NVIDIA NIM is a set of prepackaged inference microservices for running AI models on NVIDIA GPUs. Each NIM is a container that bundles the model, an optimised inference engine, runtime dependencies and standard APIs, which can also be called through the OpenAI client. The containers run in a data centre, on a workstation, at the edge or with a public cloud provider, and scale on Kubernetes.
NVIDIA Corporation is based in Santa Clara, California, and is subject to US law. You can self-host NIM on your own infrastructure, so prompts and data need not leave your environment; production use requires an NVIDIA AI Enterprise licence. The software itself is not open source and the models are licensed separately. NVIDIA also offers hosted API endpoints for prototyping, whose region is not published. The privacy policy cites standard contractual clauses and participation in the Data Privacy Framework for transfers from Europe, and acknowledges that NVIDIA is subject to US law.