Joey on SQL Server

From Fabcon to MVP Summit: The Biggest SQL Server and Fabric Updates

Recent SQL Server 2025, Azure SQL, SSMS 22 and Fabric announcements highlight new event streaming and vector search capabilities, plus expanding monitoring and ontology tooling -- with tradeoffs in platform packaging.

• By Joey D'Antoni
• 04/07/2026

First, the Microsoft-sponsored Fabcon/SQLCon event in Atlanta. Then came the MVP Summit in Redmond. Both events led to many discussions about Fabric and SQL Server. SQL Server 2025 became generally available (GA) last November at Ignite in terms of new announcements and support. Along with announcements about database savings plans (see my other column this week). At Fabcon/SQLcon, I led a full-day preconference seminar with Matt Gordon. I also presented a session on the Well-Architected Framework for Fabric. Both sessions had strong engagement and overall the conferences were refreshing experiences to not simply talk about AI all the time.

Building on the discussions at these events, the first thing I wanted to mention is Change Event Streaming (CES). It is a new feature in SQL Server 2025, and it has been available in Azure SQL Database as well. This feature enables event-driven architectures based on table activity in the database. For example, as rows are inserted or updated in a CES table, the engine sends a JSON record to an Azure Event Hub. In the future, it may support a Kafka topic.

After introducing CES, it's helpful to see how it works in practice. Once events are streamed, an application can consume them. In the Formula One-themed demo during our precon, our app reacted when an incident slowed the field. This triggered a yellow flag. The app made a call to our "race engineer," an AI service, to decide whether to pit for tires and to provide advice to our driver. If you've been confused by the term "event-driven architecture," this is it. One frustration with configuring CES was that Azure SQL only used Entra authentication, while SQL 2025, it used a Shared Access Signature, which was difficult to configure. Now, with Cumulative Update 3 for SQL 2025, Entra access for CES is supported. Additionally, CES is now available in Azure SQL Managed Instance.

Another hero feature in SQL Server 2025 was support for vector queries. Vector data converts text into a long series of floating-point values. This functionality enables comparisons of similarity based on the geometric distance between values. In simple terms, if a user queries for "red bike" and the actual string was "crimson bicycle," the vector search would return this value, while traditional full-text indexes would not get a match. While this functionality can be game changing, and eliminate whole components in an application stack (and is, read more about it here), those float values are very large and can consume a lot of database memory.

Continuing the topic of vector queries, Microsoft Research identified the memory consumption challenge early. They developed "approximate nearest neighbor" or DiskANN indexes to reduce the memory footprint and further optimize vector search. Unfortunately, in the Azure SQL DB and SQL Server 2025 implementations, DiskANN indexes could not be updated. These indexes also took a very long time to create. Azure SQL Database now supports updateable DiskANN indexes along with reported 5-6x performance improvements during index creation. While this feature is not yet in SQL Server 2025, I would expect to see it in a subsequent cumulative update.

SQL Server Management Studio (SSMS) has seen big improvements over the last couple of years. SSMS's first attempt at AI-assisted coding required complex setup, including configuring (what is now known as) Microsoft Foundry, which limited broad adoption amongst database professionals. SSMS 22, which launched in preview in Fall 2025, has a simpler configuration using GitHub Copilot, like VS Code. The GitHub Copilot integration in SSMS 22 is now generally available. Copilot supports two different modes in SSMS: if you aren't connected to a database, Copilot can help with general SQL questions; with a database connection, it uses the database's context to help you create queries using natural language. Another additional new non-AI feature is the ability to export query results to JSON, CSV, Excel, and XML. Finally, SSMS added the ability to group by schema for all types of SQL databases, a long requested feature.

This brings me to Database Hub in Fabric. The concept of Database Hub in Fabric originated in an earlier feature, Database Watcher, which served as a database monitoring tool that supported most of the Azure SQL stack. Database watcher stored data in Azure Data Explorer or Fabric Real-Time Analytics. Fabric Database Hub expands on Database Watcher's coverage of the Azure SQL database estate, adding Cosmos DB, Azure MySQL, Azure Postgres, Fabric Database, and on-premises SQL Server VMs running Arc. The Database Hub solution builds on Database Watcher and includes AI capabilities to help identify performance issues and database problems.

While I love the concept of a centralized cross-database platform monitoring solution, I don't love that you will need to become a Fabric customer to access the full functionality of Database Hub. This type of monitoring will nearly always fall back on the budget of an IT Infrastructure or DBA team, which traditionally allocates most of its resources to Azure. Additionally, it limits Azure customers who might not want to use Fabric. While Fabric offers strong resources for data visualization and log analysis, very similar functionality is available in Azure Data Explorer or Log Analytics. Microsoft already has several products in this space, with overlapping features in Azure Arc and Azure Monitor, as well as third-party tools with similar functionality. I think the concept is good (the database tooling space has seen greatly reduced investment in the last decade, due to consolidation and private equity), but I think tying this Fabric greatly limits the total available market of customers for this service.

The Fabric team also made several other announcements during the conference, mostly about its ontology and AI product, Fabric IQ. Fabric IQ was launched at Ignite last fall, but that announcement lacked many details. The new product unifies data across Fabric's data storage platform, OneLake, and organizes it using a combination of AI agents and analytics, ensuring consistent meaning and context. Fabric IQ introduces Ontologies (still in preview) to provide an enterprise vocabulary and semantic layer that unifies meaning across domains. This concept of ontology has become trendy in business intelligence tooling in recent years. recent years.

It is always refreshing to attend large, busy conferences and speak to large, engaged audiences. The recent announcements around SQL Server and Fabric show Microsoft’s continued investment across the data platform. SQL remains a strong business for Microsoft, but it needs to continue investing to stay top of mind with developers, architects, and IT executives, especially given that open-source alternatives offer lower upfront costs. Having SQLcon as part of Fabcon is a good show of Microsoft’s continued investment in the platform, I hope they continue to maintain those investments.