At Computex 2025, Nvidia announced a major expansion of its AI chip ecosystem with the launch of NVLink Fusion, a cutting-edge chip-to-chip communication technology designed to accelerate artificial intelligence systems.
During a keynote speech at the Taipei Music Centre, Nvidia CEO Jensen Huang revealed that the company will now license NVLink Fusion to external chipmakers, enabling them to develop powerful, multi-chip AI solutions. Major players like Marvell Technology and MediaTek are already planning to integrate the technology into their custom chip designs.
What Is NVLink Fusion?
Originally introduced years ago, Nvidia’s NVLink is a high-bandwidth interconnect technology used to transfer massive volumes of data between chips. The latest version, NVLink Fusion, builds on this legacy, enhancing communication speeds for complex AI workloads.
It is already employed in Nvidia’s GB200 platform, which merges two Blackwell GPUs with the Grace CPU. By licensing this tech, Nvidia is offering others the ability to develop high-performance, AI-optimised systems.
Nvidia’s Expanded Role in AI
Jensen Huang also announced plans to build a new Nvidia Taiwan headquarters in northern Taipei, reinforcing the company’s growing commitment to Asia. He highlighted Nvidia’s evolution from a video game GPU manufacturer to a global leader in AI computing.
“In the past, my presentations focused 90% on graphics chips,” Huang noted. “Today, Nvidia is at the centre of the AI revolution.”
Nvidia’s hardware now powers much of the AI boom sparked by tools like ChatGPT, and the company continues to broaden its reach, developing Arm-based CPUs for Microsoft’s Windows OS, according to earlier Reuters reports.
Future AI Chip Roadmap
At Computex and in previous development events, Huang detailed Nvidia’s future AI chip lineup, which includes:
- Blackwell Ultra—Set for release later in 2025
- Rubin processors—to follow Blackwell
- Feynman chips—expected by 2028
Additionally, Nvidia introduced DGX Spark, a desktop AI system designed for researchers, now in full production and shipping within weeks.
Computex 2025: A Strategic Platform
With over 1,400 exhibitors, Computex 2025 marks the first major gathering of tech executives in Asia since geopolitical tensions began influencing global chip supply chains. Nvidia’s presence at the event and Huang’s enthusiastic following, known as “Jensanity,” highlight the company’s outsized role in shaping the AI future.
