While many anticipated that Jensen Huang would hold his high-stakes reveals for the GTC conference in March, the NVIDIA CEO chose to strike much sooner.
At CES 2026, instead of routine announcements, NVIDIA dropped a technological bombshell by officially unveiling the Vera Rubin platform to the global audience.
This accelerated debut serves as a clear signal to the industry: the AI arms race is gaining momentum, and NVIDIA has no intention of slowing down. Rubin represents more than a mere evolution; it is a quantum leap that makes previous supercomputing architectures look rudimentary. The platform fully embraces HBM4 memory and proprietary processor cores, delivering a staggering boost in overall performance.
The cornerstone of the system is the Rubin GPU. Engineered with a massive 336 billion transistors, this chip is designed for total supremacy in AI-driven workloads. By utilizing HBM4, it achieves a memory bandwidth of 22 TB/s per chip—nearly tripling the capacity of the previous generation. In high-speed inference tasks, Rubin reaches 50 PFLOPS (NVFP4), a fivefold increase over the Blackwell architecture’s capabilities.
Source: iXBT.games