Nvidia used the SIGGRAPH 2025 conference in Vancouver to deepen its push into so-called “physical AI,” unveiling Cosmos Reason, a 7-billion-parameter vision-language model that gives robots and cameras the ability to parse video, understand physics and plan next steps. The company also expanded its Nemotron family with the Nano 2 and Llama Nemotron Super 1.5 models, designed to boost scientific reasoning, coding and tool use for autonomous software agents. Early adopters include CrowdStrike, Uber and Zoom, which are testing the new reasoning models in cybersecurity and collaboration applications. Accenture, Belden, DeepHow, Milestone Systems and Telit Cinterion plan to fold Cosmos Reason into smart-city, factory-floor and industrial-inspection systems, underscoring Nvidia’s aim to move AI beyond data-center workloads and into everyday infrastructure. To run the software, Nvidia introduced hardware based on its Blackwell architecture. The RTX Pro 6000 Server Edition GPU brings 96 GB of GDDR7 memory and up to 600 watts of configurable power to widely used 2U rack servers, while new RTX Pro 4000 SFF and RTX Pro 2000 cards target compact workstations. Cisco, Dell, Hewlett Packard Enterprise, Lenovo and Supermicro will all ship systems using the chips, with HPE’s first ProLiant servers scheduled for release on 2 September. Dell simultaneously updated its AI Data Platform—adding Elastic’s vector search and adopting Nvidia reference designs—while Amazon’s Devices & Services division said Nvidia’s digital-twin tools are helping it move toward zero-touch manufacturing. Taken together, the announcements mark Nvidia’s latest bid to extend its dominance in data-center AI into robotics, industrial automation and edge computing.