NVIDIA Open Sources Run:ai Platform as Part of DSX AI Factory

NVIDIA has released the full NVIDIA Run:ai platform as open source software, expanding the NVIDIA DSX AI Factory Platform with a foundation for AI workload and GPU orchestration. The release gives customers, developers, and ecosystem partners access to production-grade orchestration capabilities and is intended to help the broader ecosystem bring differentiated AI factory solutions to market faster.

Four Modular Components

The open source release includes four components that can be adopted independently. KAI Scheduler provides GPU-aware Kubernetes scheduling with gang scheduling, fair-share and quota allocation, topology awareness, GPU sharing, and Dynamic Resource Allocation support. NVIDIA released KAI Scheduler under the Apache 2.0 license and contributed it to CNCF as a Sandbox project, with all scheduler development happening in the community repository.

GPU Fractioning Operator brings CUDA-native GPU fractioning to Kubernetes, replacing the earlier proprietary implementation with the native capability available in CUDA 13.4. The operator includes a metrics exporter for per-pod visibility and remains scheduler-agnostic while integrating with KAI Scheduler for placement decisions. KAI Resource Management adds an organizational layer with projects, departments, quota semantics, and node pools for heterogeneous clusters. Workload Map defines a consistent workload-management contract, describing pods, components, and roles so platforms can schedule, observe, and govern diverse workload types without hardcoding each custom resource definition.

Ecosystem Adoption

Eight NVIDIA DSX ecosystem partners are integrating the open source platform into their offerings. Crusoe combines it with Crusoe Managed Kubernetes, Mirantis extends k0rdent AI with it, and Nebius offers it as a one-click deployment in Nebius AI Cloud. OpenNebula Systems integrates it through OneKS, Rafay is expanding its integration into a fully managed service, and Saturn Cloud uses it to launch white-label inference services through its Token Factory platform. Spectro Cloud integrates it into PaletteAI, while vCluster provides pre-validated deployment templates that stand up a complete environment per tenant.

The complete platform is available in the NVIDIA DSX AI Factory GitHub organization and on NVIDIA NGC as a Community Product with a Helm chart for installing the integrated stack. No paid subscription or support entitlement is required for the Community Product, though NVIDIA enterprise support is not included. Commercial offerings remain available through NVIDIA Mission Control, NVIDIA AI Enterprise, and select cloud marketplaces.

Related Open Source Projects

Two related projects complement the platform. Model Streamer reduces cold-start latency by streaming model weights directly into GPU memory and is integrated upstream in vLLM and SGLang, with documented support across AWS, Google Cloud, and Microsoft Azure. Grove provides a Kubernetes API for orchestrating multicomponent inference workloads, offering topology-aware autoscaling and hierarchical gang scheduling, and is a modular part of NVIDIA Dynamo.

Sources: Nvidia Developer Blog