NVIDIA announced at the KubeCon + CloudNativeCon conference that it is donating OpenShell, its open-source runtime for securing and governing autonomous artificial intelligence (AI) agents, to the Cloud Native Computing Foundation (CNCF), the Linux Foundation project that stewards Kubernetes. It will also contribute to Agent Substrate, a Google-led project for running long-lived agents on Kubernetes, and is unveiling three new open-source pieces for managing GPUs on Kubernetes: NVIDIA Kubernetes Engine, NVIDIA Kubernetes Recipes, and NVIDIA Snapshot. The announcement was made on November 10, 2026.
NVIDIA’s KubeCon announcements in 30 seconds
- OpenShell, NVIDIA’s runtime for controlling what AI agents do, becomes a CNCF Sandbox project.
- NVIDIA is contributing to Google’s Agent Substrate to scale agents on Kubernetes.
- NVKE and NVKR, in public preview, help providers offer managed Kubernetes on top of GPUs.
- Snapshot restores inference replicas in seconds, without reloading the model.
The announcement is signed by Warren Barkley of NVIDIA, who argues that Kubernetes has become the foundation of AI and that its demands keep growing. All of these pieces are part of what NVIDIA calls the DSX AI Factory Platform, its software stack for AI data centers, which spans everything from GPU infrastructure to orchestration and model serving.
OpenShell Moves to Community Governance
OpenShell applies deterministic controls to agent activity, so teams can enforce policies when an agent uses tools, calls inference services, or runs workloads. Its architecture allows connecting compute, inference providers, credential stores, and custom monitoring tools. According to NVIDIA, hundreds of organizations already use it, and the project has received tens of thousands of contributions in under a year; in Korea, for example, Hyundai and LG are exploring it.
As a Sandbox project, the CNCF’s entry level, OpenShell will now be governed by the organizations that build it into their products. “The cloud native community has a new infrastructure challenge: giving agents safe environments to operate in,” says Jim Zemlin, executive director of the Linux Foundation. Handing the project to a neutral foundation reduces the fear of relying on a single vendor, a key factor in making agent security a standard.
NVIDIA will also contribute to Agent Substrate, a project that originated at Google and has been accepted into the CNCF Sandbox. It preserves each agent’s environment between tasks but frees up compute when idle, suspending and resuming environments on a shared pool. NVIDIA will integrate it with Dynamo, its inference engine, and with KAI Scheduler, and both companies are working on an integration with OpenShell to govern agents across clouds and on-premises data centers.
Managed Kubernetes on GPUs, and Startups in Seconds
NVIDIA Kubernetes Engine (NVKE) and NVIDIA Kubernetes Recipes (NVKR), in public preview, are open-source components that let AI cloud operators offer managed Kubernetes on top of their own GPU infrastructure. They automate the cluster lifecycle and use signed images with pinned versions. Mirantis, Netris, OpenNebula Systems, Rafay, Spectro Cloud, and vCluster, along with Firmus and SK Telecom, have committed to integrating them into their offerings. For independent cloud providers, it’s an option for competing with hyperscalers’ Kubernetes services without building the entire layer from scratch.
NVIDIA Snapshot tackles another well-known problem: loading models and warming up workloads delays scaling. It restores saved replicas in seconds, without reloading the model. According to NVIDIA, Crusoe tested GLM-5.3 on HGX B300 systems and cut the time to a ready workload from about eight minutes to 31 seconds, and Saturn Cloud cut cold-start times for vLLM from between 2 and 8.5 minutes down to between 6 and 34 seconds. These are design-partner tests published by the vendor itself.
Frequently Asked Questions
What is NVIDIA OpenShell?
An open-source runtime that applies deterministic policies and controls to what AI agents do, now donated to the CNCF.
What is Agent Substrate?
A CNCF project that originated at Google for running long-lived agents on Kubernetes while freeing up compute when they’re idle.
What are NVKE and NVKR for?
They let cloud providers offer managed Kubernetes on top of their GPU infrastructure, with automated cluster management. They’re in public preview.
Sources:
This article was originally published in Spanish at revistacloud.com.
