The Kubernetes AI Conformance program establishes a standardized baseline for Kubernetes platforms to reliably run AI/ML workloads. It targets Kubernetes platform vendors seeking certification and aims to improve AI application portability across different environments, reducing platform-specific workarounds and providing a clear target for the AI tooling ecosystem.

How It Works

This program defines essential capabilities for Kubernetes platforms to support AI/ML workloads, building upon existing Kubernetes conformance. Vendors seeking certification must demonstrate their platform meets specific requirements across areas like accelerators, networking, scheduling, and security. The process involves a self-assessment via a checklist, gathering public evidence, and submitting a pull request to the project repository for review by the CNCF.

Quick Start & Requirements

To begin the certification process, vendors must first ensure their platform is already Kubernetes Conformant. The certification involves preparing by reviewing requirements, documenting compliance through a checklist and evidence, and submitting a pull request to the cncf/k8s-ai-conformance repository. Automated conformance tests are planned for 2026; currently, certification relies on self-assessment. Detailed instructions are available in instructions.md.

Highlighted Details

• Covers common AI/ML use cases: Training, Inference, and Agentic workloads.
• Certification is currently a self-assessment process, with automated testing slated for 2026.
• Certifications are valid for one year and must be renewed, applying per-product and per-configuration.
• Requirements encompass accelerators, networking, scheduling, observability, security, and operator support.

Maintenance & Community

This is a community-led effort governed by the Kubernetes AI Conformance project. Contributions are welcomed in documentation, research for new workload types, testing development, and general discussion. Involvement can be initiated by opening an issue or joining the project meetings. Contact is available via sig-architecture@kubernetes.io and Slack.

Licensing & Compatibility

The project is licensed under the Apache License 2.0. This permissive license generally allows for commercial use and integration within closed-source projects without significant restrictions.

Limitations & Caveats

Current certification relies on vendor self-assessment, with automated testing not expected until 2026. Certification is time-bound (one year) and specific to product configurations. Platforms must first achieve base Kubernetes conformance before pursuing AI conformance.