CES 2026 | NVIDIA's New AI Model, Chip, Robotics, Rubin, Alpamayo

NVIDIA CEO Jensen Huang unveiled a suite of groundbreaking innovations at CES 2026, centering on the new Vera Rubin AI platform and the Alpamayo reasoning model for autonomous vehicles.

The Vera Rubin platform, now in full production, is a six-chip, extreme-codesigned AI system designed to deliver up to five times the computing power of the previous Blackwell generation while reducing token generation costs by up to 10 times.

This next-generation AI supercomputer integrates 88 custom Olympus cores per Vera CPU and 336 billion transistors per Rubin GPU, with each system containing a pair of both components. The platform is designed for large-scale AI inference and training, capable of being clustered into "pods" with over 1,000 Rubin chips to form a single AI engine. It features a proprietary data format and includes innovations like "context memory storage" to improve long-form AI interactions.

The system is expected to be deployed by major cloud providers including Microsoft, Amazon Web Services, Google Cloud, and CoreWeave, as well as by AI labs such as OpenAI, Anthropic, Meta, and xAI. Cooling is optimized using hot water, and the chassis can be assembled in just five minutes.

Alpamayo, an open-source family of reasoning models, enables autonomous vehicles to handle complex, unfamiliar driving scenarios through chain-of-thought reasoning, with the first deployment in the 2025 Mercedes-Benz CLA set for early 2026.

This 10-billion-parameter chain-of-thought reasoning model is designed to guide autonomous vehicles through unpredictable situations by breaking down complex scenarios into smaller, manageable problems and explaining its decisions at each step. It is part of NVIDIA’s full-stack DRIVE platform and will be used in the 2025 Mercedes-Benz CLA, which has already earned a five-star safety rating from EuroNCAP.

Alpamayo is also being used in conjunction with NVIDIA’s Cosmos simulation environment to train robots and vehicles using synthetic data from digital twins. The model is open-sourced, including the training data, to promote transparency and trust.

In other highlights of his presentation, Huang was joined on stage by autonomous BDX droids from Star Wars, trained using NVIDIA’s Cosmos and Isaac Sim platforms, demonstrating the company’s progress in physical AI.

NVIDIA also announced partnerships with Siemens to enhance manufacturing through digital twin-based robotics training. Additionally, the company emphasized its focus on inference efficiency, with Rubin chips designed to handle real-time AI responses more effectively than previous generations.

While no new GeForce RTX GPUs were announced during the main keynote, NVIDIA covered its gaming advancements on GeForce On Community Update. This included DLSS 4.5, which introduces a second-generation transformer model to reduce ghosting and shimmering, along with support for 6x multi-frame generation and dynamic frame generation.

G-Sync Pulsar, a new flicker reduction technology, was also announced, promising perceived motion clarity equivalent to 1,000Hz through backlight pulsing and adaptive brightness/color temperature adjustments. Pre-orders for G-Sync Pulsar displays are set to open January 7, 2026.

All announced products by NVIDIA are either in full production or scheduled for rollout in the second half of 2026, with the Vera Rubin platform and Alpamayo already being integrated into early partner systems.