails or extracting model numbers). Fuzzy Search algorithms (e.g., Jaro-Winkler) address data quality issues caused by user spelling errors.

## Practical implications

- Build engineers can design more intelligent, data-centric applications by embedding AI logic (RAG) directly into the database layer, reducing reliance on external microservices for core data querying.
- Implementing event-driven architectures is simplified through Change Event Streaming, allowing immediate triggers for downstream services (e.g., CI/CD pipelines, audit logs).
- The ability to call external REST APIs from SQL simplifies system integration, allowing the database to act as a central orchestrator for data retrieval and processing.
- Improved data quality and validation can be enforced at the database level using advanced RegEx and fuzzy matching.

## Topics

SQL Server 2025, Azure SQL Databases, Vector Databases, RAG (Retrieval-Augmented Generation), T-SQL, JSON Data, Regular Expressions, Fuzzy Search, Event Streaming, NDC Conferences, Azure AI Foundry

Source: https://www.youtube.com/watch?v=rELQksklD-0