NVIDIA Open-Sources NVKE and NVKR for Managed Kubernetes

NVIDIA has officially launched two open-source infrastructure tools designed to simplify Kubernetes deployment on GPU-accelerated hardware. The NVIDIA Kubernetes Engine (NVKE) and NVIDIA Kubernetes Recipes (NVKR) were officially released on November 17, 2026 under the Apache 2.0 license, giving AI factories and research organizations the ability to run managed Kubernetes services on their own bare-metal GPU servers without relying on third-party cloud platforms.
Both tools entered public preview today, marking a significant step toward democratizing GPU orchestration for organizations building large-scale AI compute infrastructure. The release addresses a critical pain point: managing containerized AI workloads across distributed GPU clusters has historically required either expensive managed services or deep in-house orchestration expertise.
What NVKE and NVKR Deliver
NVIDIA Kubernetes Engine serves as a managed control plane for Kubernetes clusters running on GPU infrastructure. It automates cluster provisioning, scaling, and lifecycle management tasks that typically require dedicated platform engineering teams. NVKE handles GPU resource allocation, driver compatibility, and node health monitoring out of the box.
NVIDIA Kubernetes Recipes provides pre-built deployment patterns and configuration templates for common AI workload scenarios. The recipes library includes optimized setups for distributed training, inference serving, data preprocessing pipelines, and multi-tenant GPU sharing. Organizations can fork and customize these recipes rather than building orchestration logic from scratch.
Bare-Metal GPU Orchestration
The key differentiator is support for bare-metal deployments. While cloud-based managed Kubernetes services dominate the market, many AI labs and enterprises prefer to own their GPU hardware for cost, latency, or data sovereignty reasons. NVKE and NVKR eliminate the tradeoff between managed convenience and bare-metal control.
The tools integrate with NVIDIA's existing GPU operator and device plugin ecosystem. Administrators can deploy NVKE on existing hardware running NVIDIA GPUs, from H100 clusters to older A100 or V100 installations. The system automatically detects GPU topology, configures CUDA libraries, and applies best-practice scheduler policies.
Release Date and Availability
Both tools were officially launched on November 17, 2026 and are available now through NVIDIA's GitHub repositories. The Apache 2.0 license permits commercial use, modification, and redistribution without royalty obligations. NVIDIA has indicated that the public preview phase will run through Q1 2027, with general availability expected in spring 2027 pending community feedback.
Documentation includes quickstart guides for deploying NVKE on bare-metal servers, example NVKR recipes for PyTorch and TensorFlow workloads, and integration tutorials for popular MLOps platforms like Kubeflow and MLflow. NVIDIA is accepting pull requests and feature proposals through the GitHub issue trackers.
What This Means
This release lowers the barrier for organizations that want managed Kubernetes convenience without surrendering control of their GPU hardware. Research labs with dedicated compute budgets, financial services firms with compliance requirements, and AI startups scaling on owned infrastructure now have an open-source alternative to vendor-locked managed services. The Apache 2.0 licensing ensures that improvements and customizations can flow back to the community, potentially accelerating innovation in GPU orchestration patterns. For NVIDIA, the move strengthens its position in the AI infrastructure stack beyond chip sales, creating an ecosystem play that ties software tooling to its hardware dominance.
on Emergent today





