NVIDIA has made its Run:ai GPU orchestration platform fully open source under the Apache 2.0 license. The release is aimed at Kubernetes-based environments, giving teams the ability to manage GPU resources more flexibly.
The open-source package includes several key components: the KAI Scheduler for job scheduling, CUDA 13.4 GPU fractioning to split GPUs into smaller units, quota management for controlling resource usage, and a local Kind development environment for testing. These tools together address common challenges in running AI workloads on Kubernetes.
By releasing the full platform under a permissive license, NVIDIA is lowering the barrier for organizations that want to build custom GPU orchestration layers. The move could accelerate adoption of Kubernetes as the standard for AI infrastructure, though the source does not provide further details on roadmap or community governance.