NVIDIA open-sources the full Run:ai GPU orchestration platform

NVIDIA has open-sourced the full Run:ai platform, which schedules, pools and fractionally shares GPUs across Kubernetes clusters. Run:ai became NVIDIA's after a high-profile acquisition that drew regulatory scrutiny. Releasing it openly turns a paid enterprise product into free infrastructure for anyone running GPU clusters.

Run:ai sits between Kubernetes and the hardware. It handles fair-share queues, GPU fractioning for inference, preemption of training jobs, and quota management across teams. For AI platform teams, that is the layer that decides utilization, and utilization decides cost. Open-sourcing it lets organizations self-host and modify the scheduler instead of buying licenses.

The move follows NVIDIA's familiar playbook: give away the software that makes its GPUs easier to saturate. It competes with open schedulers like Kueue and Volcano and with cloud-native managed offerings. Making Run:ai free could make it the default GPU scheduler, which keeps NVIDIA hardware central even as hyperscalers push custom silicon.

Open questions: whether Run:ai supports non-NVIDIA accelerators in practice, what license applies, and whether NVIDIA keeps premium features in a commercial tier.