During the opening of CES 2026, NVIDIA unveiled the Rubin platform, which it described as its first AI platform designed with full integration across six chips, and announced that it had entered full production. The company also introduced the open Alpamayo family of reasoning models for developing autonomous vehicles, in a presentation delivered by NVIDIA founder and CEO Jensen Huang at Fontainebleau Las Vegas.
Huang said computing is changing as a result of accelerated computing and AI, and that approximately $10 trillion of computing from the past decade is being upgraded to suit this approach. According to him, Rubin aims to reduce the cost of token generation to approximately one-tenth the cost of the previous platform, making the deployment of large-scale AI systems more economical.
Rubin Platform and Data Center Components
The Rubin platform includes a set of components designed to work together, starting with data centers, including:
- Rubin GPU units with up to 50 petaflops for NVFP4 inference.
- Vera CPU units for handling data movement and agentic tasks.
- NVLink 6 networks for scaling within the system, and Spectrum-X Ethernet Photonics networks for scaling across systems.
- ConnectX-9 SuperNIC and BlueField-4 DPU units.
Huang explained that this integration across chips, racks, networks, storage and software is intended to eliminate bottlenecks when scaling AI to large sizes. NVIDIA also introduced the Inference Context Memory Storage Platform for AI-native storage, a KV-cache memory layer designed for long-context inference. The company says it delivers five times as many tokens per second, five times the performance per dollar of total cost of ownership, and five times the energy efficiency.
Open Models for Different Applications
NVIDIA said its open models, trained on its supercomputers, cover six major areas: Clara for healthcare, Earth-2 for climate science, Nemotron for reasoning and multimodal AI, Cosmos for robotics and simulation, GR00T for embodied intelligence, and Alpamayo for autonomous driving.
According to Huang, these models enable companies and developers to create, evaluate, safeguard and deploy models. Through this approach, NVIDIA aims to build an ecosystem that different companies, industries and countries can rely on.
Autonomous Driving and Physical AI
Alpamayo includes the Alpamayo R1 model, which NVIDIA described as the first open VLA model for autonomous-driving reasoning, along with AlpaSim, an open simulation framework for testing autonomous vehicles with a high degree of accuracy. The company said the suite enables the development of systems capable of Level 4 autonomy.
Huang announced that the first passenger vehicle to use Alpamayo, built on the full NVIDIA DRIVE platform for autonomous vehicles, will be the new Mercedes-Benz CLA, which is expected to reach the roads soon, with AI-defined driving in the United States this year. He also highlighted the open and shared DRIVE Hyperion platform, which is ready for Level 4 and is adopted by automakers, suppliers and robotaxi operators around the world.
In robotics, NVIDIA demonstrated robot training through Isaac Sim and Isaac Lab in simulated virtual worlds, and also showcased its collaborations with Synopsys, Cadence, Boston Dynamics and Franka. The company also announced an expansion of its partnership with Siemens, integrating the NVIDIA stack with Siemens industrial software to support physical AI from design and simulation through production.
Personal Computing and Gaming
NVIDIA demonstrated an AI agent running locally on the DGX Spark super desktop computer, embodied through a Reachy Mini robot using Hugging Face models. The company said DGX Spark delivers up to 2.6 times the performance for large models, with new support for Lightricks LTX-2 and FLUX models, and NVIDIA AI Enterprise expected to become available.
In gaming, NVIDIA announced DLSS 4.5, which adds Dynamic Multi Frame Generation, 6X Multi Frame Generation mode, and a second-generation Transformer model for DLSS Super Resolution. It said more than 250 games and applications support DLSS 4, alongside updates to RTX Remix Logic, NVIDIA ACE, G-SYNC Pulsar displays, and new GeForce NOW applications for Linux and Amazon Fire TV devices.