NVIDIA's 102.4-terabit-per-second Spectrum-6 Ethernet switch system has arrived at the world's largest AI factories, marking a critical infrastructure milestone as artificial intelligence enters the gigascale era.

The switch system delivers twice the capacity of previous-generation systems and forms part of NVIDIA's Vera Rubin platform. It's designed to handle the networking demands of AI factories running hundreds of thousands of GPUs and CPUs simultaneously.

Leading Cloud Providers Deploy First

CoreWeave, Microsoft, Nebius, Oracle Cloud Infrastructure, SpaceXAI and Tesla are among the first to deploy Spectrum-6 in their AI infrastructure. The companies are integrating the switches into their data centers to accelerate model training and inference workloads.

"CoreWeave is built for the most demanding AI workloads, and networking is central to delivering that performance at scale," said Min Jun, director of product for networking at CoreWeave. The company plans to use liquid-cooled Spectrum-X Ethernet infrastructure to help customers train frontier models faster.

Nebius emphasized the coordination challenge at gigascale. "At gigascale, performance comes down to coordination: keeping every GPU in lockstep so one slow link doesn't stall an entire job," said Laurelle Roseman, vice president of global partnerships at the company.

The switch addresses a fundamental bottleneck in large-scale AI training. Peak GPU performance alone no longer determines AI factory performance when thousands of accelerators must exchange data continuously during collective communications operations.

Purpose-Built for AI Workloads

Spectrum-6 combines with NVIDIA's ConnectX-9 SuperNIC to form the next generation of Spectrum-X Ethernet. Unlike standard Ethernet designed for enterprise north-south traffic, the platform is engineered specifically for AI's synchronized, collective-heavy communication patterns.

The system supports both pluggable and co-packaged optics, plus liquid cooling for comprehensive thermal management. NVIDIA claims Spectrum-X Ethernet delivers up to 1.6x higher AI networking performance than off-the-shelf alternatives while sustaining 95% network efficiency across deployments exceeding 100,000 GPUs.

Hardware-accelerated multiplane topologies reduce required switches by 1.7x compared to traditional architectures. The platform maintains NVIDIA's approach of vertical integration with horizontal openness, supporting standard Ethernet protocols and open network operating systems.