AICR 1.0 makes GPU cluster recipes a stable integration surface

NVIDIA’s AI Cluster Runtime pins compatible GPU cluster components into recipes, renders deployment bundles and validates running clusters against them. Version 1.0 defines compatibility rules for its public interfaces, including the command line, REST API and Go SDK. Published validation evidence records the hardware configuration and checks behind a recipe, rather than treating successful installation as proof.

NVIDIA · Oct 6, 2026

Why it matters

Platform teams can separate desired configuration, deployment and verification while keeping their existing Kubernetes delivery tools.

More context

Try a recipe matching your service, GPU and operating system, then inspect its status and signed evidence. Helm, Argo CD, Flux or Helmfile applies the generated bundle; AICR does not itself reconcile the cluster. The stable Go client supports integrations without importing internal packages.

Keep in mind

Validation evidence applies to the tested configuration and declared checks, not every possible cluster. Operators still need suitable infrastructure and must explicitly verify artifacts and results. We have not deployed or audited a recipe.

Original read