XNews

NVIDIA CES announcements and hardware platforms

NVIDIA CES announcements and hardware platforms

NVIDIA announced major hardware and model initiatives at CES 2026. - Vera Rubin platform announced for consumer and enterprise deployment with shipments planned in second half of 2026. - Rubin GPUs claim training a 100T-parameter model with one-quarter the Blackwell chips and reducing training time to one month. - NVIDIA stated inference cost reductions up to 10x on new hardware. - NVIDIA also introduced an open model ecosystem spanning Nemotron (agents), Cosmos (physics), Alpamayo (autonomous driving), Isaac GR00T (robotics), and Clara (biomed). - CES demos included a robot assistant demo built with Pipecat AI and Hugging Face Reachy Mini illustrating mixed local/cloud operation and voice interfaces.

Key insights

Vera Rubin Platform Targets Massive Training Scale with Lower Chip Counts

NVIDIA announced the Vera Rubin platform, claiming it will train a 100-trillion-parameter model using one-quarter the chip count previously required on Blackwell and reduce training time to about one month. NVIDIA stated that the new platform will lower inference costs by up to 10x and is scheduled for shipment in the second half of 2026.

NVIDIA Publishes Cross-Industry Model Suite

NVIDIA introduced five domain models/tools at CES: Nemotron for agent AI, Cosmos for physical simulation/physics AI, Alpamayo for autonomous driving, Isaac GR00T for robotics, and Clara for biomed. The announcements frame NVIDIA as providing both hardware and domain-specific model tooling to accelerate real-world AI systems.

Robot Assistant Demos Illustrate Hybrid Local/Cloud and Voice-Centric UX

CES demonstrations showcased a robot assistant built by Pipecat AI using Hugging Face Reachy Mini, emphasizing multi-model, multimodal and hybrid local/cloud operation with voice interfaces. The demos highlight NVIDIA's ecosystem push toward embedded robotics and on-device inference.

Source posts