ModelDeployment is a namespaced resource representing a deployed AI modelinstance. It references a ModelClass (which owns the Model identity andthe runtime template); credentials are scoped via spec.licenseCredential(NGC) and spec.imagePullCredentials (image pull). Model-pull credentialslive on the referenced Model's spec.source.modelPullCredential. Thecontroller creates and manages an InferenceService on the GPU cluster foreach ModelDeployment, and exposes the resulting inference endpoint viastatus.inferenceEndpointURI.