At CES 2026, Nvidia unveiled its next-generation GPU architecture, 'Rubin.' This architecture comprises six chips and is claimed to be 3.5 times faster for AI training tasks and 5 times faster for inference tasks compared to the previous Blackwell architecture. The Rubin architecture also features improvements to the Bluefield and NVLink systems to address storage and interconnection bottlenecks, and includes a new Vera CPU designed for agentic reasoning. It aims to meet the growing memory demands of AI systems, particularly for agentic AI and long-term tasks, by introducing a new tier of external storage for more efficient scaling.

Nvidia announced the 'Alpamayo' family of open-source AI models at CES 2026, designed to enable autonomous vehicles (AVs) to 'think like humans.' This 10-billion-parameter, chain-of-thought reasoning-based vision-language-action (VLA) model allows AVs to reason through complex driving scenarios, such as traffic light outages. Dubbed the 'ChatGPT moment for physical AI,' Alpamayo signifies a shift from pattern recognition to reasoning-based driving systems. The model is available on Hugging Face, empowering developers to build safer and more scalable AVs.

At CES 2026, Nvidia announced a comprehensive suite of robot foundation models, simulation frameworks, and edge computing hardware, aiming to become the default platform for generalist robotics, akin to Android's role in smartphones. This strategy reflects the broader industry shift of AI moving from the cloud into the physical world, enabling machines to reason, plan, and act in real-world environments. Central to this strategy are new open robot foundation models released on Hugging Face, designed to support generalist robotics, including world models for synthetic data generation and vision-language models for understanding physical surroundings.