TAIPEI, May 19 — At the Computex AI exhibition, Nvidia announced the launch of NVLink Fusion, a groundbreaking new technology designed to significantly improve communication between AI chips. The company plans to license NVLink Fusion to other chip designers, enabling the development of custom, high-performance AI systems.
The technology is a next-generation evolution of Nvidia’s original NVLink system, which has long been used to transfer massive volumes of data between processors. NVLink is already central to Nvidia’s GB200 platform, which pairs two powerful Blackwell GPUs with a Grace CPU for AI-intensive tasks.
Now, with NVLink Fusion, Nvidia is opening the door for external adoption. Companies like Marvell Technology and MediaTek have already announced plans to incorporate Fusion into their custom AI chip designs. 
Jensen Huang Headlines Computex with Key Announcements
Nvidia CEO Jensen Huang revealed NVLink Fusion during a keynote address at the Taipei Music Center, ahead of Computex, which officially runs from May 20 to 23. The announcement was accompanied by news that Nvidia will establish a new Taiwan headquarters in the northern suburbs of Taipei.
During his keynote, Huang reflected on Nvidia’s transformation from a graphics card manufacturer to the world’s leading AI chipmaker. “There was a time when 90% of my presentations focused on gaming GPUs,” Huang noted. “Now, it’s all about AI.”
Nvidia’s technology has become central to the AI revolution that surged after the release of ChatGPT in 2022, and the company continues to dominate the AI hardware market.
Future Roadmap and Product Rollouts
Nvidia’s vision stretches far into the future. At the company’s annual developer conference in March, Huang laid out a roadmap for upcoming chip generations, including:
-
Blackwell Ultra, launching later in 2025
-
Rubin, set to follow
-
Feynman, expected by 2028
In addition to server-class AI chips, Nvidia also launched DGX Spark, a desktop AI system aimed at researchers and developers. Huang confirmed at Computex that DGX Spark is now in full production, with shipping expected in the coming weeks.
Nvidia is also reportedly developing CPUs compatible with Microsoft Windows, utilizing Arm Holdings’ technology, signaling a growing interest in diversifying beyond GPUs.
The Return of “Jensanity” at Computex
Last year’s Computex saw “Jensanity” sweep across Taiwan, with fans and media swarming Huang during the trade show. With over 1,400 exhibitors expected this year, Computex 2025 marks a major return for tech leaders across the chip and computing sectors—especially as geopolitical shifts drive interest in re-shoring production to the U.S.
