As organizations push AI workloads into production, they’re running into a familiar struggle: fragmentation. The Cloud Native Computing Foundation’s newly launched Certified Kubernetes AI Conformance Program, announced at KubeCon + CloudNativeCon North America 2025, addresses the problem of AI’s many divergent platforms, tools and standards.

Longtime CNCF supporter and Kubernetes contributor VMware by Broadcom also announced that its vSphere Kubernetes Service (VKS) is certified as one of the first conformant platforms for the new AI standard.

In this On the Road episode of The New Stack Makers recorded at KubeCon in Atlanta, TNS Founder and Publisher Alex Williams discussed these new open, community-defined standards for running AI workloads on Kubernetes. He was joined by Broadcom’s Dilpreet Bindra, senior director of engineering, and Himanshu Singh, director of product marketing.

Bringing Order to AI Tooling Chaos

Bindra sees the new conformance program as the CNCF applying lessons learned from Kubernetes’ early days.

“There’s a lot of muddiness around delving into AI tooling,” he noted. “With the AI conformance program, the CNCF is doing as they did with Kubernetes in the beginning, making sure that applications written on top of these tools can be utilized across multiple different platforms.”

Conformance, Singh added, “at its heart is the idea of portability.” Platform vendors, infrastructure teams, enterprise AI practitioners and open source contributors in Kubernetes all need a common and interoperable foundation, he continued. “If I’m running on a public cloud and want to move my application to a private cloud, or vice versa, conformance means I should be able to take my application where it’s best suited with minimal friction.”

Running Kubernetes APIs Directly Against Your Infrastructure

VMware vSphere Kubernetes Service is the CNCF-certified Kubernetes runtime that ships as part of VMware Cloud Foundation. What makes VKS distinctive, Bindra said, is how deeply it integrates with the underlying vSphere platform rather than adding another abstraction layer. It’s VMware’s architected VKS to use Kubernetes APIs directly against the infrastructure.

“Everything is a YAML document, all your objects,” Bindra said. “This means VMs, pods, persistent volumes and Kubernetes clusters themselves can all be managed declaratively through Kubernetes APIs and tooling like Argo CD.”

To help with this, the latest VKS release includes a new add-on management system that lets users manage extensions across all their VKS clusters, whether pre-packaged components or third-party tools.

Why Kubernetes Maturity Matters for AI

All of this is possible, Singh and Bindra agreed, because Kubernetes itself has become an unshakeable foundation. That stability is what’s enabling enterprises to trust Kubernetes with production AI workloads. It’s also why Broadcom has been doubling down support for foundational projects like Cluster API and etcd, components that make reliable, scalable Kubernetes possible.

“A lot more users and applications and enterprises now fully trust the technology,” Bindra said. “Our view is that most of your application development is going to be done using Kubernetes in the future.”

Singh added, “You want the technology to mature, you want the platform to mature, and then the community can keep building to continually extend it and take things to a new level.”

“And, at this point in time,” he concluded, “AI is the new level.”

For the full conversation — including how VMware is using Kubernetes APIs to give virtual machines and containers equal footing, and why the company just open-sourced its etcd tooling — check out the complete interview.

TRENDING STORIES