The CNCF is right, just not about everybody

I spent some time reading the latest CNCF annual survey this week. It’s a solid report, well put together, and full of useful data. But as I was reading it, I couldn’t drop a recurring internal dialog… this is a bit like asking farmers if they own tractors.

If you’re responding to a CNCF survey, you’re already in the cloud native ecosystem. You’ve already accepted containers. You’re probably already running Kubernetes somewhere. You might even enjoy it. So when the headline says that most container users are running Kubernetes in production, my reaction isn’t surprise. It’s more a quiet “of course they are.”

That doesn’t make the data wrong. It just means we need to be honest about what it actually represents.

What this survey really shows is not where the enterprise market is, but where the cloud native conversation currently ends. And that distinction matters more than we tend to admit.

Inside the CNCF bubble, Kubernetes is now described as “boring,” and that’s meant as praise. Stable. Mature. Predictable. And to be fair, that’s largely true if you look purely at the core orchestration layer. Kubernetes itself has grown up.

But read between the lines, and the same survey quietly tells a very different story.

The biggest challenges organizations report are no longer technical. They’re cultural. Alignment between teams, ways of working, ownership boundaries, and expectations. If your organization still thinks in terms of packaged enterprise software, regulated releases, change advisory boards, and a generally risk-averse view of operations, that’s usually where Kubernetes starts to feel awkward rather than empowering.

GitOps is another good example. It’s often talked about as if it’s table stakes, or even a prerequisite. But the data tells a more grounded story. Even within this already self-selected audience, GitOps adoption is effectively zero at the early stages and only really shows up at the most mature end of the spectrum. That tells you something important. GitOps isn’t a starting point. It’s an outcome.

Then there’s AI, which is where things get especially interesting.

There’s a lot of noise right now about Kubernetes becoming the AI platform. And again, inside this ecosystem, that’s broadly true. But if you look at how organizations are actually using AI, the picture is far less dramatic. Most aren’t training models. Most are consuming them from a cloud service provider. Only a small minority are operating anything close to what you’d call large self-hosted models.

What’s really happening is that AI workloads are entering organizations through vendors, internal experiments, and packaged platforms. And those workloads come with very real infrastructure consequences. Cost. Scaling. Governance. Operational strain. The survey even calls out concerns around machine-driven usage putting stress on already fragile infrastructure.

That’s not a future problem. It’s already happening.

And this is where the survey becomes more interesting for what it *doesn’t* show.

If this is the reality inside the CNCF bubble, imagine what it looks like outside it.

Most enterprises don’t think of themselves as cloud native. They don’t attend KubeCon (remember, only 17,000 people attend KubeCon, likely representing around 5,000 organizations). They don’t want to assemble an ecosystem of CNCF projects. They run commercial software, internal line-of-business applications, and systems that need to be stable long before they need to be elegant.

But Kubernetes is still arriving anyway.

It’s arriving because vendors are shipping on it. Because infrastructure teams are standardizing on it. Because AI platforms assume it. And slowly, often without a formal decision, it becomes part of the furniture.

That’s where I think the real gap is.

The CNCF survey shows what Kubernetes looks like after an organization has already crossed the line and invested heavily in skills, tooling, and cultural change. What it doesn’t show is the much larger group of organizations that are about to inherit Kubernetes without wanting to become cloud native purists or platform engineering shops.

Those organizations don’t want a platform journey. They want Kubernetes to behave like infrastructure. Something that can be operated, secured, audited, supported, and if needed, handed over or unwound without drama.

They want fewer moving parts, not more. Fewer bespoke workflows, not endless YAML. And above all, they want optionality. The ability to bring things in-house, outsource them, or change direction without rewriting the entire operating model.

So my takeaway from the survey wasn’t “Kubernetes has won.” That battle is already over.

My takeaway was that the next phase isn’t about adoption. It’s about containment.

How do you stop Kubernetes from turning into a fragile dependency that only a small group understands? How do you let teams use it productively without forcing everyone to become an expert? How do you say yes to modern workloads, including AI, without quietly signing up for years of internal platform debt?

Those are the questions most enterprises are about to start asking, even if they don’t realize it yet.

And interestingly, the CNCF data already hints at the answers. Standardization beats novelty. Operational maturity beats cleverness. And boring, well-governed infrastructure turns out to be a competitive advantage.

Which, honestly, is a much more useful place for this industry to be than another year of arguing about which tool is “best.”