NVIDIA Kubernetes Engine

Overview

NVIDIA Kubernetes Engine (NVKE) is open-source software that an AI factory installs to run a managed Kubernetes service on its own NVIDIA GPU infrastructure. Operators call the NVKE API to create clusters for customers; NVKE provisions nodes, forms each cluster with its own control plane and etcd, and then operates it by scaling node pools, upgrading Kubernetes, replacing nodes, and renewing certificates. It connects to an infrastructure provider through Cluster API and uses an NVKE node agent to configure nodes without SSH.

Key Features

• An API for managed clusters that lets operators request clusters by Kubernetes version, node pools, and network settings while hiding Cluster API, kubeadm, and node configuration behind the API and nvkectl.
• Artifacts from pinned sources that operators can verify, including a reference node image built from the NVKE checkout and its pinned build inputs, with optional component and package artifacts from NVKR builds.
• Bare-metal node identity attestation through TPM and SPIFFE/SPIRE before a node receives a join credential, with optional etcd encryption at rest per cluster.
• GPU-aware day-2 operations through node pools, including upgrades, node replacement, certificate renewal, and remediation requested when NVSentinel detects a GPU fault.
• Built on upstream Kubernetes, kubeadm, and Cluster API, producing standard Kubernetes clusters that can be used with kubectl and clusterctl alongside nvkectl.
• An OpenTelemetry-based telemetry pipeline, NKTP, for metrics, logs, and traces from the management cluster and NVKE clusters.

Use Cases

• AI factory operators build artifacts with NVKR, install and run NVKE, and create and operate Kubernetes clusters for their customers.
• AI factory customers receive a Kubernetes API endpoint and node pools from their operator and run AI workloads on them.
• Evaluators and contributors try NVKE on one Linux machine against a simulated infrastructure provider, or build and change NVKE and NVKR.

Getting Started

• Evaluate locally with mock-NICo on one Linux machine with at least 6 CPUs and 12 GiB of memory.
• Run NVKE end to end on one machine.
• For hardware installation, plan the installation, build and host node OS images, prepare nodes, bootstrap the management cluster, install the management plane, configure NICo as the infrastructure provider, and secure the NVKE API.

Deployment & Requirements

• NVIDIA GPU machines and the networks that connect them, managed by NICo.
• Nodes for the management cluster prepared as described in the NVKE documentation.
• A reference-image build host with the required tools, an HTTP(S) boot-artifact endpoint, and registries for container and package artifacts.
• An identity provider that issues signed JWTs if NVKE API authentication is enabled.
• At preview, NVIDIA Infra Controller (NICo) as the supported infrastructure provider.

Before You Adopt

• License: Apache-2.0. Review its terms before using, modifying, or distributing the project.
• NVKE is a preview release; APIs, configuration, and behavior may change, and production use is not recommended.
• At preview, NVKE's infrastructure-provider support is limited to NVIDIA Infra Controller (NICo).
• API authentication, when enabled, depends on an external identity provider that issues signed JWTs.
• Only NVKR Releases from the hosted pipeline are signed and carry build provenance; local Runs are unsigned.
• No NVKR Release is qualified at preview.