Key points
- The full Run:ai platform, including the KAI Scheduler and GPU fractioning operator, is now available on GitHub under the Apache 2.0 license.
- Platform teams can inspect, fork, and self-host the code for production GPU scheduling, quotas, and workload management on Kubernetes or OpenShift.
- The release integrates into the NVIDIA DSX AI Factory platform but lacks enterprise support from NVIDIA in this community edition.
The short version
- NVIDIA has published the source code for its Run:ai GPU orchestration platform on GitHub, licensing it under Apache 2.0.
- The open-source release includes the KAI Scheduler, GPU fractioning capabilities, and resource management tools for Kubernetes clusters.
- Organizations can now self-host the platform to manage AI workloads without relying on a commercial subscription, though enterprise support is not included.
The details
According to MarkTechPost, NVIDIA released the complete Run:ai platform as open-source software, placing the code on GitHub under the Apache 2.0 license. This move allows platform teams to inspect, fork, and self-host the stack for production GPU scheduling, quota management, and workload handling without needing a paid subscription. The release integrates Run:ai into the broader NVIDIA DSX AI Factory platform, shifting control of these scheduling mechanisms directly into developers' hands for independent operation.
The monorepo contains the control plane, cluster services, installer, command-line interface, and an MCP server. Local development builds involve approximately 50 Go services. The platform operates on Kubernetes or OpenShift clusters equipped with NVIDIA GPUs. For local testing, the system runs on macOS or Linux with eight simulated Tesla-A10 GPUs, requiring no actual GPU hardware. However, local builds on Linux demand 16 GB of RAM and 100 GB of disk space.
NVIDIA divided the release into four distinct components that teams can adopt separately. The KAI Scheduler, now a CNCF Sandbox project, handles Kubernetes-native scheduling tasks such as gang scheduling, time-based fair-share, topology awareness, bin-packing, and DRA support. The GPU Fractioning Operator introduces CUDA 13.4-native GPU fractions to Kubernetes, with the driver enforcing hard memory limits per workload.
The KAI Resource Management component adds project, department, quota, and node pool capabilities. NVIDIA’s example demonstrates guaranteeing 50% of a cluster’s resources to a specific research team. The Workload Map is a Go library describing workload types like PyTorchJob, RayCluster, and KServe services. It works offline and does not handle scheduling, queuing, or autoscaling. NVIDIA published no platform-level benchmarks for this specific open-source release.
The context
NVIDIA Run:ai is an AI workload and GPU orchestration platform designed for Kubernetes. It schedules development, training, and inference workloads across shared GPU clusters without replacing Kubernetes itself. The control plane manages workload submission, resource management, and monitoring, while cluster services run on each Kubernetes cluster to add GPU-aware scheduling. Users interact with the platform via a web UI, CLI, REST API, and an MCP server for AI assistants.
This release follows NVIDIA’s broader push into open-source AI infrastructure. As TechSec Headlines reported on 2026-10-15, NVIDIA pledged open-source tools and training resources to help European researchers and startups access advanced AI capabilities. The Run:ai open-source launch extends that strategy by removing barriers to entry for GPU orchestration, allowing organizations to build their own infrastructure rather than purchasing it.
What to watch
- Monitor the GitHub repository for updates to the KAI Scheduler and GPU Fractioning Operator, as these are critical for production stability.
- Observe community adoption rates and third-party contributions, which will determine the long-term viability of the open-source version.
- Track any announcements regarding enterprise support options that might emerge alongside the community edition.
- Watch for performance reports from early adopters, as NVIDIA provided no official benchmarks for this release.
What it means for organisations
Platform engineering and security teams must evaluate whether self-hosting this open-source stack aligns with their operational capacity. Organizations should determine who will maintain the code, apply security patches, and manage updates, as NVIDIA does not provide enterprise support for this version. IT directors need to assess the risk of running unmanaged code in production environments and ensure their teams have the expertise to customize and secure the platform. Clear ownership and maintenance procedures must be established before deployment.
What to do and how to stay safe: NVIDIA
- Audit the open-source code for security vulnerabilities before deploying it to production GPU clusters.
- Verify that your Kubernetes environment meets the hardware requirements for local builds and production runs.
- Establish a process for monitoring the GitHub repository for critical updates and security fixes.
- Ensure your team understands the differences between this community edition and the previous proprietary version.
Step-by-step guide: Secure Home Routers for Small Business Networks
General security guidance from the TechSec Headlines newsroom. It is not confirmed advice from the organisations named in this story.
Frequently asked questions
Does NVIDIA provide enterprise support for the open-source Run:ai platform?
No, the community edition has no NVIDIA enterprise support, meaning users are responsible for maintenance and troubleshooting.
Can I run Run:ai without a physical GPU for local development?
Yes, the local development environment runs on macOS or Linux with simulated Tesla-A10 GPUs and requires no actual GPU hardware.
What license is the Run:ai platform released under?
The source code is released under the Apache 2.0 license on GitHub.
