This is based on NVIDIA’s own announcement on the NVIDIA Technical Blog, 13 October 2026.
NVIDIA has released the complete NVIDIA Run:ai platform as open source software [1].
“Today, NVIDIA is making the full NVIDIA Run:ai platform available as open source software, giving customers, developers, and ecosystem partners access to production-grade AI workload and GPU orchestration capabilities,” NVIDIA said [1]. The release is part of the NVIDIA DSX AI Factory platform.
What Run:ai does, in plain words
Companies that train and run AI models often share a pool of expensive graphics processors (GPUs) between many teams and jobs. GPU orchestration is the software that decides which job gets which GPU, when, and for how long, so the chips are kept busy and no team grabs everything. Run:ai does this on Kubernetes, the widely used system for running software in containers [1].
GPU fractioning means splitting one GPU so several smaller jobs can share it, each getting its own slice of the chip’s memory. NVIDIA says this suits work that doesn’t need a whole GPU, such as notebooks and smaller models [1].
The four parts
NVIDIA says the platform has four components that can also be used on their own [1]:
- KAI Scheduler, the engine that decides where jobs run. NVIDIA had already released it under the Apache 2.0 licence and contributed it to the Cloud Native Computing Foundation (CNCF) as a Sandbox project [1].
- GPU Fractioning Operator, which lets Kubernetes use the GPU-sharing feature now built into CUDA 13.4. NVIDIA says its own “proprietary implementation has been retired” in favour of this native capability [1].
- KAI Resource Management, which handles projects, departments, quotas and groups of machines, so a company can, for example, guarantee one team a share of the cluster [1].
- Workload Map, a common way for tools to understand different kinds of AI jobs. NVIDIA says it “ships as a Go library” [1].
TSN checked the GitHub repositories NVIDIA links to. The main platform and all four components show the Apache 2.0 licence [2].
Is it free?
The source code is on GitHub under open source licences [1][2]. NVIDIA also offers the full platform on its NGC catalogue as a “Community Product” with a Helm chart for installing it [1].
There are caveats. NVIDIA says downloading the Community Product from NGC “requires acceptance of the applicable NVIDIA Software License Agreement and Product-Specific Terms for NVIDIA AI Products.” No paid Run:ai subscription is needed, but “NVIDIA enterprise support is not included” [1].
NVIDIA’s paid, supported versions continue, including as a standalone product and through NVIDIA Mission Control and NVIDIA AI Enterprise [1].
Who is using it
NVIDIA named eight partners integrating the open source platform: Crusoe, Mirantis, Nebius, OpenNebula Systems, Rafay, Saturn Cloud, Spectro Cloud and vCluster [1]. Nebius, for example, offers it as a one-click deployment in Nebius AI Cloud, and Rafay plans to offer it as a fully managed service [1].
Related projects
NVIDIA also pointed to two related open source projects [1]:
- Model Streamer, which loads AI model weights into GPU memory faster. It is built into the vLLM and SGLang inference engines. According to NVIDIA and Microsoft, whose Azure Blob Storage integration was jointly developed, published benchmarks on eight NVIDIA H100 virtual machines showed 3x to 6x faster loading than the default loader, and a steady 80 Gbps compared with about 40 Gbps [1]. TSN has not tested this.
- Grove, a Kubernetes tool for running AI jobs made of several connected parts. It is part of NVIDIA Dynamo but can be used separately [1].
What this means
- The whole platform is now open. NVIDIA says anyone can inspect the code, contribute, or build products on it [1].
- You can take just one piece. Each of the four components can be adopted without the rest [1].
- “Free” has limits. The code is Apache 2.0, but the NGC download requires accepting NVIDIA’s licence terms, and enterprise support isn’t included [1][2].
Sources
- NVIDIA Technical Blog (Ekin Karabulut and Sam Heywood), “NVIDIA Run:ai Is Now Open Source as Part of NVIDIA DSX AI Factory Platform”, 13 October 2026. https://developer.nvidia.com/blog/nvidia-runai-is-now-open-source-as-part-of-nvidia-dsx-ai-factory-platform/
- GitHub repositories, licence information checked 13 October 2026: https://github.com/dsx-ai-factory/run-ai ; https://github.com/kai-scheduler/KAI-Scheduler ; https://github.com/kai-scheduler/gpu-fractioning ; https://github.com/kai-scheduler/kai-resource-management ; https://github.com/dsx-ai-factory/workload-map
