Nvidia is thinking big again. In this case, the vision is not for a more powerful GPU to go into a more powerful rack to be deployed in a more powerful data center, but instead a vision for how to leverage the connectivity and networking technologies Nvidia has been developing to harness the computing power of multiple data centers to create “one gigantic GPU,” according to Dave Salvator, director of accelerated computing at Nvidia
Anyone attending Hot Chips 2025 in Palo Alto, California Aug. 24-26 should expect to hear a lot about this vision, which Salvator distinguished from previous notions about scaling up and scaling out by describing it as a “scale across” strategy for distributed AI computing.
Salvator, speaking during a media briefing prior to the event, added that for AI “it’s a really interesting development in terms of a way to take scale beyond data center scale… we're talking now about multi-data center scale as an as an enabler, basically, for applications to get to the scale [of computing power] that they need, whether that's for training or inference.”
At the center of Nvidia’s vision is an update to the Spectrum-X Ethernet platform the company announced two years ago. The new Spectrum-XGS Ethernet uses new algorithms that “dynamically adapt the network to the distance between data center facilities,” enabling the interconnection of multiple, distributed data centers to form “massive AI super-factories capable of giga-scale intelligence,” according to a Nvidia blog post.
With performance governed by auto-adjusted distance congestion control, precision latency management and end-to-end telemetry, Salvator claimed that Spectrum-XGS Ethernet nearly doubles the performance of the NVIDIA Collective Communications Library, accelerating multi-GPU and multi-node communication to deliver predictable performance across geographically distributed AI clusters.
The result is a networking platform composed of Nvidia’s Spectrum-X switches and ConnectX-8 SuperNICs that provides 1.6x greater bandwidth density than off-the-shelf Ethernet for multi-tenant, hyperscale AI factories, the company added.
Among the first companies deploying Spectrum-XGS Ethernet is the AI specialist hyperscaler CoreWeave. “CoreWeave’s mission is to deliver the most powerful AI infrastructure to innovators everywhere,” said Peter Salanki, co-founder and chief technology officer 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.”