Categories
TAIPEI 20th May 2025 – Nvidia has announced plans to market its next-generation chip-to-chip interconnecting technology, dubbed NVLink Fusion. The move is an important advancement in the speed of communication inside AI systems. This announcement came during the Computex 2025 event in Taipei, Taiwan, where CEO Jensen Huang also outlined the future plans of this company’s AI processors.
NVLink Fusion was developed at Nvidia’s Israeli research center. It is based on technology acquired from Mellanox in 2019. This technology allows faster data transfers between chips. Speedy transfers are essential for high-performance AI in both training and inference. Unlike previous versions, the latest NVLink Fusion will be available for licensing. Third-party chipmakers such as Marvell Technology and MediaTek can license this technology. The move marks a shift in Nvidia’s approach to chip communication technology.
“This is the first time we’re opening NVLink to the industry,” Huang declared. “By allowing integration with CPUs and AI accelerators from other vendors, we’re expanding the AI ecosystem.”
This shift indicates Nvidia’s move to a more flexible hardware platform designed to accelerate the development of large-scale AI systems by supporting custom configurations created by external partners.
Alongside revealing NVLink Fusion, Nvidia introduced new chip architectures for the future—Blackwell Ultra in 2025, Rubin in 2026, and Feynman scheduled for 2027—in a way of reaffirming its determination to push the boundaries of AI computing.
It also said that it will set up a new corporate headquarters in the northern region of Taipei to highlight its growing presence across Asia and continued investments within the industry of semiconductors in Taiwan.
For the latest information and news, follow AtoZTechWorld or find us on Facebook Page.
