Teams can use the full platform or select tools, but enterprise support still requires a separate offering.
NVIDIA says the full Run:ai platform is now available as open-source software for managing artificial-intelligence workloads and GPUs on Kubernetes. The announcement covers an immediate release, not an independently measured improvement in speed or GPU use.
Run:ai is designed to manage work across the AI lifecycle, including data processing, model training and inference. Platform teams can install the integrated stack, build on its code, or adopt four components separately.
The KAI Scheduler handles GPU-focused placement. NVIDIA says it supports gang scheduling, which starts multipod jobs together, along with fair sharing, quotas, topology-aware placement, bin-packing and GPU sharing. That could help teams divide a shared cluster without hand-managing every job.
KAI Resource Management organizes capacity by projects, departments and node pools. GPU Fractioning Operator lets multiple workloads share GPU capacity through CUDA-native fractions, while Workload Map gives different Kubernetes workload types a common structure for scheduling, monitoring and policy.
The code is available through the NVIDIA Run:ai repository in the NVIDIA DSX AI Factory GitHub organization. NVIDIA also offers the integrated platform through NGC as a Community Product, with a Helm chart for installation.
The Community Product does not require a paid Run:ai subscription or NVIDIA support entitlement, according to NVIDIA. Enterprise support is not included, and downloads require acceptance of NVIDIA’s applicable license terms.
For operators, the practical change is more choice: evaluate the full stack or fill one capability gap. Before production use, teams should check licensing, compatibility, maintenance and support needs, then compare scheduling and utilization results against their current Kubernetes tools.