NVIDIA Run:ai is open-source software that decides how AI jobs share graphics chips (GPUs) across Kubernetes clusters, the systems that run programs over many computers.
In plain words
Companies that own many GPUs want several teams to use them without waste. Run:ai places training, inference and development jobs onto GPUs using fair-share rules, priorities and preemption, and it can hand out a fraction of one GPU to a small job. A central control plane sets quotas per department or project and watches every connected cluster. NVIDIA has released the full platform as open source, along with four parts that can be used separately, such as KAI Scheduler.
A quick example
Imagine a university with 64 GPUs shared by three research groups. A scheduler like this could guarantee each group its quota, lend idle GPUs to whoever is busy, and take them back when the owner's jobs arrive.
Where you'll meet it
- Sharing GPU clusters between teams in companies and AI clouds.
- Running training and inference jobs on Kubernetes.
Related
Sources
- DocsNVIDIA Run:ai Is Now Open Source as Part of NVIDIA DSX AI Factory Platform, NVIDIA Developer · read 13 Oct 2026
- Docsdsx-ai-factory/run-ai, NVIDIA DSX AI Factory (GitHub) · read 13 Oct 2026
Checked by WhatIsAI against the sources above · 13 Oct 2026. A full five-minute explainer with a diagram may follow. Spotted an error? Report it.