NVIDIA has officially announced its latest GPU architecture, named Rubin, marking a significant milestone in the semiconductor industry. This announcement comes as the industry is still digesting the impact of the Blackwell series, signaling NVIDIA’s transition to a rapid, one-year product release cycle to meet the insatiable global demand for Artificial Intelligence power.
According to the latest technical specifications, Rubin-based products have already entered the early stages of production. AI computing solutions leveraging the Rubin GPU are expected to be available through partner companies starting in the second half of 2026. The Rubin architecture is the direct successor to the Blackwell series and introduces groundbreaking performance metrics. Specifically, its inference performance using the NVFP4 precision format reaches a staggering 50 petaflops, which represents a five times increase in performance compared to the Blackwell GPU.
What makes the Rubin architecture particularly impressive is its efficiency in design. While the transistor count in Rubin has increased by one point six times relative to Blackwell, the actual performance gains significantly outpace this hardware growth. This achievement is attributed to a completely revamped architectural design that optimizes data flow and computational efficiency. Additionally, the Rubin platform is expected to integrate the upcoming Vera central processing unit and advanced High Bandwidth Memory 4 technology, further solidifying its position at the core of future AI data centers.
