NVIDIA Launches NVLink Fusion for Custom AI Silicon Collaboration

NVIDIA has significantly advanced AI computing through its collaboration with TSMC and innovations like NVLink, a high-speed interconnect technology that enables rapid GPU-to-GPU and CPU-to-GPU communication with bandwidths up to 130TB per second. The company recently launched NVLink Fusion, a program that allows partners and customers to develop semi-custom AI infrastructure by integrating non-NVIDIA CPUs and AI accelerators with NVIDIA's GPUs, facilitating scalable AI environments tailored to enterprise needs. This initiative has attracted major industry players such as Qualcomm, Fujitsu, MediaTek, and Marvell, who plan to incorporate NVLink Fusion to build powerful AI data centers. NVIDIA's proprietary NVLink technology, which significantly outperforms standard PCIe interfaces, serves as a critical component in scaling AI workloads across rack-scale architectures, enabling massive GPU clusters for complex AI model training and inference. Additionally, NVIDIA offers two main GPU architectures for AI servers—NVLink (SXM Edition) and PCIe Edition—where the SXM design, integrated with NVLink switches, provides ultra-high bandwidth and efficient GPU interconnectivity within systems like DGX and HGX. These innovations underpin NVIDIA's dominant position in AI computing infrastructure, supporting the growing demands of data-intensive AI applications.