Nvidia Launches NVLink Fusion To Accelerate Communication Between AI Chips, MediaTek And Marvell Ready For Adoption

NVLink Fusion is the latest version of NVLink technology (photo: x @_JoseNajarro)

JAKARTA The United States chip giant, Nvidia, announced the launch of their latest technology, NVLink Fusion, which is designed to accelerate communication between chips in an artificial intelligence (AI) system. This announcement was made directly by Nvidia CEO Jensen Huang in a keynote speech at Computex 2025 which was held at Taipei Music Center, Taiwan.

NVLink Fusion is the latest version of NVLink technology that has previously been used to transfer large amounts of data between Nvidia chips, such as the GB200 unit which combines two Blackwell GPUs with Grace processors.

With NVLink Fusion, Nvidia will not only use it for their own products, but will also start selling this technology to other companies. Two major companies, Marvell Technology and MediaTek, have stated their plans to adopt NVLink Fusion in the development of custom chips for their AI systems.

"This technology allows the design of a stronger and more efficient AI system by connecting multiple chips more quickly and optimally," Huang said.

In addition to the launch of the new technology, Nvidia also announced that it would set up a headquarters in Taiwan, precisely in the northern suburbs of Taipei. This move demonstrates Nvidia's commitment to strengthening its presence in Asia.

Huang also highlighted how Nvidia's role has evolved from graphic chip manufacturers to games to become a global leader in AI chips since the boom in ChatGPT's popularity in 2022. He revealed that now most of its presentations are focused on AI systems and software, no longer on graphics chips.

On that occasion, Nvidia also launched an AI desktop computer called DGX Spark aimed at AI researchers. Huang ensures that the device is in full production phase and will be available in the next few weeks.

Computex 2025 itself is expected to be followed by around 1,400 exhibitors, and will be the first major meeting of computer and chip industry executives in Asia after trade tensions between the United States and Asia escalated following US President Donald Trump's statement regarding the threat of new tariffs.

Previously, at Nvidia's annual developer conference in March 2024, Huang had revealed next-generation roadmaps of chips such as Blackwell Ultra which will be present later this year, as well as Rubin and Feynman which are planned to launch until 2028.

With the launch of NVLink Fusion and various other innovative products, Nvidia has further emphasized its position as a major player in the AI-based computing era.