Computex: Nvidia to sell tech to advance AI development, launches new chip-linking tech

Nvidia has unveiled a new technology called NVLink Fusion at Computex in Taiwan. The new technology is aimed at enhancing chip-to-chip communication, a critical component for building and deploying artificial intelligence tools, Reuters reported.

Nvidia said it plans to offer NVLink Fusion to other chipmakers, enabling them to build custom AI systems by linking multiple chips together. The announcement was made by Nvidia CEO Jensen Huang during his keynote at the Taipei Music Center, the venue for the Computex AI exhibition running from 20 to 23 May.

According to Reuters, Marvell Technology and MediaTek are among the companies expected to adopt NVLink Fusion in their upcoming chip designs. Nvidia also named Alchip, Fujitsu and Qualcomm as partners.

The NVLink technology is currently used to facilitate high-speed data transfer between chips in Nvidia’s platforms, such as the GB200, which combines two Blackwell GPUs with a Grace CPU.

However, Nvidia is facing challenges in the Chinese market due to the restrictions on the export of Nvidia's H20 chips, imposed by the US government. As Reuters noted, Nvidia said last month it would incur $5.5 billion in charges related to these export controls.

In addition to the product announcement, Nvidia revealed plans to build a new Taiwan headquarters in the northern suburbs of Taipei.

Huang also used his keynote to reflect on the company’s evolution, highlighting Nvidia’s transition from a gaming-focused GPU manufacturer to the dominant force behind the AI boom that began with the launch of ChatGPT in 2022. He added that the company is now developing CPUs capable of running Microsoft’s Windows OS using technology from Arm Holdings.