At CES 2026 in Las Vegas, Nvidia CEO Jensen Huang confirmed that the company’s Vera Rubin AI supercomputing platform is already in full production. Comprising six distinct chip types, the flagship configuration houses 72 of Nvidia’s latest graphics processing units and 36 new central processors per device. These chips are being tested in Nvidia’s labs and by leading AI firms ahead of a public launch later this year. Huang demonstrated how pods containing more than 1,000 Rubin chips can be networked together, delivering a combined throughput that Nvidia says will accelerate training of large-scale models in simulated environments and physical applications such as robotics and manufacturing.

During his keynote, Huang highlighted that the next-generation Rubin chips deliver up to five times the AI computing capability compared to the previous generation, despite only increasing transistor count by 60%. This leap is driven by a proprietary data format that Nvidia is promoting as a new industry standard for AI workloads. The company also introduced a context memory storage layer designed to reduce latency in conversational AI systems handling complex, multi-turn interactions by more efficiently managing long contextual histories across millions of concurrent users.

Nvidia unveiled Alpamayo, an open-source AI model tailored for real-time decision-making in autonomous vehicles. Huang announced that both the model and its training dataset will be publicly released, enabling automakers and Tier 1 suppliers to validate safety and performance metrics independently. Alpamayo leverages neural architectures optimized on Vera Rubin hardware and supports traceable decision logs for post-drive analysis. Nvidia projects that automotive partners deploying Alpamayo can reduce development cycle times by up to 30%, accelerating road testing and regulatory approval processes.