1 of 19

Advancing Kubernetes AI Conformance: Current State and Roadmap

Janet Kuo, Staff Software Engineer, Google

Yuan Tang, Senior Principal Software Engineer, Red Hat

Mario Fahlandt, Customer Delivery Architect, Kubermatic

#KubeCon #CloudNativeCon

2 of 19

Janet

Kuo

Yuan

Tang

Mario Fahlandt

Rita

Zhang

3 of 19

Intro

4 of 19

How It Started

  • AI introduces new requirements: GPU-intensive, synchronously scheduled workloads, specialized hardware, networking security, etc.
  • Without standardized conformance, organizations suffered from significant fragmentation.
  • This fragmentation resulted in major difficulties when migrating AI models and created vendor lock-in.

5 of 19

Kubernetes AI Conformance

  • Kubernetes Conformance isn’t enough
    • Some parts are outside of Kubernetes core
    • Faster iterations
  • WG AI Conformance was officially established in 2025 to solve this
    • Sponsored by SIG Architecture
  • Original AI Conformance proposal (SHOULD & MUST requirements)
  • Additional requirements for platforms, on top of Kubernetes Conformance
    • It doesn’t limit how end users want to run AI on Kubernetes
    • It require platforms to support capabilities / APIs most use cases need

6 of 19

Goal

  • Define a standardized set of capabilities, APIs, and configurations for running AI/ML workloads on Kubernetes.
  • Aims to simplify AI/ML operations on Kubernetes, accelerate adoption, guarantee interoperability and portability for AI workloads, and enable ecosystem growth on an industry-standard foundation.
  • “Write once, run anywhere”, for AI workloads too!

7 of 19

The Current State

  • Working Group is temporary in nature, and can’t own code
  • Now a SIG Architecture Subproject
  • A permanent home for the program's long-term maintenance and evolution (same path as Kubernetes Conformance)
  • https://github.com/kubernetes-sigs/ai-conformance

​

8 of 19

What have we accomplished?

9 of 19

Program Launch at KubeCon North America 2025

https://www.cncf.io/announcements/2025/11/11/cncf-launches-certified-kubernetes-ai-conformance-program-to-standardize-ai-workloads-on-kubernetes/

10 of 19

Certified Kubernetes - AI Platform in CNCF Landscape

https://landscape.cncf.io/?group=certified-partners-and-providers

https://github.com/cncf/landscape

11 of 19

Conformance Checklists

​

Defined and released conformance checklists for multiple Kubernetes versions, providing a clear baseline for AI-ready clusters:

metadata:

kubernetesVersion: v1.35

platformName: "[Platform Name]"

platformVersion: "[Platform Version]"

vendorName: "[Vendor Name]"

…

spec:

accelerators:

- id: dra_support

description: "Support DRA APIs ..."

level: MUST

status: ""

evidence: []

notes: ""

- id: driver_runtime_management

description: "Provide a verifiable mechanism"

level: SHOULD

…

https://github.com/cncf/k8s-ai-conformance

12 of 19

Requirement Definition and Graduation Process

13 of 19

What’s next?

14 of 19

What’s next?

1.36: Inference

​

1.37: agents?

​

Automate tests

go test -v ./test [-run <TestName>] [-kubeconfig=<path/to/kubeconfig>] [-accelerator-type=<type>]

We need more conformance tests for specific areas, and more standardization.

​

15 of 19

Conformance

programs in a nutshell

​

16 of 19

Why? & Who?

Why?��Improves compatibility among solutions built upon the project

​

Guarantees that users get the expected functionality and a consistent interface, even when moving from one vendor offering to another.

​

Creates a defined standard for what it means to be a "conformant" or "certified" solution in your projects ecosystem.

​

Who?��The Project

​

The Governing Board of the CNCF

​

Project Team of CNCF

​

If it is overarching the TOC

​

17 of 19

How to get involved?

18 of 19

How to get involved

  • AI Conformance Biweekly Meeting
    • Thursdays at 7 PM CET | 10 AM PDT
  • Communications:
    • Slack: #ai-conformance on K8s Slack
    • Mailing list: sig-architecture mailing list
  • https://github.com/kubernetes/community/tree/main/sig-architecture

19 of 19