Nvidia Announces “ChatGPT Moment for Physical AI” with Alpamayo and Vera Rubin

At CES 2026, Nvidia CEO Jensen Huang declared the industry has reached the "ChatGPT moment for physical AI," signaling a shift toward autonomous systems capable of reasoning in the real world. The keynote introduced the Alpamayo open-source VLA models and the Vera Rubin supercomputer platform, designed to provide the massive compute required for the next generation of robotics and autonomous vehicles.

​Alpamayo: Open-Source Reasoning for Autonomous Driving

​The Alpamayo 1 model, a 10-billion parameter Vision-Language-Action (VLA) model, is designed to help developers solve the "long tail" of driving scenarios—complex edge cases that traditional AI often fails to navigate.

• ​Chain-of-Thought Reasoning: Alpamayo 1 integrates reasoning capabilities, allowing it to process visual data and generate trajectories while providing a logical rationale for its decisions.
• ​Teacher Model Strategy: Rather than acting as the primary runtime brain for consumer cars, Alpamayo serves as a powerful "teacher" model to train and validate production-ready autonomous driving (AV) stacks.
• ​Open Ecosystem: To accelerate AV development, Nvidia has released the Alpamayo 1 weights on Hugging Face, accompanied by the AlpaSim simulation framework and the Physical AI Open Datasets, which include over 1,700 hours of diverse driving data.

​Mercedes-Benz and the MB.OS Integration

​The collaboration between Nvidia and Mercedes-Benz enters a new phase with the upcoming CLA model. This vehicle will be the first to feature a full-stack autonomous driving suite powered by Nvidia technology and integrated into the MB.OS platform.

• ​Global Rollout: The launch is scheduled to begin in the U.S. market before a broader global expansion.
• ​Scalable AI: The partnership leverages Nvidia's end-to-end infrastructure, using Alpamayo-derived insights to enhance safety and autonomy features in production vehicles.

​Vera Rubin: Infrastructure for the AI Industrial Revolution

​To power these advancements, Nvidia introduced the Vera Rubin supercomputer platform, the successor to the Blackwell architecture.

• ​Architecture and Performance: The platform features the Vera CPU and the Rubin GPU. The NVL72 rack system is engineered for extreme inference and training workloads, with the Rubin GPU providing 50 PFLOPS of NVFP4 compute—a five-fold increase in inference performance over the previous generation.
• ​Efficiency Gains: The platform is designed to handle Mixture-of-Experts (MoE) models with significantly higher efficiency, projecting a tenfold reduction in the cost of inference per token.
• ​AI Superfactories: Microsoft has been confirmed as a key partner in deploying the Fairwater AI superfactories based on the Rubin NVL72 architecture, treating the entire data center as a single unit of compute.

​Expanding the Robotics Frontier

​Beyond vehicles, Nvidia's "Physical AI" strategy encompasses the Cosmos and GR00T frameworks for humanoid and industrial robotics. By providing standardized open-source models and the heavy-duty hardware of the Vera Rubin platform, Nvidia aims to establish the foundational infrastructure for all embodied intelligence, from logistics to consumer service robots.