NVIDIA open-sources NVKE, a system to run Kubernetes on GPU fleets

NVIDIA has opened NVKE, the NVIDIA Kubernetes Engine, in preview: an open-source system that lets AI factory operators create and run Kubernetes clusters on their own GPU infrastructure, without building a cluster management system from scratch.

The official documentation, updated on November 6, 2026, describes NVKE as the Kubernetes management layer of DSX OS, NVIDIA's software stack for AI factories. Each workload cluster the system creates has its own control plane and its own etcd, isolating customers from one another on the same infrastructure.

What NVKE does

Through the NVKE API or the nvkectl CLI, the operator creates, describes, updates and deletes clusters and node pools. The system provisions bare-metal nodes, forms the cluster and handles ongoing operations: scaling node pools, upgrading Kubernetes, replacing nodes and renewing certificates.

Features highlighted by NVIDIA include:

  • Verified node identity: on bare metal, each node attests its identity with its TPM via SPIFFE/SPIRE before receiving the credential to join the cluster; etcd encryption at rest can be enabled per cluster.
  • Verifiable artifacts: the companion project NVKR (NVIDIA Kubernetes Recipes) builds node OS images and components from pinned sources, records a software bill of materials (SBOM) and signs releases.
  • GPU-aware operations: with NVSentinel installed, NVKE reboots or replaces nodes with a detected GPU fault.
  • Open standards: the system uses upstream Kubernetes, kubeadm and Cluster API, and the CNCF has certified that the clusters it creates are conformant Kubernetes — NVKE is a Certified Kubernetes AI Platform.

Preview, not production

NVIDIA is explicit about the project's stage: NVKE is available as a preview, APIs, configuration and behavior can change between releases, and the company does not recommend it for production workloads yet. The code is licensed under Apache 2.0 and accepts outside contributions.

NVIDIA also draws the scope clearly: NVKE is not a Kubernetes distribution and not a Cluster API flavour — it is a self-hosting factory for workload clusters. And it does not manage bare metal: that job belongs to NICo (NVIDIA Infra Controller), which NVKE drives through CAPNICo.

Why it matters

For cloud providers and AI infrastructure operators, offering managed Kubernetes requires building a system with API-driven provisioning, isolated control planes per customer, upgrades, node identity and observability. NVKE packages that work as open software — arriving in the same year NVIDIA open-sourced the Run:ai platform in October. The move reinforces the company's strategy of turning its software stack into an open standard for anyone operating GPU fleets.

Also on Geekiki: CIQ launches Stratum Kubernetes Engine, open Kubernetes for the AI era and NVIDIA Vera Rubin NVL72: up to 30x more throughput per dollar with only 3% more HBM.

Image credit: NVIDIA logo, via Wikimedia Commons. Original file by NVIDIA, licensed under Apache 2.0.

Sources

Transparency: this article was produced with the help of artificial intelligence, cross-checked against public posts and sources on the internet, and reviewed against the original sources cited. We recommend consulting the original sources to confirm the details.

By Muse