A DH2i customer in Germany couldn't move its SQL Server infrastructure off Windows. Not because the technology wasn't ready. Because the team running it wasn't. "They basically said we had to wait for guys to retire before we could look at moving to Kubernetes," says Don Boxley, co-founder and CEO of DH2i. The database administrators who'd run Windows failover clustering for two decades weren't interested in relearning their jobs, and nobody was going to force them out the door over it. So the company waited.
That story explains why Kubernetes hasn't replaced Windows in the enterprise better than any technical limitation does. I've covered DH2i since I started writing for DZone 11 years ago, and the "when does Windows finally give way" conversation has been a fixture the whole time. What's changed isn't the technology. It's how trapped customers feel by their own infrastructure.
"People are now starting to feel a little more trapped with their Windows infrastructure," Boxley told me. Applications have largely already made the jump to Linux and containers. Databases haven't. That creates what Boxley calls a schism: modern application layers running on Kubernetes, sitting on top of SQL Server instances still running Windows failover clustering underneath, sometimes on versions of SQL Server that are close to a decade old.
The numbers back that up. DH2i added telemetry to its latest DxEnterprise release, version 26, and among customers who've opted in, the split is roughly 60 percent Windows, 40 percent Linux — moving, in Boxley's words, "slowly but surely." Only 20 percent of DH2i's install base is running SQL Server 2025. Most customers are still on versions older than 2017. That matters, because real Kubernetes support requires SQL Server 2022 at minimum. Only about a quarter of DH2i's customers are even on a version that makes the jump technically realistic.
So when people say Kubernetes "isn't replacing Windows," what they usually mean is SQL Server specifically — the database layer that's proven far stickier than anything running on top of it.
Boxley's read on why goes beyond technical debt, though there's plenty of that. It's generational and organizational too. DBA teams that built their careers on Windows failover clustering don't necessarily want to relearn their infrastructure around Kubernetes, and companies aren't always willing to push the issue.
There's also a money story underneath the technology story, and it's arguably the more interesting one. Boxley told Channel Insider recently that AI infrastructure spending is freeing up budget that's pushing some customers toward Linux and Kubernetes. When I asked him to square that with how slowly Windows is actually giving ground, his answer wasn't about AI workloads driving technical need. It was about AI as budget cover. "The AI became the cover, if you will, because the expenditures were so significant that you could hide two, three million dollars to get it done," he says. Modernizing legacy database infrastructure is hard to justify on its own. Attached to an AI initiative, it's an easier ask.
Broadcom's VMware licensing changes are doing something similar for a different set of customers. Boxley says the enterprises coming to DH2i because of VMware costs tend to be large, already carrying big bills, and highly motivated to move — often landing on Kubernetes, frequently on Red Hat OpenShift, in whatever cloud fits their budget.
For everyone else, DH2i's approach is deliberately incremental rather than a rip-and-replace. Customers running Windows failover clustering can move to DH2i's own clustering with roughly 15 to 20 seconds of downtime, no bigger infrastructure change required. From there, DH2i can add a Linux node to the same cluster, or a Kubernetes node, running it side by side with the existing Windows node until the customer is ready to fail over and make Kubernetes primary. Boxley says a committed customer can go from Windows to Kubernetes in about a week.
The pitch to IT teams with limited headcount is blunt: tell DH2i what you're spending on Windows Server licensing, and the company commits to cutting that by 30 percent, locked in for multiple years. "It's just money," Boxley says. "How much money are you going to save me? Then we can have the conversation."
What Boxley pushes back on is the idea that any of this needs to wait for a bigger AI roadmap to justify it. "There's no reason for customers to wait," he says. Moving off Windows and onto Kubernetes is, in his view, no longer a question of whether it's possible. It's a question of how fast a given organization wants to go.
Still, he's candid about what's holding the rest back. "The Windows infrastructure gravity is really, really strong," he says. "We want to get everybody onto a different planet, if you will. But it's tough to get out of that atmosphere."
That gravity, more than any Kubernetes roadmap, is the real story.