NVIDIA open-sources the full Run:ai platform under Apache 2.0: GPU scheduling, CUDA-native GPU fractioning and team quotas for Kubernetes

Why it matters

【Key points】

On October 13 NVIDIA open-sourced the full Run:ai platform as part of its DSX AI Factory platform. Run:ai, acquired in late 2024 for about $800 million, orchestrates AI workloads and GPUs on Kubernetes and was previously commercial software. The release covers the integrated, self-hostable platform plus four standalone components: KAI Scheduler (a GPU-aware Kubernetes scheduler, already a CNCF Sandbox project), GPU Fractioning Operator (GPU sharing built on native CUDA 13.4 fractioning, replacing the proprietary implementation), KAI Resource Management (team and department quotas and guarantees) and Workload Map (a Go library for describing workload types consistently). Code is under Apache 2.0 in the dsx-ai-factory GitHub organization, with a free community Helm chart on NGC; supported versions remain available commercially.

【Verifiable】

NVIDIA blog (2026-10-13 14:19 UTC): https://developer.nvidia.com/blog/nvidia-runai-is-now-open-source-as-part-of-nvidia-dsx-ai-factory-platform/

Source: https://github.com/dsx-ai-factory/run-ai

NGC community product: https://catalog.ngc.nvidia.com/orgs/nvidia/runai-oss/collections/runai/-

Coverage (Calcalist): https://www.calcalistech.com/ctechnews/article/jqdtr2997

【Our read】

1. GPU utilization drives AI compute cost, and this scheduling, sharing and quota layer used to be bought or built in-house. Smaller clouds and self-run clusters can now adopt it directly.

2. It also pulls infrastructure toward NVIDIA's DSX AI factory blueprint: the more the free orchestration layer spreads, the more standardized deployment around NVIDIA hardware becomes.

3. Real value depends on the gap between community and supported editions and on support for non-NVIDIA hardware, which the announcement does not detail.