Catena, Tuesday, May 20, 2025-Nvidia on yesterday (Monday) launched NVLink Fusion, a new chip-to-chip communication technology designed to enhance the speed and efficiency of artificial intelligence systems by tightly linking multiple processors.
Unveiled by CEO Jensen Huang during a keynote at the Computex exhibition in Taipei, NVLink Fusion marks Nvidia’s latest step to solidify its dominance in AI infrastructure.
The technology will be sold to other chipmakers, including MediaTek, Marvell Technology, Alchip, Fujitsu and Qualcomm, to help build high-performance, custom AI platforms.
NVLink, already central to Nvidia’s flagship GB200 system, enables rapid data exchange between its Grace CPU and Blackwell GPUs.
The new Fusion variant extends this architecture across chips from multiple vendors, improving scalability and enabling broader ecosystem collaboration.
Huang also announced the company will open a new Taiwan headquarters in northern Taipei.
He emphasized Nvidia’s transformation from a graphics chip manufacturer into a global leader in AI computing, noting the firm’s expanded focus beyond model training to deploying AI-driven applications.
The company also confirmed its DGX Spark desktop AI workstation, designed for researchers, is in full production and will ship in the coming weeks.
Nvidia’s roadmap includes next-generation Rubin chips, to be followed by Feynman processors in 2028. Huang’s appearance again drew significant public attention, echoing last year’s “Jensanity” fanfare in Taiwan.
Computex, running May 20–23, marks Asia’s first major semiconductor industry event since renewed US pressure on firms to boost domestic manufacturing.
