NVIDIA has released a public preview of its open-source NVIDIA Kubernetes Engine (NVKE) and NVIDIA Kubernetes Recipes (NVKR). These tools provide building blocks for delivering managed services for cloud-native computing applications.
NVIDIA announced the tools at KubeCon + CloudNativeCon North America 2026. Mirantis, Netris, OpenNebula Systems, Rafay, Spectro Cloud, vCluster, Firmus, and SK Telecom have already pledged their support.
NVKE is an open-source Kubernetes management system. It enables teams to provision, scale, and operate Kubernetes clusters. NVKR provides best practices for deploying NVKE.
Kubernetes evolves for AI workloads
Organizations have used Kubernetes clusters to run cloud-native applications for years. Now, adoption is increasing as more organizations deploy AI applications.
Many developers build these applications with containers and use Kubernetes to orchestrate them. The Cloud Native Computing Foundation (CNCF) oversees the open-source Kubernetes platform.
However, AI workloads continue to evolve. As organizations deploy more AI agents, Kubernetes must also adapt to run these new workloads more efficiently.
Google, for example, recently contributed its open-source Agent Substrate framework to the CNCF. The framework helps Kubernetes clusters run long-running agent workloads more efficiently at scale.
Agent Substrate separates stateful agent sessions and identities from active compute. The framework can create snapshots of idle agents and store them for later use. It can then restore those agents in under 500 milliseconds. This approach reduces costs by using infrastructure resources more efficiently.
Currently, Kubernetes assigns every active agent its own pod. That approach consumes memory and CPU resources even when an agent sits idle. As a result, the Kubernetes control plane may need to track millions of persistent objects it was never designed to manage.
Performance and security requirements grow
AI agents also create performance challenges for the Kubernetes control plane. As agents make exponentially more application programming interface (API) calls, they can put additional pressure on control plane performance.
Organizations also need stronger security capabilities to protect AI agents. Google contributed gVisor to the CNCF to help address this challenge. The technology isolates code inside a lightweight userspace application kernel that can run as a container.
NVIDIA also contributed OpenShell to the CNCF. The open-source runtime helps organizations securely deploy artificial intelligence (AI) agents.
Other technologies could provide another option for agentic AI workloads. Frameworks such as WebAssembly (Wasm), for example, could help organizations run these workloads more efficiently.
Cosmonic recently demonstrated this potential by running 1.42 million AI agents in sandboxes. The company ran the agents on a 64-core server provided by Advanced Micro Devices (AMD).
An emerging opportunity for MSPs
These initiatives may take time to reach the Kubernetes distributions that MSPs deploy. However, the number of AI workloads continues to grow.
That growth creates an opportunity for service providers with the expertise to deploy and manage AI applications. MSPs can build on Kubernetes’ existing presence in enterprise IT environments. They can then extend their cloud-native expertise into AI infrastructure and managed services.
Photo: DC Studio / Shutterstock

