Nvidia Unveils SpectrumXGS to Link Data Centers for AI

Nvidia rolled out a comprehensive upgrade to its AI infrastructure strategy, announcing SpectrumXGS, a networking technology designed to knit together multiple data centers into a single, scalable AI environment. The company said the new capabilities enable “scale across” multisite AI workloads, improving inter-center communication, reducing jitter and latency, and pushing AI systems toward what Nvidia calls “AI superfactories” that span cities or even continents. The announcements arrive ahead of a major industry conference and come alongside advances in the company’s Dynamo inference framework and a broader push into photonics-based interconnects planned for 2026.

SpectrumXGS sits atop Nvidia’s SpectrumX Ethernet platform and is designed to orchestrate and interlink data centers so that clusters can operate as a unified GPU fabric. Nvidia executives described the technology as addressing a core bottleneck in current AI deployments: the limits imposed by power budgets and thermal constraints within individual data centers. By coordinating cross-site traffic with advanced network optimization, SpectrumXGS aims to deliver predictable performance for models running across...