NVIDIA has released its full Run:ai platform as open-source software, giving AI infrastructure teams access to tools for scheduling and sharing GPUs across Kubernetes clusters. Four components can also be adopted separately, making this a substantial shift in how teams can use the platform, not just a new label on the box.
NVIDIA Watch analysis
What happened
The release includes KAI Scheduler for GPU-aware Kubernetes scheduling, GPU Fractioning Operator for sharing CUDA GPUs, KAI Resource Management for quotas and organisational hierarchies, and Workload Map for a consistent workload-management contract. NVIDIA says teams can use the integrated platform or choose individual components.
The code is available through NVIDIA’s DSX AI Factory GitHub organisation, and the integrated stack can be installed from NVIDIA NGC using a Helm chart. NVIDIA says the Community Product does not require a paid Run:ai subscription or support entitlement, though NGC downloads require acceptance of applicable NVIDIA software licence terms. Enterprise support is not included in that Community Product.
NVIDIA also names eight partners integrating the platform into their offerings, including Crusoe, Mirantis, Nebius and Rafay. Its announcement describes the release and partner plans; it does not establish how much GPU utilisation or throughput operators will gain in practice.
Why it matters
GPU orchestration is the plumbing that decides which AI workloads get access to scarce accelerator capacity, and how that capacity is shared. Opening the full platform, while allowing teams to adopt individual pieces, gives infrastructure operators more room to inspect, adapt or build on those tools instead of treating Run:ai as a single bundled product.
The practical distinction is worth keeping clear: access to the Community Product is not the same as NVIDIA enterprise support, and licence terms still apply. But a lower-friction route to the software could broaden who can test it and give cloud providers more options for packaging AI infrastructure.
Our read
This is a meaningful opening of AI infrastructure software, with useful modularity rather than a token code release. The next test is whether the components are straightforward to run outside NVIDIA’s own commercial offerings, and whether operators can demonstrate benefits on their actual workloads. “Open source” is a good invitation; the licence, support and operational details are where the real relationship begins.
What to watch
- How the four components are licensed and maintained as the open-source projects develop.
- Whether partners make their integrations generally available and on what terms.
- Whether operators publish results for GPU utilisation and workload throughput.
- Which Run:ai capabilities remain tied to NVIDIA-supported commercial offerings.
Discussion spark: Would you adopt GPU orchestration software like Run:ai without enterprise support, or is support part of the price of running it in production?
Sources and evidence
- Vertiv CDU qualifies as NVIDIA DSX Ready – Comms Business (2 October 2026, 09:25 UTC)
- NVIDIA Run:ai Is Now Open Source as Part of NVIDIA DSX AI Factory Platform – NVIDIA Developer (2 October 2026, 09:25 UTC)
Independent WittyWires tracker for public updates about NVIDIA. Not affiliated with or endorsed by NVIDIA; this is not an official account.