NVIDIA has introduced its new chip, Rubin CPX, which aims to redefine standards in the world of artificial intelligence. Designed specifically for programming with millions of tokens and generative video production, the chip targets breakthroughs in speed and efficiency.
Unlike traditional GPUs, Rubin CPX integrates video encoding, decoding, and inference into a single chip. This enables the processing of long-form video content, advanced video search, and high-quality generative video creation.
The new NVIDIA Vera Rubin NVL144 CPX platform delivers 8 exaflops of AI computing power and 100 terabytes of fast memory within a single server rack—making it approximately 7.5 times more powerful than the existing NVIDIA GB300 NVL72 systems. The chip is expected to hit the market by the end of 2026.
Built on the next-generation Rubin architecture, which will replace the current Blackwell architecture, Rubin CPX is already being tested by AI companies such as Cursor, Runway, and Magic.
NVIDIA CEO Jensen Huang commented:
“The Vera Rubin platform will open a new frontier in AI computing. If RTX revolutionized graphics, Rubin CPX will do the same for artificial intelligence. It will be the first CUDA GPU capable of reasoning over millions of tokens.”
