During the keynote speech at CES 2026, Jensen Huang, CEO of NVIDIA, made a groundbreaking announcement: the NVIDIA Vera Rubin chip architecture has officially commenced full-scale production. This architecture represents a holistic overhaul, encompassing enhancements from the CPU and GPU to the very methods of interconnectivity. It integrates six cutting-edge chips: the Vera CPU, Rubin GPU, NVLink 6 Switch Chip, ConnectX-9 SuperNIC, BlueField-4 DPU, and Spectrum-6 Ethernet Switch Chip.

The Rubin GPU stands out with remarkable performance gains. Its NVFP4 inference computing capability soars to an impressive 50 PFLOPS, marking a fivefold increase over the Blackwell architecture. When it comes to training computing power, it reaches 35 PFLOPS, 3.5 times that of Blackwell. Despite these leaps in performance, the transistor count has only risen by 1.6 times, showcasing a significant leap in architectural efficiency. Moreover, through its meticulously coordinated design, the Vera Rubin platform seamlessly integrates these six chips. This integration ensures comprehensive coverage across computing, networking, storage, and security, thereby offering robust support for the age of AI inference.