At CES 2026, instead of routine announcements, a technological bombshell was presented — the Vera Rubin platform was officially shown to the public.
The decision to accelerate the presentation served as a signal to the market: the AI arms race is not slowing down, and NVIDIA has no intention of yielding to competitors. Rubin is not an evolution, but a true leap that makes previous supercomputers look like toys. The platform fully utilizes HBM4 memory and proprietary processor cores, delivering impressive performance.
The heart of the system is the Rubin GPU. This chip, with 336 billion transistors, is designed for absolute superiority in AI computing. Thanks to HBM4, bandwidth reaches 22 TB/s per chip, nearly three times higher than the previous generation. In high-speed inference tasks, Rubin achieves 50 PFLOPS (NVFP4), which is five times the capability of Blackwell.