k0s is now Certified Kubernetes AI Conformant
)
)
Summarize with AI
){kind=icon}
Claude){kind=icon}
ChatGPT){kind=icon}
Gemini){kind=icon}
Perplexity){kind=icon}
Grokk0s has passed the CNCF Certified Kubernetes AI Conformance program at Kubernetes v1.35. The submission, k0s v1.35.3+k0s.0, is public, reproducible, and listed alongside every other certified platform in the cncf/k8s-ai-conformance repository.
k0s was already a Certified Kubernetes distribution. AI conformance is the next bar, and a more demanding one. In plain terms, it is an independent confirmation that the AI and machine learning workloads people actually deploy today, things like model training, inference serving, and AI agents, run reliably on k0s using standard, off-the-shelf Kubernetes tools. For a single-binary distribution that prides itself on staying small, clearing it was a satisfying result.
What this means
The Certified Kubernetes AI Conformance program launched at KubeCon North America in November 2025. It defines a baseline set of capabilities every platform needs before it can credibly claim to run AI workloads, and it asks each vendor to prove those capabilities with public, re-runnable evidence rather than marketing claims.
We ran that proof on a real cluster: k0s with NVIDIA T4 GPUs on Azure. Across the program's requirements, k0s showed it can:
Hand GPUs to workloads safely. Containers get access to exactly the accelerators they ask for and nothing more, and the right drivers are installed and managed automatically.
Route model and inference traffic. A standard Gateway API implementation directs requests to the right model backend.
Schedule large, distributed jobs. All-or-nothing ("gang") scheduling means a big training job either gets everything it needs or waits, instead of half-starting and wedging the cluster. Clusters and pods scale up and down with GPU demand.
Show what the hardware is doing. Per-GPU utilization, memory, temperature, and power flow into standard monitoring, alongside metrics from the AI workloads themselves.
Run complex AI tooling. Heavyweight operators like KubeRay install and run reliably, including self-healing when something is deleted.
The program separates mandatory capabilities from optional, nice-to-have ones. k0s covered all of the mandatory set, plus extras like sharing a single GPU across several workloads (time-slicing).
The full, requirement-by-requirement breakdown, with commands anyone can rerun, lives on the k0s documentation site.
Why it matters
k0s is a zero-friction, single-binary Kubernetes distribution. The instinct with a lean distribution is to wonder what it gives up. AI conformance answers that directly: nothing that matters for running AI workloads.
The certification proves that the components the AI ecosystem depends on, from GPU scheduling to traffic routing to monitoring, all work on k0s in their standard, upstream forms. There are no custom abstractions to learn and nothing proprietary to adopt. If a project runs on upstream Kubernetes, it runs on k0s.
For teams deciding where to run training and inference, that is independent verification rather than a vendor promise. The test was simple and public: a CNCF reviewer could rerun our documentation on their own infrastructure and get the same result.
What's next
k0s AI conformance is the first half of a two-product story. k0rdent, Mirantis's open source cluster lifecycle platform, provisions k0s clusters under the hood, and it pursued AI conformance too. Because a k0rdent cluster is a k0s cluster underneath, most of this evidence carried straight over, and k0rdent's submission then went further to cover the optional capabilities as well. That neat hand-off between the two was one of the nicer payoffs of doing the work in the open: get it right once on k0s, and the platform on top mostly inherits it.
Resources
k0s AI Conformance submission: the public submission and per-requirement evidence (v1.35/k0s).
k0s documentation: the AI conformance evidence page and full project docs.
k0s on GitHub: source, releases, issues.
CNCF Certified Kubernetes AI Conformance: the program, instructions, and the full list of certified platforms.
CNCF launch announcement: the program's debut at KubeCon NA.
)
)
)
)
)
