NVIDIA Introduces NVLink Fusion for Custom AI Infrastructure

NVIDIA has introduced NVLink Fusion, a new silicon solution designed to enable industries to create semi-custom AI infrastructure. This announcement was made at Computex 2025 in Taipei, Taiwan. NVLink Fusion allows system architects to integrate non-NVIDIA CPUs or accelerators with NVIDIA's products in rack-scale architectures, opening up new possibilities for custom AI solutions.

The NVLink Fusion program has attracted several partners, including Qualcomm and Fujitsu, who plan to integrate the technology into their CPUs. Additionally, custom AI accelerators from companies like MediaTek, Marvell, and Alchip will also be part of the NVLink Fusion ecosystem. This initiative aims to enhance the scalability and performance of AI workloads by leveraging NVIDIA's proprietary interconnect technology, which offers superior bandwidth and latency compared to standard interfaces.

NVIDIA's NVLink Fusion is expected to provide cloud providers with a straightforward path to scale AI factories to millions of GPUs, utilizing NVIDIA's rack-scale systems and end-to-end networking platform. This development marks a significant step in NVIDIA's efforts to expand its AI ecosystem and facilitate broader AI adoption across various industries.