NVIDIA Debuts NVLink Fusion to Power Custom AI Supercomputers

Rewiring AI Infrastructure

NVIDIA has launched NVLink Fusion, a high-speed, scalable chip-to-chip and system-to-system interconnect designed to help enterprises and original equipment manufacturers (OEMs) build semi-custom AI infrastructure. Announced at COMPUTEX 2024, NVLink Fusion allows organizations to leverage NVIDIA’s powerful GPUs, networking, and software stacks while tailoring their infrastructure to meet domain-specific performance needs. This initiative is part of NVIDIA’s effort to democratize supercomputing by enabling a broader range of industries—from life sciences to automotive—to create purpose-built systems optimized for their unique AI workloads.

Partnering for Scale and Flexibility

NVLink Fusion is at the core of the NVIDIA Partner Network’s new initiative—the NVLink Fusion initiative, which brings together system builders, OEMs, ODMs, and hyperscalers. Early partners include Dell Technologies, Hewlett Packard Enterprise, Lenovo, Supermicro, and cloud providers such as CoreWeave. These companies can now customize system configurations using NVIDIA’s data center GPU architectures, like Blackwell, while integrating proprietary technologies or accommodating specific customer demands. This emerging model of collaborative design and delivery demonstrates how NVIDIA is reshaping the AI infrastructure landscape by combining large-scale innovation with modular adaptability.