NVKE is a preview management plane for GPU Kubernetes clusters on bare metal
NVIDIA's Apache-2.0 NVKE lets platform teams run a managed Kubernetes service on their own GPU machines, with isolated workload clusters and TPM-based node identity, still in preview.
What NVKE is
NVIDIA Kubernetes Engine (NVKE) is a Go project, released under the Apache-2.0 license, that describes itself as a management plane for GPU Kubernetes on bare metal. The README pitches it as a way to run a managed Kubernetes service on your own NVIDIA GPU infrastructure: platform teams install and operate the NVKE management plane, and their customers use the resulting clusters through the Kubernetes API and kubectl. Each workload cluster gets its own Kubernetes control plane, etcd and node pools.
The repository was created on 2026-08-04. In the GitHub trending snapshot it had 791 stars, 345 of them gained in the last day, and 109 forks.
The maintainers are explicit that this is a preview. APIs, configuration and behavior can change between releases, and the README says they do not recommend NVKE for production workloads.
How it is put together
The README says NVKE is not a Kubernetes distribution and not a Cluster API flavour. It calls it a self-hosting factory for Kubernetes clusters. A throwaway seed cluster stands up a management cluster, and the management plane running there provisions and manages the lifecycle of many isolated workload clusters on demand.
An installation therefore has two kinds of cluster:
- The management cluster runs the NVKE management plane and infrastructure controllers.
nvkectlforms it with a temporary seed cluster; installing the management-plane services is a separate step. - Workload clusters run customer applications. The management plane creates and operates them through NICo, Cluster API and a node agent.
NICo stands for NVIDIA Infra Controller and is the sole compatible infrastructure provider at preview. Operators request clusters through the NVKE API or nvkectl, then create, scale and delete node pools, upgrade Kubernetes and replace nodes. Nodes attest through SPIFFE/SPIRE using their TPM before they receive join credentials, and encryption of etcd data is available per cluster. A companion project, NVIDIA Kubernetes Recipes (NVKR), builds node images and NVKE components from pinned inputs; NVKE deploys those artifacts.
Building from source uses Go 1.27.3 and GNU Make 4 or later:
git clone https://github.com/dsx-ai-factory/nvke.git
cd nvke
make build
According to the README this writes nine binaries to bin/, including nvkectl, controller, agent, tenant-api and operator-api, without starting a management plane or creating a cluster.
Trying it without GPUs
A local quickstart creates a development management plane and a workload cluster on simulated machines, with no NVIDIA account required. It needs a Linux host or VM (amd64 or arm64) with at least 6 CPUs, 12 GiB of free memory and 100 GiB of free disk, plus Docker and virtual TPM support. On macOS the evaluation runs inside a Colima virtual machine.
Preview limitations
The README lists several gaps an operator has to handle. After installation, API authentication is off and the API uses plain HTTP, so authentication and protected access must be configured before admitting callers. The management cluster does not upgrade, scale or replace itself, and there is no management-plane upgrade or backup/restore procedure. FDE node images are for evaluation and provide no data-at-rest confidentiality. Node replacement and upgrades also have preview limits that the node pool guide documents as manual steps.
Source: dsx-ai-factory/nvke on GitHub