Nvidia Introduces NVLink Fusion to Boost AI Data Center Bandwidth and Scalability

At Computex 2025, Nvidia unveiled NVLink Fusion, its latest generation of GPU interconnect technology designed to meet the growing demands of artificial intelligence (AI) data centers. Building on the company’s original NVLink architecture introduced in 2014, the new solution offers a 14-fold increase in bandwidth compared to PCIe Gen5, achieving up to 1.8 TB/s per GPU. This fifth-generation NVLink is central to Nvidia’s ‘AI factories’ strategy and is integrated into the new GB200 NVL72 and GB300 NVL72 rack models.

NVLink Fusion introduces mesh-style GPU communication and bypasses the traditional central hub approach. It also includes NVLink Switch chips capable of delivering up to 130 TB/s in a 72-GPU domain. Major tech partners such as MediaTek, Marvell, Alchip Technologies, Astera Labs, Synopsys, Cadence, Fujitsu, and Qualcomm are already collaborating with Nvidia to incorporate NVLink Fusion in their products.

The solution is designed to enable scalable AI infrastructure, with flexible integration options and a unified orchestration platform called Mission Control. This software component simplifies the complex management of large-scale AI environments. NVLink Fusion positions Nvidia at the forefront of next-generation AI data center architecture, promising greater efficiency and customization in high-performance computing environments.