Nvidia unveiled plans on Monday to license its NVLink Fusion technology to other chip designers, aiming to accelerate chip-to-chip communication critical for building advanced AI systems.
The move, announced by CEO Jensen Huang at Computex in Taipei, will enable companies like Marvell Technology and MediaTek to develop custom AI solutions with interconnected chips, mirroring the architecture of Nvidia’s own GB200 superchip, which pairs Blackwell GPUs with Grace CPUs.
Huang, speaking at the Taipei Music Centre, emphasized Nvidia’s evolution from a gaming graphics specialist to the linchpin of the AI boom, noting that his presentations now focus overwhelmingly on AI infrastructure. The company also revealed plans to establish a Taiwan headquarters in Taipei’s northern suburbs, reinforcing its commitment to the region amid global supply chain tensions.
The NVLink Fusion announcement follows Nvidia’s March unveiling of its Blackwell Ultra and Rubin chip roadmaps, with next-generation Feynman processors slated for 2028. Huang confirmed the imminent release of DGX Spark, a desktop AI system for researchers currently in full production.
Nvidia’s licensing strategy marks a pivot toward ecosystem control, as AI workloads demand ever-faster interchip communication. The move comes as global chip leaders gather at Computex amid U.S. tariff threats aimed at reshoring semiconductor production.