Nvidia doubles down
Nvidia unveiled a full‑stack robotics play at GTC — new Isaac tools, a Cosmos world model and GR00T robotics foundation models aimed at production‑scale ‘physical AI’, plus a booming networking business now bringing in multibillion‑dollar revenue. The pitch: make simulation, open models, and high‑speed networking the backbone for fleets of robots, AVs and industrial automation. (trendforce.com)
NVIDIA said dozens of global robotics leaders — including ABB Robotics, AGIBOT, Agility, CMR Surgical, FANUC, Figure, Hexagon Robotics, KUKA, Medtronic, Skild AI, Universal Robots, World Labs and YASKAWA — will build on NVIDIA’s Physical AI stack announced at GTC. (nvidianews.nvidia.com) The company pushed new GR00T family builds into production: GR00T N1.7 is being distributed while Jensen Huang previewed GR00T N2, a DreamZero‑derived “world‑action model” that NVIDIA says enables robots to complete new tasks in unfamiliar environments more than twice as often as leading vision‑language‑action models and currently tops MolmoSpaces and RoboArena rankings. (nvidianews.nvidia.com) NVIDIA also upgraded Cosmos to Cosmos 3 and published Omniverse/Isaac blueprints for massive synthetic‑data generation, naming early adopters such as Agility Robotics, Figure and Skild AI for post‑training and validation workflows. (nvidianews.nvidia.com) On autonomy, NVIDIA expanded DRIVE Hyperion partnerships to BYD, Geely, Isuzu and Nissan and committed to a robotaxi deployment with Uber across 28 cities by 2028, beginning service in Los Angeles and San Francisco in the first half of 2027. (investor.nvidia.com) NVIDIA’s data‑center networking arm reported roughly $10.98 billion in networking sales in the company’s fiscal fourth quarter and topped $31 billion for fiscal 2026, a surge the company says places networking as its second major revenue engine behind compute. (cnbc.com) NVIDIA quantified ecosystem scale at GTC — engaging about 110 robot‑brain developers — and released open toolkits and a GR00T‑Dreams GitHub blueprint that generates large synthetic trajectory datasets from single images plus language prompts. (therobotreport.com)
