Glonce

NVIDIA releases NVKE preview, a managed Kubernetes plane for GPUs

infrastructureThe Glonce editorsHacker News🔥64Glonce rating64Impact45Novelty50Relevance75Surprise25Players80Credibility75
NVIDIA releases NVKE preview, a managed Kubernetes plane for GPUs

A GitHub repository, dsx-ai-factory/nvke, presents NVIDIA Kubernetes Engine (NVKE). The README describes it as the management plane that platform teams install to create and operate Kubernetes clusters for their customers on their own NVIDIA GPU infrastructure. Each workload cluster gets its own Kubernetes control plane, etcd and node pools. The operator runs the NVKE management plane, and customers use their clusters through the Kubernetes API and kubectl. The project is under Apache License 2.0, with a 2026 NVIDIA CORPORATION & AFFILIATES copyright line.

The status is stated up front: NVKE is a preview. APIs, configuration and behavior can change between releases, and the README says NVIDIA does not recommend NVKE for production workloads.

The README is explicit about what NVKE is not: it 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 on it then provisions and lifecycle-manages many isolated workload clusters on demand. An installation therefore has two kinds of cluster. The management cluster runs the management plane and infrastructure controllers; the nvkectl tool forms it from a temporary seed cluster, and installing the management-plane services is a separate step. Workload clusters run customer applications, and the management plane creates and operates them through NVIDIA Infra Controller (NICo), Cluster API and a node agent.

The README lists four capabilities. First, it creates independent workload clusters: they are requested through the NVKE API or nvkectl, and NVKE provisions machines through NICo. Second, it manages cluster lifecycle: create, scale and delete node pools, upgrade Kubernetes and replace nodes, with manual steps and preview limits on replacement and upgrades. Third, it establishes node identity: nodes attest through SPIFFE/SPIRE using their TPM before receiving join credentials, and encryption of etcd data is available per cluster. Fourth, it connects builds to deployment: the companion project NVIDIA Kubernetes Recipes (NVKR) builds node images and NVKE components from pinned inputs, NVKE deploys those artifacts, and NVKR owns their build and release evidence.

For a real installation, the operator supplies the machines, networking, node images and registries, and configures access to NICo, which is the sole compatible infrastructure provider at preview. A platform and version reference lists validated combinations and pinned dependencies.

The README spells out several preview limits. API authentication is off and the API uses plain HTTP after installation, so operators must configure authentication and protected access 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.

For a trial, a local quickstart runs on a Linux host or VM, amd64 or arm64, with at least 6 CPUs, 12 GiB of free memory and 100 GiB of free disk. It requires Docker, virtual TPM support and the tools listed in the guide, and it walks through workload-cluster creation until the cluster's nodes report Ready. On macOS the evaluation runs inside a Colima virtual machine. To build from source you need Go 1.27.3 (the version pinned in go.work) and GNU Make 4 or later; make build writes nine binaries to bin/: agent, updater, controller, nvkectl, capd-reboot-controller, nvke-os-image-check, management-engine, tenant-api and operator-api. Building does not start a management plane or create a cluster. Contributions are welcome, with CONTRIBUTING.md as the starting point and an issue first for substantial work.

Key facts

  • NVKE is a management plane for running managed Kubernetes on your own NVIDIA GPU infrastructure; each workload cluster gets its own control plane, etcd and node pools.
  • It is a preview: APIs, configuration and behavior can change between releases, and the README says NVIDIA does not recommend it for production workloads.
  • Machines are provisioned through NVIDIA Infra Controller (NICo), the sole compatible infrastructure provider at preview; the companion project NVKR builds node images and components.
  • Preview limits include API authentication off, plain HTTP after installation, and no management-plane upgrade or backup/restore procedure.
  • The code is under Apache License 2.0; the local quickstart needs 6 CPUs, 12 GiB of free memory and 100 GiB of free disk, and make build produces nine binaries.

Why it matters

NVKE is a first-party NVIDIA project that lets an operator run its own managed Kubernetes service on NVIDIA GPU machines, with isolated clusters per customer. The README frames it as a self-hosting factory rather than a Kubernetes distribution or a Cluster API flavour: a throwaway seed cluster builds a management cluster, which then creates and manages many workload clusters. It is published as open source under Apache License 2.0, but only as a preview.

Who it affects

The README addresses platform teams that install the management plane to offer Kubernetes clusters to their customers, and those customers who then use their clusters through the Kubernetes API and kubectl. Real installations need NICo as the infrastructure provider and operator-supplied machines, networking, node images and registries. Contributors are invited through CONTRIBUTING.md.

How to use it

Start with the local quickstart on 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. Follow it through workload-cluster creation and confirm the nodes report Ready. On macOS, use the local site guide, which runs the evaluation inside a Colima VM. Clusters are requested through the NVKE API or nvkectl. To build, use Go 1.27.3 and GNU Make 4 or later, then run make build in a clone of the repository. The licence is Apache License 2.0.

How solid is it

The source is the project's own README in a repository carrying a 2026 NVIDIA copyright line, so the descriptions are the vendor's. It labels itself a preview and says APIs, configuration and behavior can change between releases. No release date, version number or general-availability timeline is given. No benchmarks, performance figures, customer names or pricing are given.

Risks and caveats

NVIDIA does not recommend NVKE for production workloads. After installation, API authentication is off and the API uses plain HTTP, so authentication and protected access must be configured before callers are admitted. 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 Kubernetes upgrades involve manual steps and preview limits. NICo is the only compatible infrastructure provider at preview.

“NVKE is not a Kubernetes distribution and not a Cluster API flavour.”

— NVKE README