Nvidia Makes Run:ai Open Source for Its DSX AI Factory

Nvidia just open-sourced the full Run:ai platform, the GPU orchestration software it closed a roughly $700 million acquisition for at the end of 2024. The code is live on GitHub under the DSX AI Factory org and on NGC as a free Community Product, with no paid Run:ai subscription required for that build.

The release breaks into four pieces you can take separately: KAI Scheduler (already Apache 2.0 and a CNCF Sandbox project), a GPU Fractioning Operator that wires CUDA 13.4 memory isolation into Kubernetes, KAI Resource Management for team quotas and node pools, and Workload Map, a contract so PyTorchJobs, Ray clusters and custom CRDs look the same to schedulers and dashboards. Eight DSX partners, including Crusoe, Nebius, Saturn Cloud and vCluster, are shipping it in their stacks.

Enterprise customers still get the commercial Run:ai path through Mission Control, AI Enterprise and cloud marketplaces. Opening the scheduler and sharing layer is Nvidia telling every AI cloud the same thing: run more jobs on the GPUs you already bought from us.

— via NVIDIA Developer Blog.