NVIDIA Open-Sources the Full Run:ai Platform: GPU Orchestration Leaves the Subscription, Four Components Stand Alone
Synopsis
NVIDIA has released the complete Run:ai platform as open source software within its DSX AI Factory platform, giving customers, developers and ecosystem partners access to production-grade AI workload and GPU orchestration. The platform previously shipped as a commercial product that dynamically allocated GPU resources on Kubernetes and managed workloads across the full lifecycle, from data processing and training to inference. The release includes four components that can also be adopted independently: KAI Scheduler, GPU Fractioning Operator, KAI Resource Management and Workload Map, so partners can take the whole platform, build on it, or integrate single pieces.
Interpretation
The full Run:ai platform is now available as open source, so platform teams and ecosystem partners can adopt it whole or take one component to fill a specific gap in an existing platform. Run:ai had been a commercial product that helped organizations run AI workloads on Kubernetes at scale, dynamically allocating GPU resources and managing workloads from data processing and training through inference. The release includes the complete platform code plus four independently adoptable components, with source in the NVIDIA DSX AI Factory GitHub organization and a Helm chart on NGC as a Community Product for installing the integrated stack.
The four components each address a distinct problem: scheduling, GPU fractioning, resource ownership and workload description. KAI Scheduler is the Kubernetes-native scheduler that powers Run:ai, released under Apache 2.0 and contributed to CNCF as a Sandbox project, with all scheduler development happening in the community repository. KAI Scheduler coordinates all-or-nothing placement of multipod jobs, fair-share and quota allocation, topology awareness and bin-packing, GPU sharing, and Dynamic Resource Allocation; GPU Fractioning Operator brings CUDA 13.4-native GPU fractions to Kubernetes, with the driver enforcing a hard memory limit per workload and a metrics exporter for per-pod visibility.
The other two components govern who gets what at the organizational level and give many workload types one consistent description. KAI Resource Management governs who gets what after the scheduler decides placement, mapping GPU capacity to projects, departments and node pools with fair-share and quota semantics. Workload Map gives PyTorchJob, RayCluster, JobSet, KServe InferenceService and other workload types a consistent description, ships as a Go library that works offline on unstructured objects without adding anything to the cluster, and is deliberately narrow: it describes workload structure and does not schedule, queue, autoscale or propagate across clusters.
Perspective
The release targets platform teams, AI clouds and DSX ecosystem partners running AI workloads on Kubernetes, who can adopt the platform whole or integrate individual components. KAI Scheduler can be deployed in an existing cluster alongside the current software stack without the complete platform; Workload Map works offline as a Go library without cluster changes. The NGC Community Product requires no paid subscription or NVIDIA support entitlement but does not include enterprise support; Enterprise AI Factory deployments remain available through NVIDIA-supported commercial offerings, with availability, licensing and support terms varying by offering.
The open source Community Product carries no NVIDIA enterprise support, and deployments needing support still go through commercial offerings. Workload Map is explicitly limited to describing workload structure and does not schedule, queue, autoscale or propagate across clusters. GPU Fractioning Operator depends on CUDA 13.4-native fractioning, and the earlier proprietary implementation has been retired. The release does not give an item-by-item feature comparison between the open source and commercial versions.