Nvidia announced on Monday that it will sell a technology enabling faster chip-to-chip communication, essential for building and deploying AI systems. The new version of its NVLink technology, called NVLink Fusion, will be made available to other chip designers to help create powerful custom AI setups by linking multiple chips together.

Companies like Marvell Technology and MediaTek plan to adopt NVLink Fusion for their custom chip projects. Originally developed years ago, NVLink facilitates the exchange of large data volumes between chips, such as in Nvidia’s GB200, which combines two Blackwell GPUs with a Grace processor. Nvidia CEO Jensen Huang revealed these plans during his keynote at the Taipei Music Center, where the Computex AI exhibition runs from May 20-23.

Huang also announced plans to establish a Taiwan headquarters near Taipei. His speech highlighted Nvidia’s evolution from a graphics chip maker focused on gaming to the leading producer of AI chips powering the recent surge in AI technology, especially after ChatGPT’s 2022 launch. Nvidia is also developing CPUs based on Arm technology for Windows, as Reuters has reported.

At Computex last year, Huang attracted significant attention, dubbed “Jensanity” by the media. In March, at Nvidia’s developer conference, he outlined strategies to shift from building large AI models to supporting applications running on those models.

Nvidia introduced several new AI chips, including the Blackwell Ultra coming later this year, followed by Rubin and Feynman processors due by 2028. The company also launched the DGX Spark desktop AI system for researchers, which Huang said is already in production and will be available soon.

With around 1,400 exhibitors, Computex is Asia’s first major computer and chip industry event since U.S. tariff threats aimed at boosting domestic manufacturing.