Nvidia CEO Jensen Huang on Monday introduced a new initiative aimed at expanding the company’s role in artificial intelligence infrastructure, announcing “NVLink Fusion” at Computex 2025. The move marks a significant shift for the chipmaker, allowing its interconnect technology to work with processors beyond its own.
Previously limited to Nvidia’s proprietary hardware, NVLink will now support third-party CPUs and application-specific integrated circuits (ASICs), enabling more flexible and semi-custom AI systems. “NVLink Fusion is so that you can build semi-custom AI infrastructure, not just semi-custom chips,” Huang said during his keynote at Asia’s largest electronics event.
The program opens the door for Nvidia’s customers and partners — including Fujitsu and Qualcomm Technologies — to integrate their own processors alongside Nvidia’s GPUs in AI data centers. Chipmakers MediaTek, Marvell, Alchip, Astera Labs, Synopsys, and Cadence have already joined as early collaborators.
Industry experts say the change positions Nvidia to stay competitive as more companies design custom chips tailored for specific workloads. “NVLink Fusion consolidates Nvidia as the center of next-generation AI factories — even when those systems aren’t built entirely with Nvidia chips,” said Ray Wang, a Washington-based technology analyst.
While Nvidia leads in general-purpose AI training hardware, cloud giants like Google, Amazon, and Microsoft — all of which use Nvidia products — are developing in-house processors that could pose future challenges. Wang believes NVLink Fusion may ease that tension by fostering cooperation instead of competition.
Still, the strategy carries some risks. By enabling customers to use alternative CPUs, Nvidia may see reduced demand for its own processor offerings. But analysts say the tradeoff could be worthwhile. “At the system level, the added flexibility improves the competitiveness of Nvidia’s GPU-based solutions versus emerging architectures,” said Rolf Bulk of New Street Research.
Notably, rivals Broadcom, AMD, and Intel have not yet joined the NVLink Fusion ecosystem.
Beyond NVLink, Huang also provided an update on Nvidia’s Grace Blackwell platform. The upcoming “GB300” system, expected in the third quarter, will offer improved performance for demanding AI tasks.
The company also launched DGX Cloud Lepton, a new AI platform designed to connect developers with global GPU resources through a unified compute marketplace. Nvidia said the service will help users access high-performance infrastructure through a network of cloud providers.
Capping off the keynote, Huang announced plans to open a new Nvidia office in Taiwan and revealed a major collaboration with Foxconn to build an AI supercomputer on the island.
“We are delighted to partner with Foxconn and Taiwan to help build Taiwan’s AI infrastructure,” Huang said, emphasizing the country’s critical role in global chip production and innovation.
