Wearing his signature black leather jacket, NVIDIA CEO Jensen Huang took the CES 2026 stage and declared that the "screen era" of generative AI has peaked. AI is now "growing hands and legs, gaining eyes" to enter the physical world. He unveiled the new Rubin computing platform and the open-source Alpamayo model family, pushing inference compute to new heights.
Rubin GPU: 288GB HBM4 and 3.6 ExaFLOPS Inference
The Rubin platform is built for reasoning. The Rubin GPU features 288GB of HBM4 memory with 22 TB/s bandwidth. When 72 Rubin GPUs team up with 36 Vera CPUs in the Vera Rubin NVL72 rack, inference performance reaches 3,600 PFLOPS (3.6 ExaFLOPS). Huang emphasized "test-time scaling," noting that while training costs may flatten, inference demand will grow exponentially. Volume production timelines are unannounced, but the technology gap is already wide.
Alpamayo: Bringing Chain-of-Thought to Autonomous Driving
The open-source Alpamayo family stole the spotlight. Its flagship Alpamayo 1 (R1) integrates chain-of-thought reasoning into vision, language, and action models. When facing unexpected road hazards like broken barriers or fallen cargo, R1 doesn't just react instantly — it reasons through causes and selects the best evasive path. Mercedes-Benz will adopt the system in its upcoming CLA model, while Uber and JLR are also partners. The autonomous driving market has entered a "reasoning race."
From Protein Folding to Climate Simulation: Compute Reaches Science
Huang didn't stop at cars. He highlighted protein synthesis model OpenFold3 and the Earth-2 climate platform, calling DGX Cloud "a microscope for modern science". Drug discovery and precision weather forecasting require massive matrix operations that the Rubin architecture enables in real time. NVIDIA is transforming from a graphics card vendor into an infrastructure provider for global research and industrial automation.
Policy and Capital: AI Factories Ride the Automation Wave
Against a policy backdrop emphasizing manufacturing reshoring, Huang pitched "AI factories" and robotics to capture domestic automation opportunities. His repeated mention of test-time scaling paints a rising compute demand curve for investors. Market watchers believe that once Rubin chips go into mass production and Alpamayo models leave the lab, Industry 4.0 will shift from concept to assembly lines.
The CES 2026 stage lights dimmed, but NVIDIA's vision — AI leaving screens to illuminate streets, factories, and labs — is now clear. Next time in Las Vegas, rows of robots with silicon brains may greet visitors, and Wall Street will price PFLOPS with updated valuation models.