By Don Boxley, CEO and Co-Founder, DH2i
Microsoft’s upcoming release of SQL Server 2025 marks one of the biggest shifts in the platform’s history – not because of performance tweaks or UI updates, but because AI is now a first-class citizen inside the database.
For years, developers have had to bolt AI onto SQL Server from the outside – using separate vector databases, separate LLM integrations, and separate infrastructure stacks. That fragmentation forced teams to juggle traditional DBA workflows in VMs while AI engineers built their workloads in Kubernetes or cloud-native environments.
SQL Server 2025 aims to end that divide.
SQL Server Becomes AI-Native
Microsoft added two long-requested capabilities:
1.) Native vector database support – No more exporting embeddings to an external store. Vectors are now just another data type right inside SQL Server.
2.) Built-in interoperability with multiple LLMs – OpenAI, Google Gemini, Meta’s LLaMA, Perplexity, and more. Developers can pick the model they want instead of being forced into a proprietary stack.
In other words, SQL Server is no longer just a database. It’s becoming a one-stop hub for transactional + AI workloads, whether running on VMs, bare metal, or Kubernetes.
But AI Infrastructure Has a New Problem: It Can’t Go Down
AI applications don’t sleep. Whether it’s a customer-facing AI sales agent, chatbot, or 24/7 copilot, downtime isn’t just an inconvenience – it’s a public outage.
Traditional backend systems like ERP or CRM could fail quietly – users might complain, but few saw it happen.
AI agents? They’re front office. If they crash, customers see it instantly.
High Availability (HA) Becomes the AI Safety Net
That’s where HA platform solutions step in.
In the SQL Server 2025 world, HA isn’t just about keeping a database online – it’s about keeping AI agents alive.
The ideal platform should provide automatic failover, cluster-aware intelligence, and support for spreading workloads across VMs, physical servers, or Kubernetes nodes in order to ensure:
- If one SQL node goes down, the AI app keeps running
- If you’re scaling vector-heavy AI workloads, you can distribute load across regions or clouds
- If you’re migrating from Windows VMs to containers, HA acts as the safety bridge between both worlds
As one customer put it to me recently, “AI doesn’t work unless it works 24/7 – and that means HA has never been more important.”
Will Developers Stick With SQL Server for Vectors?
It’s an open debate.
Some AI teams will insist on purpose-built vector databases. Others, especially enterprises managing cost and governance, will prefer consolidating back into SQL Server, as long as performance holds up.
The real battle will be political, not technical:
- DBAs will argue: “Let’s keep everything in SQL Server – one platform, one governance model.”
- AI engineers will counter: “We already built everything elsewhere – why change?”
HA platforms will play a critical role in mediating that debate and transition, enabling hybrid environments where vector extensions run in SQL, while legacy systems stay alive in VMs or containers during migration.
What’s Next?
The industry is moving fast from “AI as a feature” to “AI as infrastructure.”
SQL Server 2025 gives Microsoft shops a direct runway into that future, but only if their availability strategy evolves alongside it.
- AI agents can’t crash
- AI databases can’t stall
- AI platforms must span VMs-to-Kubernetes gracefully
In short, AI is only as good as the HA beneath it.
##
ABOUT THE AUTHOR
Don Boxley Jr is a DH2i Co-founder and CEO. He has more than 20 years in management positions for leading technology companies. Boxley earned his MBA from the Johnson School of Management, Cornell University.
