Overview

• Nvidia's NVLink Fusion program allows non-Nvidia CPUs and custom AI accelerators to integrate with its high-speed NVLink interconnect for rack-scale AI infrastructure.
• Early NVLink Fusion partners include Qualcomm and Fujitsu for CPUs, and MediaTek, Marvell, Alchip, Synopsys, Cadence, and Astera Labs for accelerators and design tools.
• NVLink delivers up to 14× the bandwidth of PCIe 5.0, enabling faster communication between GPUs and CPUs in AI data centers.
• Rivals AMD, Intel, and Broadcom remain absent from the NVLink Fusion ecosystem, backing the open Ultra Accelerator Link standard instead.
• Nvidia also launched DGX Cloud Lepton, a GPU compute rental platform, now in early access with support from multiple cloud providers offering tens of thousands of GPUs.