Nvidia announces new chip-linking tech to support custom AI

Nvidia has just announced a new technology aimed at improving how chips communicate with each other in large-scale artificial intelligence (AI) systems.

At the Computex tech trade show in Taiwan on Monday, the company introduced NVLink Fusion, an updated version of its existing NVLink technology. This system enables faster chip-to-chip communication, a crucial part of building powerful AI tools that require multiple chips working together. Nvidia said it plans to make the technology available to other chipmakers, helping them design custom AI hardware more efficiently.

The announcement reflects Nvidia's shift from its original focus on graphics chips for video games to a broader role in the AI industry, states a recent report by Reuters on Nvidia's latest announcements. Since the launch of ChatGPT in 2022, demand for high-performance AI chips has surged, placing Nvidia at the centre of the trend.

At the event, CEO Jensen Huang also highlighted Nvidia's wider roadmap, adds the report. The company is working on new generations of AI processors, including the Blackwell Ultra, due later this year, and the Rubin series, which will be followed by Feynman chips in 2028. Nvidia also revealed a desktop AI system for researchers, the DGX Spark, which is now in full production and expected to ship in the coming weeks.

In addition, Nvidia is developing CPUs that run Microsoft's Windows operating system using technology from Arm Holdings, according to previous reports by Reuters.

Computex 2025 marks a major gathering for the tech and chipmaking industry in Asia, with more than 1,400 exhibitors participating, as per the Reuters report.