NVIDIA Kubernetes Engine (NVKE) is a management plane for operating self-hosted Kubernetes clusters on NVIDIA GPU infrastructure, currently in preview with explicit limitations against production use. It provisions independent workload clusters, each with dedicated control planes, etcd, and node pools, using the NVIDIA Infra Controller (NICo) as the sole infrastructure provider. The architecture employs a throwaway seed cluster to bootstrap a management cluster running the NVKE control plane, which then lifecycle-manages workload clusters via Cluster API and a node agent. Nodes attest identity through SPIFFE/SPIRE using TPM before receiving join credentials, with optional per-cluster etcd encryption. The companion project NVKR builds node images and components from pinned inputs, while NVKE deploys the artifacts. Local evaluation requires a Linux host or VM (amd64/arm64) with 6 CPUs, 12 GiB RAM, 100 GiB disk, Docker, and virtual TPM; macOS uses Colima. Building from source uses Go 1.27.3 and GNU Make 4, producing nine binaries including nvkectl, controller, agent, and management-engine. Preview limitations include disabled API authentication, plain HTTP, no management-plane upgrade or backup/restore, and FDE node images lacking data-at-rest confidentiality. Operators must explicitly configure authentication and protected access before admitting callers.