Overview

• NVLink Fusion, Nvidia's fifth-generation interconnect technology, provides 1.8 TB/s per GPU, a 14x bandwidth improvement over PCIe Gen5.
• The technology is integrated into Nvidia's GB200 and GB300 NVL72 racks, supporting 130 TB/s across 72 accelerators for AI data centers.
• For the first time, Nvidia is offering NVLink as an open IP standard, enabling third-party CPUs and ASICs to integrate directly into its ecosystem.
• Qualcomm and Fujitsu are the inaugural adopters, with Qualcomm planning to incorporate NVLink Fusion into its Oryon CPUs for AI inference workloads.
• The move challenges the Ultra Accelerator Link consortium's competing open standard, positioning Nvidia to maintain dominance in modular AI infrastructure.