Nvidia is looking beyond hardware to solve one of AI’s biggest bottlenecks: moving massive amounts of data quickly and reliably across long distances. The company has introduced Spectrum-XGS Ethernet algorithms, designed to accelerate GPU-to-GPU communication when spread across multiple servers and even between distant data centers.
These algorithms are essentially making Nvidia’s existing Spectrum-X infrastructure smarter.
“It’s not a new hardware element, but it’s leveraging the Spectrum-X infrastructure, and the new algorithms effectively move more data across longer distances between sites,” said Gilad Shainer, senior vice president of networking at Nvidia.
Making distributed GPUs act like one machine
So why does this matter? AI workloads are very quickly outgrowing what a single data center can handle. With limits on power and space, companies are spreading GPUs across multiple facilities, sometimes hundreds of miles apart. The real challenge is ensuring that all those GPUs still perform as if they were in the same room.
Normally, Ethernet doesn’t care whether the connection is across a rack or across the country; it treats them the same. Spectrum-XGS changes that by adjusting performance on the fly. It analyzes real-time data, such as traffic patterns, congestion, and the distance between sites, and then adjusts parameters like congestion control, routing, and load balancing.
The result is that data centers spread out over long distances can still run like one giant AI supercomputer.
A new pillar for AI factories
Nvidia is framing Spectrum-XGS Ethernet as the “third pillar” of AI computing, adding scale-across to the more familiar concepts of scale-up and scale-out.
“The AI industrial revolution is here, and giant-scale AI factories are the essential infrastructure,” said Jensen Huang, founder and CEO of Nvidia. “With NVIDIA Spectrum-XGS Ethernet, we add scale-across to scale-up and scale-out capabilities to link data centers across cities, nations and continents into vast, giga-scale AI super-factories.”
The technology pretty much doubles the performance of Nvidia’s Collective Communications Library, which handles GPU-to-GPU and multi-node communication. That means faster training times, more predictable performance, and the ability to run super complex AI models across distributed clusters.
Who’s first in line
Early adopters include CoreWeave, a hyperscale infrastructure provider.
“CoreWeave’s mission is to deliver the most powerful AI infrastructure to innovators everywhere,” said Peter Salanki, cofounder and CTO of CoreWeave. “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.”
Spectrum-XGS is already integrated into Spectrum-X switches, ConnectX-8 SuperNICs, and systems running Blackwell GPUs. For companies wanting to scale up AI factories, the idea is, simply, that smarter Ethernet keeps everything connected and humming.
NVIDIA, Dell, and Elastic recently announced new updates to the Dell AI Data Platform, designed to help customers support the full lifecycle of AI workloads, from ingestion and transformation to agentic inference and AI-powered knowledge retrieval. Read more here.