NVIDIA AI Foundation Models AUP

This document sets out the rules for how you can and cannot use NVIDIA NIM software. You can build things with it and offer them as a service through your own products, but you cannot use developer-program software in production, deploy it in shared commercial systems, or use any part of it for emotion recognition. NVIDIA may collect telemetry data from NIM containers, but you can turn that collection off by setting an environment variable.

This document establishes the acceptable use conditions under which NVIDIA NIM licenses its Software, defining permitted and prohibited deployment contexts, hardware restrictions, and data practices. Customers may offer the Software, Derivative Samples, and Derivative Models as a service within a Customer Product on owned or leased infrastructure, but developer-program Software is confined strictly to internal, non-production evaluation and testing. The document imposes absolute prohibitions on emotion recognition use and multi-user commercial deployments, restricts proprietary components such as CUDA toolkit, TensorRT, cuDNN, NVIDIA drivers, and NVIDIA Omniverse to systems with NVIDIA Platforms, and limits DGX Software and firmware to designated DGX Systems. NVIDIA NIM collects telemetry data in NIM container releases for product configuration, optimization, and improvement, and retains a right to request periodic usage reporting from customers no more than monthly.

For an individual user or customer, this document means that the Software comes with firm boundaries: developer-program Software is for internal testing only and cannot be shipped in a product or run in production, and no use of the Software—or any component—for emotion recognition is permitted under any circumstances. NVIDIA may ask you to report which software your personnel and end users are using, up to once a month by email. Telemetry data is collected in NIM container releases; you can disable this collection by setting the environment variable NIM_TELEMETRY_MODE=0.