NVIDIA to License NVLink Fusion to Marvell and MediaTek

NVIDIA has announced plans to offer its advanced chip interconnect technology, NVLink Fusion, to other semiconductor design companies. This move aims to accelerate the development and deployment of high-performance AI systems by enabling faster communication between chips.

The latest iteration of NVLink, known as NVLink Fusion, is engineered to link multiple chips together, facilitating the seamless exchange of massive data workloads essential for AI model training and inference. Companies like Marvell Technology and MediaTek have already confirmed plans to adopt NVLink Fusion in their custom chip initiatives, signaling broad industry interest.

NVLink has been a cornerstone of NVIDIA's high-bandwidth interconnect solutions, previously used in products such as the GB200 platform, which combines two Blackwell GPUs and one Grace CPU to create a powerful compute engine for AI applications.

The announcement was made by NVIDIA CEO Jensen Huang during his keynote at Computex 2025 in Taipei, an event held from May 20 to 23. Alongside unveiling NVLink Fusion, Huang also revealed plans to establish a new NVIDIA Taiwan headquarters in northern Taipei, underscoring the company's long-term commitment to the region.

During his keynote, Huang reflected on NVIDIA's evolution from a graphics-focused company to a dominant force in AI computing. "Historically, 90% of my talks were about GPUs for gaming," said Huang. "But today, we are shaping the future of AI with chips and systems that are transforming industries worldwide."

NVIDIA continues to innovate across its product lines. At the company's annual GPU Technology Conference (GTC) in March, Huang outlined NVIDIA's strategy to meet growing computational demands—from training large-scale AI models to deploying AI-powered applications. New offerings include the upcoming Blackwell Ultra chips, expected later this year, and the future Feynman CPU, scheduled for release in 2028, following the Rubin generation.

Additionally, NVIDIA introduced the DGX Spark, a desktop AI workstation designed for researchers. According to Huang, production has ramped up, and the system will be available for deployment within weeks.