What the CNCF Survey Reveals About Cloud Native’s AI Future

Cloud native has quietly crossed a point of no return. The 2025 CNCF Annual Survey, built on responses from 628 practitioners across industries and geographies collected in September 2025, reads like a snapshot of an ecosystem becoming the default operating environment for modern software. That matters because once a platform becomes default, the next question is never “Should we use it?” Now the question is, “What do we build on top of it?”

The survey shows AI being pulled into the same gravity field as containers, Kubernetes, CI/CD, and developer platforms. In other words, the future of AI will be decided by the capabilities that let organizations reliably run complex systems at scale and not just better models. 

Let’s dig into what the survey reveals and what it implies for the AI-shaped cloud native decade ahead.

Kubernetes is turning into the AI infrastructure layer

There was a time when Kubernetes was described as a container orchestrator. That description now feels too small.

Among container users, 82% report using Kubernetes in production in 2025, up from 66% in 2023. When a platform becomes this common, it stops being a technology choice and starts functioning like shared infrastructure.

Now add the AI signal, where 66% of organizations report running generative AI workloads on Kubernetes.

Most people think that the AI race is between those who have the best model. The survey suggests that the race may be between those who can make AI work reliably as a production workload. 

Maturity follows a pattern, and GitOps exposes where you are

The survey not only talks about adoption, but it also shows how mature the market is. The researchers have grouped organizations into four profiles that form a progression: explorers, adopters, practitioners, and innovators. Explorers represent 8% of organizations, adopters 32%, practitioners 34%, and innovators 25%.

Rather than experimenting, most teams are now operationalizing and standardizing. They are discovering the difference between “we run Kubernetes” and “we run Kubernetes well.”

The survey suggests you can predict maturity by what practices show up consistently at each stage. There’s a standout indicator that functions almost like a truth serum: GitOps.

Explorers haven’t used GitOps yet, but 58% of innovators have deployments that work with GitOps.

When inference services, data pipelines, and model-serving components get stitched into production, drift becomes more than annoying. It becomes a risk multiplier. GitOps is attractive here not because it’s fashionable, but because it forces environments to be reproducible and changes to be auditable.

AI’s ambition is high, but the delivery rhythm is slow

The survey reveals a gap that should make every AI strategy slide deck a little more uncomfortable.

The survey found that 52% of people do not train their AI models at all, while 47% only use them sometimes.

Training is expensive and specialized, and many organizations can create value by consuming models rather than building them. If you aren’t training models, your edge doesn’t come from inventing algorithms. It comes from how reliably and quickly you can deploy AI capabilities into products, update them, observe them, and keep them safe.

And the survey points directly at that. It suggests the real advantage comes from robust delivery pipelines and optimization discipline, the kind of “boring” engineering that becomes incredibly exciting the moment AI is involved. When models are updated, the blast radius can be bigger than a typical microservice release. That should change how you think about rollback, testing, and monitoring.

Open source sustainability is becoming a real risk factor

With the help of more than 270,000 contributors from around the world, the CNCF ecosystem now includes 234 projects at different stages of development. The survey’s themes point to growing strain on the systems that build, test, deploy, and distribute software, especially as automated and machine-driven usage ramps up.

Simply put, AI doesn’t just use computers; it also uses supply chains. It amplifies the load on registries, build systems, dependency graphs, and security tooling. The future isn’t only about what you run in production. It’s also about whether the foundations beneath production remain healthy.

If your organization’s AI roadmap assumes infinite free infrastructure from the open source world, the survey is a reminder to take that assumption seriously and support the projects you rely on.

The next frontier might not be WebAssembly

The report also offers an interesting counter-signal. About 65% of organizations report no WebAssembly experience, a number that has remained consistent across recent years, and only 5% report full deployment experience in 2025.

This doesn’t mean WebAssembly won’t matter. It does suggest that the near-term cloud native AI future is not about swapping runtimes or chasing novelty. It’s about industrializing what already works. Kubernetes, containers, delivery pipelines, and observability are where the energy is concentrated right now.

Cloud Native’s AI future

The CNCF survey doesn’t predict a sudden rupture where AI rewrites infrastructure overnight. It shows a more realistic evolution: cloud native practices absorbing AI workloads the way they absorbed microservices and distributed systems, by turning them into operationally manageable units.

If your cloud native foundation is strong, AI becomes an extension of what you already do well. If it’s fragile, AI will expose that fragility quickly and publicly.

So here’s the question: when AI arrives in your environment, will it feel like a new product opportunity, or like a stress test you aren’t ready to take?

Because the survey suggests the future won’t wait for comfort. It will reward readiness.