NVIDIA Launches Open Source Kubernetes Engine for GPU Infrastructure
New NVKE and NVKR projects streamline managed Kubernetes services for AI factories.
1New Open Source Frameworks
NVIDIA has introduced two open source projects, NVIDIA Kubernetes Engine (NVKE) and NVIDIA Kubernetes Recipes (NVKR), to support AI factories. These tools are designed to build and operate managed Kubernetes services specifically on GPU-accelerated infrastructure. Both projects are currently available in public preview under the Apache 2.0 license for community evaluation.
2Separation of Management Tasks
The architecture divides responsibilities between cluster management and software builds. NVKE provides the software plane for provisioning and managing clusters via a REST API. Meanwhile, NVKR focuses on creating reproducible node and container images, ensuring that validated combinations of software versions and GPU hardware generations are consistently deployed across the environment.
3Bootstrapping the Management Plane
Deployment begins with nvkectl, which initializes a temporary seed cluster on a bootstrap host. This seed cluster facilitates the creation of a permanent management cluster. Once established, the management plane uses the NVIDIA Infra Controller (NICo) to request bare-metal servers and form the foundation for customer-facing workload clusters with independent control planes.
4Reproducible Image Pipeline
NVKR enables teams to build node images from version-pinned sources, recording every input for future reproducibility. This pipeline compiles the NVKE node agent and OVN-Kubernetes from source, integrating SPIRE agents for security. The resulting QCOW2 disk images allow factories to deploy validated Kubernetes versions without manually managing cloud-init configurations or OS-level settings.
5Customer Cluster Provisioning
AI factories publish release sets, allowing customers to select specific Kubernetes versions and configurations. Through the NVKE API, teams submit cluster requests that define node pools, instance types, and VPC prefixes. The system automates the lifecycle, including kubeadm initialization and the installation of essential networking components like NodeLocal DNSCache for each tenant.
6Security and Telemetry Integration
Security is enforced through attested SPIFFE identities, which nodes must present before receiving join credentials. For observability, the NVKE telemetry pipeline utilizes OpenTelemetry-based Helm charts. This system collects logs, metrics, and traces from the infrastructure, exporting them over OTLP to a monitoring backend to ensure visibility into cluster health and performance.
7Ecosystem and Partner Adoption
Industry partners including Firmus, SK Telecom, Mirantis, Rafay, and Spectro Cloud are already integrating these components into their managed offerings. Netris is incorporating NVKE into its network automation solutions. These collaborations aim to help AI factories provide scalable, GPU-accelerated Kubernetes environments while maintaining control over their specific operational and service requirements.
8Current Preview Limitations
Both NVKE and NVKR are in public preview and are not yet recommended for production workloads. NVIDIA notes that APIs, configuration behaviors, and project roadmaps are subject to change between releases. Teams are encouraged to review the preview status page and known limitations before evaluating the stack against their specific infrastructure requirements.
9Strategic Impact on AI
By standardizing GPU infrastructure management, these tools address the complexity of scaling AI workloads. The ability to automate bare-metal lifecycle management and provide consistent Kubernetes environments allows factories to focus on service delivery. Future contributions from the open source community will likely shape the evolution of the DSX AI Factory platform.