NVIDIA Releases Run:ai GPU Orchestration Platform as Open Source

NVIDIA announced on October 13, 2026 that it is releasing the complete Run:ai platform as open source software, making production-grade AI workload and GPU orchestration capabilities available to customers, developers and ecosystem partners without requiring a paid subscription or enterprise support entitlement, according to a post by Tanya Lenz on the NVIDIA Developer Blog.

The release expands the NVIDIA DSX AI Factory Platform by establishing Run:ai as an open foundation for the broader ecosystem. NVIDIA completed its acquisition of Run:ai on December 30, 2024, as reported by TechCrunch at that time.

Platform Capabilities and Components

Run:ai is a Kubernetes-based platform that manages AI workload scheduling and GPU resource allocation for interactive development, model training and inference. According to NVIDIA documentation current as of October 13, 2026, the platform's capabilities include GPU pooling and sharing, fractional GPU allocation, workload prioritization, organizational quotas, role-based access control and utilization monitoring. These are intended operational capabilities, not performance guarantees.

The October 13 release includes the integrated Run:ai platform and four modular components that can be adopted independently: KAI Scheduler, GPU Fractioning Operator, KAI Resource Management and Workload Map. NVIDIA released KAI Scheduler under the Apache 2.0 license and contributed it to the Cloud Native Computing Foundation as a Sandbox project, according to the announcement.

KAI Scheduler coordinates gang scheduling with in-memory simulation, fair-share and quota allocation, topology awareness, GPU sharing across multiple methods including MIG and MPS, and Dynamic Resource Allocation support. The GPU Fractioning Operator enables CUDA 13.4-native GPU fractioning on Kubernetes with hard memory isolation enforced by the CUDA driver. KAI Resource Management provides an organizational layer mapping GPU resources to projects, departments and teams with defined capacity shares and node pools for heterogeneous hardware. Workload Map defines a contract for treating diverse Kubernetes workload types—including PyTorchJob, RayCluster, JobSet and KServe InferenceService—as first-class objects throughout their lifecycle.

Availability and Ecosystem Adoption

The complete open source Run:ai platform source code is available in the NVIDIA DSX AI Factory GitHub organization. It is also distributed as a Community Product on NVIDIA NGC with a Helm chart for installation. The Run:ai source code is governed by applicable open source licenses; the NGC Community Product requires acceptance of the NVIDIA Software License Agreement and Product-Specific Terms for NVIDIA AI Products. No paid subscription is required for the Community Product, and NVIDIA enterprise support is not included, according to the announcement.

NVIDIA-supported commercial offerings of Run:ai continue to be available as a standalone product, through NVIDIA Mission Control and NVIDIA AI Enterprise, and on select cloud marketplaces, with availability, licensing and support terms varying by offering.

Eight ecosystem partners detailed their integrations in announcements linked from the October 13 post. Crusoe integrates Run:ai with Crusoe Managed Kubernetes. Mirantis extends k0rdent AI with Run:ai for lifecycle-managed infrastructure provisioning. Nebius offers Run:ai as a one-click application in Nebius AI Cloud. OpenNebula Systems integrates Run:ai with the OpenNebula AI Cloud Platform through OneKS. Rafay is expanding its integration to deliver Run:ai as a fully managed service with automated infrastructure and control-plane management. Saturn Cloud integrates Run:ai to help operators launch white-label AI inference services through its Token Factory platform; Saturn Cloud announced this integration on September 17, 2026, according to PR Newswire. Spectro Cloud integrates Run:ai into PaletteAI for lifecycle automation. vCluster provides Run:ai through pre-validated deployment templates with per-tenant environments.

Related Open Source Projects

The announcement also detailed two complementary open source projects. Model Streamer reads model weights concurrently and streams them directly into GPU memory to reduce cold-start latency for inference workloads. It is integrated upstream in vLLM and SGLang, ships in the AWS vLLM Deep Learning Container, and is documented by Google Cloud and Microsoft Azure. Storage platform vendors WEKA, VAST Data and DDN have independently published benchmarks; WEKA reported loading into GPU memory in 3.5 seconds, approximately 40 percent faster than local NVMe, while DDN reported under four seconds for the 65.5 GB Qwen3-32B model across eight GPUs. These are vendor-reported figures for specific configurations, not universal performance claims.

Grove provides a Kubernetes API for orchestrating multi-component workloads including agentic pipelines, disaggregated serving and mixture-of-experts training. It offers topology-aware autoscaling and hierarchical gang scheduling, and on NVIDIA GB200 NVL72 it manages ComputeDomain resources automatically. Grove is a modular part of NVIDIA Dynamo that can be used independently.

Context and Limitations

The October 13, 2026 announcement marks a shift from Run:ai's prior proprietary distribution to an open source model within the DSX AI Factory ecosystem. Platform teams can inspect the code, contribute to development, deploy the integrated stack or adopt individual components. The community distribution does not include enterprise support; organizations requiring support entitlements must use NVIDIA's commercial offerings.

NVIDIA stated that all KAI Scheduler development happens in the community repository and that improvements flow through the same codebase. The company continues to support the project's maintainers and stated it will work with the broader ecosystem to advance AI workload scheduling.

Source: developer.nvidia.com.