NVIDIA has made the full Run:ai platform open source. The Kubernetes software that decides how AI workloads share expensive GPUs — previously a paid, proprietary product — can now be downloaded, inspected and self-hosted for free as part of NVIDIA’s DSX AI Factory platform. Why it matters: GPU time is the scarcest resource in AI, and the tool that squeezes more work out of each chip just became free for anyone to run.
October 13, 2026
What NVIDIA released
Run:ai is an orchestration layer for Kubernetes clusters. Instead of one team’s job reserving a whole GPU while other work waits in line, it schedules jobs across a shared fleet, lets several workloads split a single GPU, and applies quotas so teams can borrow idle capacity and hand it back when its owner needs it. NVIDIA says the open release contains the complete platform, not a stripped-down edition.
Four pieces can also be adopted on their own. KAI Scheduler is the scheduling engine, released under the Apache 2.0 licence and contributed to the Cloud Native Computing Foundation as a Sandbox project. The GPU Fractioning Operator brings GPU sharing with hard, driver-enforced memory limits to Kubernetes, replacing Run:ai’s older proprietary approach. KAI Resource Management handles the organisational layer — projects, departments, quotas and node pools. Workload Map is a library that lets tools handle different workload types, such as PyTorch jobs and Ray clusters, in a consistent way.
How you get it, and what stays paid
The code sits in a public GitHub repository under NVIDIA’s DSX AI Factory organisation, and the integrated stack is also offered through NVIDIA’s NGC catalogue as a Community Product with a Helm chart for installation. That community version needs no paid Run:ai subscription — but it also comes with no NVIDIA enterprise support. Supported commercial versions continue to exist, including through NVIDIA AI Enterprise and cloud marketplaces, so NVIDIA is giving away the software while still selling the safety net.
Partners moved the same day: Nebius, Mirantis, OpenNebula, Lambda, Crusoe, Rafay and vCluster are among the companies already packaging or integrating the open platform into their own AI cloud offerings. CTech notes the move also keeps a promise made when NVIDIA acquired the Israeli-founded company in late 2024, in a deal estimated at around $800 million, when its founders said they intended to open-source the software.
Why it matters
For anyone running AI infrastructure, this changes the maths. GPU orchestration was a line item — a licence on top of already expensive hardware. Now the orchestration layer is a free, inspectable foundation that cloud providers and enterprises can build on, which puts pressure on every vendor selling similar scheduling software. For smaller teams, the practical win is utilisation: sharing GPUs safely, with memory isolation and fair quotas, means the hardware you already own can carry more training and inference work. The catch to weigh is support — if a scheduling failure idles a costly cluster at 3 a.m., community code will not answer the phone. That trade-off, free software versus paid assurance, is exactly the “everything AI, tested” question buyers now get to answer for themselves.
Frequently asked questions
Is Run:ai now completely free?
The full platform is available as open source and as a free Community Product through NGC, with no paid subscription required. NVIDIA enterprise support is not included; supported commercial versions are still sold separately.
Do I have to take the whole platform?
No. The four main components — KAI Scheduler, GPU Fractioning Operator, KAI Resource Management and Workload Map — can each be adopted independently.
What does Run:ai actually do?
It schedules AI workloads on Kubernetes, allocates and shares GPU capacity across teams, enforces quotas, and manages jobs from data processing and training through to inference.
Sources: NVIDIA developer blog, CTech, HPCwire