NVIDIA CEO Jensen Huang unveiled Rubin, a next-generation “extreme-codesigned” AI platform that succeeds the Blackwell architecture. Built from the data center outward, the Rubin ecosystem integrates six brand-new chips, including the Rubin GPU, the 88-core Vera CPU, and the NVLink 6 Switch into a single, unified AI supercomputer.
According to NVIDIA, the platform is purpose-built for the era of Agentic AI and advanced reasoning, delivering 50 petaflops of inference performance and a staggering 10-fold reduction in the cost of generating AI tokens. Huang confirmed the hardware is already in production, with the first systems expected to roll out to partners like Microsoft, AWS, and Google Cloud in the second half of 2026.
