NVIDIA has announced the NVIDIA Spectrum-XGS Ethernet, a new technology designed to connect multiple distributed data centers into unified, large-scale AI clusters. This offering expands the capabilities of the Spectrum-X Ethernet platform, introducing a “scale-across” networking approach that links geographically separated facilities to function as a single, high-performance AI super-factory. According to NVIDIA, this addresses the industry challenge of increased power and capacity limitations within individual data center buildings—a critical consideration as AI workloads accelerate.
Spectrum-XGS Ethernet builds on the existing Spectrum-X platform by adding infrastructure to interconnect AI clusters across cities, nations, and continents. The system incorporates algorithms that automatically adapt network behavior based on inter-site distances. Key technical features include advanced, dynamic congestion control tailored for different distances, precision latency management, and end-to-end telemetry. NVIDIA claims that, with this design, Spectrum-XGS Ethernet nearly doubles the performance of the NVIDIA Collective Communications Library for multi-GPU and multi-node communication across distributed clusters.
The platform integrates NVIDIA Spectrum-X switches and NVIDIA ConnectX-8 SuperNICs, which together offer what NVIDIA reports is 1.6 times greater bandwidth density than conventional Ethernet solutions. This is targeted directly at multi-tenant, hyperscale data centers aiming to interconnect at giga-scale for AI workloads. NVIDIA states that this results in predictable performance and ultralow latency across distributed sites, allowing multiple facilities to act as a single AI resource.
NVIDIA indicates that CoreWeave, a hyperscale cloud provider, will be among the first to implement Spectrum-XGS Ethernet for its own distributed data centers. “With NVIDIA Spectrum-XGS, we can connect our data centers into a single, unified supercomputer, giving our customers access to giga-scale AI that will accelerate breakthroughs across every industry,” said Peter Salanki, cofounder and chief technology officer of CoreWeave.
The Spectrum-XGS Ethernet solution is available as of August 2025. This launch follows previous networking announcements from NVIDIA, such as its Spectrum-X and Quantum-X silicon photonics networking switches, intended to support massive GPU connectivity across sites while lowering operational costs and energy usage.
Source: NVIDIA
