NVIDIA Releases AICR v1.0 for Stable GPU Cluster Configurations

NVIDIA released AI Cluster Runtime (AICR) version 1.0 on October 6, 2026. The release provides version-locked, validated recipes for GPU-accelerated Kubernetes clusters, aiming to make cluster configuration reproducible and verifiable.

AICR v1.0 defines a stable compatibility contract for its CLI, REST API, Go SDK, bundle layout, and artifact schemas. This contract means integrators can rely on these interfaces without unexpected changes. Breaking changes to stable interfaces now require a new major version, and the project has added committed compatibility baselines and merge-blocking checks for public interfaces.

Core capabilities

The release centers on four capabilities: Snapshot, Recipe, Bundle, and Validation. A Snapshot records observed cluster state, including OS, kernel, GPU, and topology. A Recipe pins component versions and declares validation phases. A Bundle renders a recipe into deployment artifacts for Helm, Argo CD, Flux, or Helmfile. Validation compares observed state to the recipe and can run conformance and performance checks.

Validation evidence is signed by the hardware it was tested on and displayed on a live dashboard. Operators can run the aicr evidence verify command to check signed validation evidence. The dashboard lets operators filter recipes by service, GPU generation, OS, workload intent, and platform.

Ecosystem and governance

Pulumi Labs offers an AICR provider for infrastructure-as-code, and Mirantis's k0rdent integration packages AICR for multi-cluster management. Over 100 distinct contributors have participated, nearly half from outside NVIDIA. Contributors may propose recipes for unsupported hardware, OS, or services and submit signed evidence.

The Go SDK package github.com/NVIDIA/aicr/pkg/client/v1 exposes the supported workflow without internal imports. Issues, feature requests, and bug reports are handled through the GitHub issue tracker. AICR's public interfaces are governed by a v1.x compatibility policy for integrators.

For inference engineers and edge AI developers, AICR v1.0 offers a way to lock down cluster configurations and verify that the running environment matches the intended setup. The signed evidence provides a tamper-evident record of what was tested, which can be useful for auditing and troubleshooting. The compatibility contract reduces the risk of breakage when updating tooling or integrating with other systems.