One-sentence summary
NVIDIA announced on October 13, 2026 that the Run:ai platform is open source under the Apache 2.0 license, with its Kubernetes-native GPU scheduler KAI Scheduler donated to the CNCF Sandbox; this guide explains how to obtain the open-source components, deploy them in a Kubernetes cluster, and configure GPU quotas and fractioning for multi-team sharing.
Use Cases
- Infrastructure teams that share one GPU cluster across multiple teams and need fair scheduling and quota management.
- Platform engineers who want GPU fractioning for AI training and inference workloads on Kubernetes to improve GPU utilization.
- Developers who want to evaluate Run:ai scheduling capabilities without buying the commercial platform.
Steps
- Confirm the open-source scope and license: the Run:ai platform is open source under Apache 2.0; the core scheduling engine KAI Scheduler is a Kubernetes-native GPU scheduler donated to the CNCF Sandbox; check component and third-party dependency licenses before redistribution.
- Obtain the KAI Scheduler source: get source code, release versions and Helm charts from the kai-scheduler/KAI-Scheduler repository on GitHub.
- Prepare the Kubernetes cluster: KAI Scheduler targets Kubernetes clusters running AI workloads; confirm the cluster version and GPU nodes (for example, with the NVIDIA GPU Operator installed) before deployment.
- Deploy the scheduler: use the official installation method (Helm or manifests) to deploy KAI Scheduler as a cluster-level scheduler with GPU-aware scheduling.
- Configure GPU quotas and fractioning: set GPU quotas per team or namespace, and enable fractioning so multiple tasks share a single physical GPU to improve utilization.
- Try it locally: without a ready cluster, build a local Kubernetes environment with Kind and deploy KAI Scheduler to verify basic scheduling behavior.
Notes
- KAI Scheduler is the open-source scheduling engine of the Run:ai platform; other components and advanced features of the commercial platform are still delivered through NVIDIA Run:ai.
- The functional boundary between the open-source and commercial versions follows NVIDIA's official documentation; verify component and third-party dependency licenses before redistribution.
- Run:ai joined NVIDIA after the acquisition received European regulatory clearance in December 2024; this open-sourcing publishes the full platform codebase (TechRepublic, October 15, 2026).
References
- NVIDIA Technical Blog: NVIDIA Run:ai Is Now Open Source as Part of NVIDIA DSX AI Factory Platform (October 13, 2026)link
- NVIDIA Technical Blog: NVIDIA Open Sources Run:ai Scheduler to Foster Community Collaborationlink
- TechRepublic: Nvidia Run Is Now Open Source: What's Free and Paid (October 15, 2026)link