Nvidia introduces NVLink Fusion, a breakthrough chip-to-chip communication tech, enabling faster AI data processing and collaboration with Marvell and MediaTek.
 |
| From Computex 2025 in Taipei, Nvidia’s Jensen Huang reveals NVLink Fusion and plans for a new Taiwan HQ, strengthening the company’s AI leadership. Image: Collected |
TAIPEI, Taiwan — May 19, 2025:
Nvidia on Monday unveiled NVLink Fusion, its latest chip-to-chip communication technology, marking a major step toward enabling faster, more scalable artificial intelligence (AI) systems. Announced during CEO Jensen Huang’s keynote at Computex 2025 in Taipei, the new interconnect platform is designed to be shared with other chip-makers — including Marvell Technology and MediaTek — to build powerful custom AI infrastructures.
The NVLink Fusion system builds on Nvidia’s years of interconnect innovation, allowing multiple processors to share data at high speed and low latency. It’s already at work inside Nvidia’s advanced GB200 platform, which combines two Blackwell GPUs with a Grace CPU.
“Fusion is the fabric that will allow everyone to build AI factories,” said Huang at the Taipei Music Center, as Computex opened its doors to over 1,400 exhibitors from across the globe.
Huang also announced plans to build a new Nvidia headquarters in the northern suburbs of Taipei, reinforcing Taiwan’s importance in the global semiconductor landscape. The facility will serve as a strategic hub for Nvidia’s Asia-Pacific operations and research.
In a wide-ranging keynote, Huang recounted Nvidia’s transformation from a graphics card manufacturer to a global leader in AI computing. “What used to be 90% about graphics is now 100% about AI,” he said, drawing cheers from the tech-savvy crowd.
The launch of NVLink Fusion is part of Nvidia’s larger strategy to support the next wave of AI computing — not just training large models, but deploying them efficiently in real-world applications. Earlier this year, Huang unveiled Nvidia’s upcoming chip lineup, including the Blackwell Ultra (set to launch later in 2025), followed by the Rubin and Feynman processors by 2028.
Additionally, Huang confirmed that the DGX Spark — Nvidia’s new AI desktop platform for researchers — is now in full production and will ship within weeks.
This year’s Computex marks the first major chip and computing conference in Asia since major geopolitical shifts prompted calls for tech manufacturing repatriation. Against that backdrop, Nvidia’s announcements underscore its commitment to openness, global partnerships, and technological leadership.
As the world races to power the AI-driven future, Nvidia’s NVLink Fusion could be the connective tissue that binds together the next generation of computing.