# NVIDIA move to secure AI agents Kubernetes favors operators _NVIDIA donated OpenShell to CNCF Sandbox and added Kubernetes tools for AI factories, with early tests showing startup cuts from minutes to seconds._ **Published:** 2026-11-10 **Source:** https://www.startuphub.ai/ai-news/technology/2026/nvidia-move-to-secure-ai-agents-kubernetes-favors-operators --- [NVIDIA](https://blogs.nvidia.com/blog/open-source-openshell-infrastructure/) will contribute OpenShell to the CNCF Sandbox and help scale Agent Substrate to secure AI agents Kubernetes operations with shared controls. NVIDIA says hundreds of organizations already use OpenShell and the project drew tens of thousands of contributions in less than a year. That adoption pace gives the donation weight beyond a code drop. Control, not compute, is now the bottleneck. Jim Zemlin, CEO of the Linux Foundation that hosts CNCF, said the cloud native community needs shared foundations for agent security because agents need secure environments in which to operate. His view matters because CNCF governance decides how fast OpenShell controls get reviewed and adopted across vendors. GitHub currently describes OpenShell as the safe, private runtime for autonomous AI agents, per its public repository listing. Deterministic policy enforcement is the pitch. NVIDIA says OpenShell plugs into existing compute, inference and credential tools so teams can enforce tool and access policy. NVIDIA says it will align Agent Substrate with Dynamo for inference and KAI Scheduler for shared allocation, suspending idle agent environments instead of holding dedicated compute. NVIDIA also released NVIDIA Kubernetes Engine and NVIDIA Kubernetes Recipes in public preview to automate cluster lifecycle work with signed, version pinned images. Mirantis, Rafay, Spectro Cloud and SK Telecom have aligned on integration plans, which spreads NVIDIA tooling through managed service paths. NVIDIA says [Crusoe](https://www.startuphub.ai/startups/crusoe), an upstream contributor, cut a GLM-5.3 workload ready time from roughly eight minutes to 31 seconds on HGX B300 systems using Snapshot. Saturn Cloud benchmarked Snapshot for scale from zero inference and cut vLLM startup from between two and 8.5 minutes to six and 34 seconds, NVIDIA says. Snapshot restores checkpointed replicas without model reload, so recovery and elasticity improve directly. StartupHub.ai data tracks Crusoe and Saturn Cloud as AI cloud operators running NVIDIA hardware, not rival silicon vendors, so faster startup lowers their idle cost rather than threatening NVIDIA sales. StartupHub.ai data shows no verified standalone funding figure for Snapshot as of November 2026, since NVIDIA positions it as open infrastructure. That read would change if operators publish sustained throughput or cost data beyond lab tests. The open question is whether CNCF shared control can evolve as fast as agent behavior. --- Original analysis from [startuphub.ai](https://www.startuphub.ai), the #1 AI startup directory. © StartupHub.ai. All rights reserved. You may not republish this article in full without a license. Search engines and AI research tools may crawl and summarize for reference. Bulk reproduction or model training on this content requires a license. See https://www.startuphub.ai/terms.