Nvidia used the SIGGRAPH 2025 conference to broaden its portfolio of foundation models for so-called physical AI, introducing the 7-billion-parameter Cosmos Reason vision-language model alongside two new Nemotron models—Nemotron Nano 2 and Llama Nemotron Super 1.5. The models are designed to let robots and enterprise agents plan, reason and generate synthetic data, and can run from edge devices powered by Jetson Thor chips to Blackwell-based RTX Pro 6000 workstations and DGX Cloud clusters. Early industrial users include CrowdStrike, Uber and Zoom, which are building smarter AI agents, while Amazon Devices & Services said the new stack is enabling a move toward zero-touch manufacturing through digital twins. Nvidia also highlighted fresh extensions to its Metropolis video-analytics platform and Isaac Sim simulation tools, positioning the software as a bridge between simulation, training and real-world deployment for smart cities, factories and infrastructure.