CNCF Adds Program to Standardize AI Workloads on Kubernetes Clusters
The Cloud Native Computing Foundation (CNCF) today launched a program that aims to bring more standardization to how artificial intelligence (AI) workloads are deployed on Kubernetes clusters.
Announced at the KubeCon + CloudNativeCon North America 2025 conference, the Certified Kubernetes AI Conformance Program defines a set of capabilities and configurations required to run AI and machine learning frameworks on Kubernetes.
Jonathan Bryce, executive director for the CNCF, said the overall goal is to make it simpler to ensure that AI and ML workloads are truly portable across multiple Kubernetes environments. That’s especially critical in an era where more organizations need to adhere to sovereign cloud computing mandates that require them to deploy applications within the physical border of a country, he added.
CNCF CTO Chris Aniszczyk additionally noted that the program will also make it easier to deploy AI and ML workloads that adhere to these guidelines across hybrid cloud computing environments. Many AI workloads are already being deployed using inference engines that will be running a wide range of classes of processors, so ensuring interoperability is becoming a more pressing issue, he added.
The challenge is that while many AI workloads already run on Kubernetes clusters, many of the data science teams that built them did not always fully understand how to avoid being locked into a specific platform, noted Aniszczyk.
Initially launched earlier this year as a beta project, version 1.0 of the Certified Kubernetes AI Conformance Program has drawn support from Broadcom, Google, Microsoft, Oracle and Red Hat. Version 2.0 of the program is now being developed with a release goal for next year.
The Certified Kubernetes AI Conformance Program is similar in scope to the Certified Kubernetes Conformance Program that the CNCF maintains to certify more than 100 interoperable distributions of Kubernetes.
It’s not clear just how many AI/ML workloads are running on Kubernetes clusters, but the number is expected to exponentially increase. While many of the first wave of these applications were built and deployed by data science teams, deployment of AI/ML workloads is rapidly shifting to centralized IT teams that have more experience managing applications at scale in production environments.
Those teams typically have a greater appreciation for ensuring that their organization doesn’t find itself locked into a specific IT infrastructure platform because developers, for example, relied too much on a set of proprietary application programming interfaces (APIs), noted Aniszczyk.
Regardless of approach, the one clear thing is that as more AI/ML workloads are built and deployed, the number of Kubernetes clusters running in production environments should continue to steadily increase. The challenge then becomes finding and retaining IT professionals who have the expertise required to successfully manage those clusters.
In the meantime, IT teams would be well advised to assess how existing AI/ML workloads were constructed with an eye toward re-engineering them, to ensure they are as portable as possible in an era where moving them from one platform to another can, for any number of reasons, suddenly become a very high priority.


