#nvlinkfusion

Nvidia continues to solidify its position as a global AI powerhouse with the launch of NVLink Fusion, a groundbreaking technology designed to accelerate chip-to-chip communication for advanced AI systems. Announced ahead of the highly anticipated Computex 2025 trade show, this move signals Nvidia’s strategic shift from a GPU-centric past to a future driven by AI innovation.

What is NVLink Fusion?

NVLink Fusion is the latest evolution of Nvidia’s NVLink technology, originally developed to enable high-speed data transfer between multiple GPUs. The new iteration goes a step further, allowing third-party chip designers to harness Nvidia’s interconnect technology to build custom AI systems at scale. This will be crucial for companies seeking to deploy AI tools that require seamless data sharing across a network of processors.

Currently, NVLink plays a critical role in Nvidia’s flagship GB200 chip platform, where it connects two Blackwell GPUs with a Grace CPU, optimizing data throughput for intensive AI workloads.

Major Announcements at Computex 2025

The announcement came during Nvidia CEO Jensen Huang’s keynote at the Taipei Music Center, just a day before Computex officially opens its doors from May 20–23. Huang’s speech was packed with major developments, including:

A New Taiwan Headquarters: Nvidia plans to establish a regional HQ in northern Taipei, a move that reinforces its long-term commitment to Taiwan and its pivotal role in global semiconductor manufacturing.

Shift in Focus: Huang reflected on Nvidia’s transformation, noting that while his past presentations focused heavily on gaming GPUs, today’s spotlight is firmly on AI and data-centric computing.

Next-Gen Chips Roadmap: Huang also detailed Nvidia’s ambitious roadmap:

Blackwell Ultra chips arriving later this year.

Rubin chips to follow.

Feynman processors set for release in 2028.

DGX Spark Desktop AI Chip in Full Production

In a nod to AI researchers and developers, Nvidia also confirmed that its DGX Spark, a desktop-class AI computing system, has entered full-scale production. It will be available in the coming weeks and is expected to be a game-changer for AI innovation at the desktop level.

Nvidia’s Expanding AI Ecosystem

Alongside hardware, Nvidia is deeply invested in the software and platforms required to train and deploy generative AI. At the company’s GTC developer conference in March, Huang emphasized a strategic pivot—from building massive AI models to enabling practical, scalable AI applications.

Nvidia is also expanding into CPU development, reportedly working on chips to run Microsoft Windows, based on Arm Holdings’ architecture, marking another step in its effort to dominate every level of the AI stack.

The Return of "Jensanity"

If last year’s Computex was any indication—where Huang was mobbed by fans and media alike—the Nvidia CEO is expected to draw massive crowds again this year, highlighting the cultural impact he has had on Taiwan’s tech landscape.

Final Thoughts