NVIDIA Corporation offers hosted inference for AI models at build.nvidia.com. Inference means running a trained model to get its output, as opposed to training it. The hosted endpoints are NVIDIA NIM APIs. NIM is NVIDIA's family of prebuilt, optimized inference microservices: packaged services that each run one model behind a standard API.
A developer picks a model from the catalog, for example one of NVIDIA's Nemotron models or a DeepSeek model, and calls its endpoint over the internet; the hosted APIs run on NVIDIA DGX Cloud. The same NIM microservices can be downloaded and self-hosted on NVIDIA-accelerated machines in the cloud, a data center or a workstation, so an application can move from the hosted endpoint to its own servers without code changes. The site also lists agent skills, blueprints (reference workflows with code samples) and the NemoClaw agent stack.
Access is free for development and testing: hosted APIs for prototyping, and self-hosted NIM for research and development as part of the NVIDIA Developer Program. For production use, NVIDIA directs customers to an NVIDIA AI Enterprise license.