Kubernetes Got Boring, and That Is the Best News in Infrastructure

There is a particular moment in every technology’s life when it stops being interesting. The conference talks get less breathless. The blog posts stop explaining what it is and start explaining how to run it cheaply. Somebody writes “we moved off it” and nobody clicks. Kubernetes reached that moment a while ago, and the numbers from the CNCF’s 2025 annual survey, published this January, make it official.

82% of container users now run Kubernetes in production. In 2023 that figure was 66%. A sixteen-point jump in two years is not the shape of a technology finding its audience. It is the shape of one that has already found it and is now mopping up the stragglers.

What the number does and does not say

Read that statistic carefully, because the qualifier matters. It is 82% of container users, not 82% of all organizations. If your shop runs everything on virtual machines and has no container story, you are not in the denominator. The survey’s broader claim is the more startling one: 98% of surveyed organizations report having adopted cloud native techniques in some form, and 59% say that “much” or “nearly all” of their development and deployment now works that way.

The remaining 10% are in early stages or not doing it at all. That is a small enough minority to be a rounding error in most hiring conversations.

So the honest summary is not “everyone uses Kubernetes.” It is closer to “if you have containers in production, you almost certainly orchestrate them with Kubernetes, and most organizations now have containers in production.”

Boring is a compliment

When a technology is exciting, the people who know it are paid for their enthusiasm. When it is boring, they are paid for their reliability. The second arrangement is better for anyone planning a career longer than eighteen months.

Boring means the interfaces stopped moving. It means the failure modes are documented, the war stories have been told, and somebody has already written the runbook for the thing that broke at three in the morning. It also means the questions asked in interviews changed. Five years ago you might have been asked to explain what a Pod was. Now you are more likely to be handed a cluster that will not schedule and asked what you check first.

That shift is exactly why the Certified Kubernetes Administrator exam is structured the way it is, which is next week’s subject.

The AI wrinkle

The survey’s headline framing is that Kubernetes has become the operating system for AI workloads, and the supporting figure is that 66% of organizations hosting generative AI models use Kubernetes to manage some or all of their inference workloads.

Worth pairing that with the number nobody put in a headline: 44% of respondents do not yet run AI or ML workloads on Kubernetes at all. Both things are true. Among the organizations doing AI in production, Kubernetes is the default substrate. Plenty of organizations are still not doing AI in production, whatever the earnings calls suggest.

There is a second sobering figure in there. Only 7% of organizations deploy models daily, and 47% deploy occasionally. The gap between “we have a model” and “we ship models continuously” remains wide, and most of what closes it is unglamorous platform work.

The part that is still hard

If Kubernetes has become boring, the hard part moved somewhere else. The survey asked what gets in the way, and the top answer was not a technical one. 47% of respondents cited cultural changes within the development team as their leading challenge.

That tracks with the other split in the data. Among organizations the survey classes as cloud native innovators, 58% use GitOps principles extensively. Among those it classes as adopters, that figure is 23%. The tooling is available to everyone at the same price, which is to say free. The difference between the two groups is not access to software. It is whether the organization has agreed on how it wants to work.

This is the least surprising finding in the report and the one most likely to be ignored, because a culture problem cannot be solved by installing anything.

What to do with this if you are learning

Three practical readings.

The learning curve is now a one-time cost with a long payoff. When a platform churns, knowledge decays and you pay the cost repeatedly. Kubernetes still changes, but the parts you would learn first have been stable for years. Time spent on Pods, Services, Deployments, and the scheduler is not going to expire.

Operational skill is worth more than architectural opinion. The market has enough people who can draw a cluster diagram. What the 82% figure implies is a very large installed base of running clusters, and running clusters break. Debugging skill is the scarce good.

The adjacent skills are the differentiator. GitOps, storage, networking, and the ability to explain a change to a team that is nervous about it. The survey’s own data says the blocker is rarely the container runtime.

Boring technologies are where careers are actually built. The exciting ones are where they are occasionally made and more often lost.

Sources: CNCF 2025 Annual Cloud Native Survey announcement.