SQL Server 2025: Enterprise AI Without the Learning Curve

With the release of SQL Server 2025, Microsoft is firmly positioning its flagship database platform as an enterprise-ready AI solution. Having spent time with the preview builds and working with the feature set, I believe this release represents a significant opportunity for organizations to leverage AI capabilities and their existing investments in SQL Server infrastructure, skills, and processes.

While many cloud-native and niche vector databases have emerged to support AI workloads, SQL Server 2025 brings these capabilities to your existing operational data platform. This integration offers unique advantages that standalone AI solutions can’t match without significant integration efforts and security considerations.

T-SQL Programming Language: Familiar Skills for New AI Challenges

The most compelling aspect of SQL Server 2025’s AI capabilities is their integration directly into T-SQL. This isn’t a bolt-on solution but a thoughtful expansion of what you can do with the language developers and DBA teams already know.

Key here is that teams don’t need to learn a new programming language or platform. The same T-SQL skills they’ve used for years now extend to AI operations. AI functionality can be incorporated into stored procedures, functions, and applications with minimal changes using skills, tools, and techniques familiar to developers and DBAs.

In my experience working with clients implementing AI-based solutions, the challenge is integration with existing operational data. SQL Server 2025 eliminates that gap. Your AI features run right where your data already lives, leveraging existing security and availability features, simplifying architectures, and accelerating time to value.

Let’s examine Microsoft’s approach to integrating AI into SQL Server 2025 using a code sample they provided as part of the demos in the public preview. The code below shows an AI-enabled semantic search implementation, highlighting the platform’s new vector capabilities.

It follows a four-step process:

  1. Define an external model connection, here I’m using an Ollama embedding generation REST Endpoint. You’re free to use your model of choice.
  2. Add a VECTOR(768) column for storing embeddings to a table in your database, here it’s Product. We are also adding a column chunk this is the text representation of the values used to generate the embedding.
  3. Generate the embeddings for Products by combining product details texts into a chunk of text and passing that into the new AI_GENERATE_EMBEDDINGS function to generate the embeddings by passing them to our external model ollama. Then storing the generated embedding in our newly created embeddings column.
  4. Perform a natural language vector similarity search in which a conversational query about a red bike that doesn't cost much is converted to a vector embedding using AI_GENERATE_EMBEDDINGS and compared against product embeddings stored in our table using cosine distance using the function vector_distance.
-- Step 1: Create an External Model used for embedding generation ----------------------
CREATE EXTERNAL MODEL ollama
WITH (
    LOCATION = 'https://model-web:443/api/embed',
 API_FORMAT = 'Ollama',
 MODEL_TYPE = EMBEDDINGS,
 MODEL = 'nomic-embed-text'
);
GO


-- Step 2: Altering a Table to Add Vector Embeddings Column ----------------------------
USE [AdventureWorks2025]
GO
ALTER TABLE [SalesLT].[Product]
ADD embeddings VECTOR(768), chunk NVARCHAR(2000);
GO


-- Step 3: CREATE THE EMBEDDINGS (This demo is based off the MS SQL 2025 demo repository)
UPDATE p
SET 
 [chunk] = p.Name + ' ' + ISNULL(p.Color, 'No Color') + ' ' + c.Name + ' ' + m.Name + ' ' + ISNULL(d.Description, ''),
 [embeddings] = AI_GENERATE_EMBEDDINGS(p.Name + ' ' + ISNULL(p.Color, 'No Color') + ' ' + c.Name + ' ' + m.Name + ' ' + ISNULL(d.Description, '') USE MODEL ollama)
FROM [SalesLT].[Product] p
JOIN [SalesLT].[ProductCategory] c ON p.ProductCategoryID = c.ProductCategoryID
JOIN [SalesLT].[ProductModel] m ON p.ProductModelID = m.ProductModelID
LEFT JOIN [SalesLT].[vProductAndDescription] d ON p.ProductID = d.ProductID AND d.Culture = 'en'
WHERE p.embeddings IS NULL;


-- Step 4: Performing a vector similarity search ----------------------------------------
DECLARE @search_text NVARCHAR(MAX) = 'I am looking for a red bike and I dont want to spend a lot';
DECLARE @search_vector VECTOR(768) = AI_GENERATE_EMBEDDINGS(@search_text USE MODEL ollama);

SELECT TOP(4)
    p.ProductID,
    p.Name,
    p.chunk,
    vector_distance('cosine', @search_vector, p.embeddings) AS distance
FROM [SalesLT].[Product] p
ORDER BY distance;
GO

This example shows how SQL Server 2025 integrates AI directly into the database using familiar T-SQL syntax. This allows developers to implement semantic search without learning new platforms or languages. You keep the AI processing close to where the data already resides, keeping it secure, protected, and highly available.

Get your hands on this code right now and head over to my GitHub repo Ollama SQL FastStart for a full stack solution showcasing SQL Server 2025 and Vector all implemented in docker to get your development stack up and running in a single line of code.

Enterprise Grade Availability and Disaster Recovery

One of the challenges with many purpose-built vector databases is their relative immaturity in critical enterprise requirements like high availability and disaster recovery. SQL Server, on the other hand, brings a long history of hardened capabilities to support your database platform:

  • Backup and Restore - The same familiar backup and restore operations work with vector data, allowing you to leverage your existing backup strategies, including features like T-SQL Snapshot Backup.

  • AlwaysOn Availability Groups - Your AI-enabled databases can participate in availability groups, providing high availability and disaster recovery for your AI workloads. This is crucial for mission-critical AI applications that can’t afford downtime.

  • Failover Cluster Instances - For those preferring the shared storage model, FCIs continue to provide an excellent HA solution compatible with vector data.

This is particularly important as AI becomes more central to business operations. When your recommendation engine or document processing systems are directly tied to revenue or customer experiences, having mature HA/DR capabilities becomes core to any system architecture.

Enterprise Grade Security

Security remains one of the most critical considerations when implementing AI systems. SQL Server 2025 brings its comprehensive security model to AI workloads:

  • Row-Level Security (RLS) - Apply fine-grained access control to your AI data, ensuring users only see what they’re authorized to access. This is vital for data-driven applications or when dealing with sensitive information, controlling who can access exactly what information in a database, row store, or vector store.

  • Multi-layered Encryption - From TDE for data at rest to Always Encrypted for sensitive columns to TLS encrypted connections, SQL Server provides comprehensive encryption capabilities that extend to vector data.

  • Active Directory/Entra ID Integration - Leverage your organization’s existing identity management for authentication and authorization, making security management consistent with your other enterprise applications.

When we consider that AI often involves sensitive data, these security capabilities become even more critical. SQL Server’s mature security model protects embeddings and other AI assets at the same level as your traditional structured data.

DBA Operational Skill Sets

Perhaps the most overlooked advantage of using SQL Server for enterprise AI is the ability to leverage existing DBA skills and support systems:

  • Familiar Monitoring and Management Tools - The same tools DBAs use to manage traditional SQL Server workloads apply to AI-enabled databases, from Query Store to Extended Events and mature third-party monitoring solutions, like Redgate’s SQL Monitor.

  • Enterprise Support - Microsoft’s support infrastructure and a vast ecosystem of partners and consultants means you have reliable assistance when issues arise.

  • Deep Community - The SQL Server community is one of the largest and most active in the database space, with vast resources for troubleshooting, optimization, and knowledge sharing.

Wrapping Things Up

SQL Server 2025 is introducing AI capabilities to traditional enterprise data platforms (I can’t believe I just said traditional databases, but here we are). Organizations can implement AI solutions without the complexities of integrating separate systems, transferring data, or retraining their teams. Customers can bring AI into their enterprise utilizing the existing T-SQL skills of developers and DBAs, along with proven availability and disaster recovery mechanisms, enterprise-grade security, and established operational practices and not increase the technical complexity of their platforms.

While many vendors push organizations toward complete re-platforming for AI capabilities or cloud-based solutions, Microsoft’s approach with SQL Server 2025 stands out by enabling businesses to build on their existing investments and move forward with next-gen AI-based solutions. This represents an efficient path to implementing AI at scale for many enterprises.

As I continue to explore SQL Server 2025’s capabilities, I’ll be sharing more detailed technical implementations and best practices. In the meantime, you should start experimenting with the preview releases to understand how these capabilities can enhance your data platform.