AI news · October 15, 2026
NVIDIA opens Run:ai for shared GPU orchestration
Shared GPU capacity is becoming an open infrastructure problem: teams can inspect the scheduler, fractioning and quota layers instead of accepting one closed control plane.
- modular components
- 4
- KAI Scheduler license
- Apache 2.0
- named ecosystem partners
- 8
NVIDIA said on October 13 that the full Run:ai platform is now available as open source as part of the DSX AI Factory platform. The release includes KAI Scheduler for GPU-aware Kubernetes scheduling, a GPU Fractioning Operator, KAI Resource Management for organisational quotas and hierarchy, and Workload Map for a common workload-management contract. NVIDIA says eight ecosystem partners are integrating the platform, while KAI Scheduler is released under Apache 2.0 and contributed to the CNCF Sandbox.
The practical angle is modular adoption. A platform team can inspect or adopt the scheduler without taking the complete stack, or use fractioning to share underused GPU capacity between notebooks, smaller models and inference jobs. Licensing and support boundaries still apply to NVIDIA commercial offerings, so test the open components against the existing Kubernetes topology before treating them as a replacement.
What you can do with it
Try it: run the KAI Scheduler repository against a small, non-production Kubernetes cluster. Compare queue time, GPU utilisation and fairness for one multi-pod training job, one interactive notebook and one inference workload before changing production scheduling.
Our take
This is a strong infrastructure release because it exposes the control points that determine GPU cost and access. The value is the option to measure scheduling and fractioning in the existing cluster. Treat support terms and migration effort as separate decisions.
Links Try it
Source: NVIDIA Technical Blog — Made With Models writes the brief; the reporting is theirs.