Nvidia has introduced its next-generation Spectrum-X Ethernet networking platform, positioning it as a key building block for the next wave of “gigascale” AI factories designed to connect millions of GPUs while reducing power consumption and operational costs.
The networking platform is part of Nvidia’s broader Rubin architecture, which integrates six major components—including the Vera CPU, Rubin GPU, NVLink 6 switches, ConnectX-9 SuperNICs, BlueField-4 DPUs and the new Spectrum-6 Ethernet switches—into a tightly coupled AI infrastructure stack.
The company says this level of integration underscores the growing importance of networking in AI. In its most recent quarter, networking sales were $11 billion, up 263% year-over-year, prompting the ever-subtle CEO Jensen Huang to declare “We’re … now the largest networking company in the world” during Nvidia’s earnings call.
While GPUs have dominated headlines during the AI boom, networking has increasingly become a performance bottleneck as models grow larger and require faster communication between compute nodes. Spectrum-X is a comprehensive platform consisting of Spectrum Ethernet switches, Spectrum-X SuperNICs, ConnectX NICs, BlueField DPUs, LinkX cabling and transceivers and Spectrum-XGS for networking between multiple AI data centers.
At the heart of the platform is the Spectrum-6 switch, a 102.4-terabit-per-second Ethernet switch system delivering 2x the capacity of previous-generation systems and built as part of the Vera Rubin platform.
Spectrum-6 is designed to operate an AI factory as one end-to-end computing system. It combines new Ethernet switches, network interface cards, silicon photonics and software designed to improve bandwidth while lowering latency and power usage.
The new Spectrum-X technology intelligently balances traffic across available paths, rapidly bypasses failures and precisely recovers when data traveling across a network fails to reach its destination. Plus, support for open network operating systems and a choice of RDMA transport models gives AI builders flexibility without compromising performance.
Nvidia says its latest silicon photonics technology integrates optical communications directly into networking hardware, reducing power consumption while increasing bandwidth density compared with conventional optical networking approaches.
The announcement reflects a broader shift in AI infrastructure strategy. Early AI clusters were primarily limited by GPU availability, but hyperscale operators are increasingly finding that networking, storage and power delivery determine how efficiently massive GPU deployments perform. By integrating networking more tightly with compute, Nvidia aims to eliminate communication bottlenecks that emerge as AI systems scale beyond a single data center or even multiple campuses.
New to the platform is Nvidia’s previously announced Spectrum-XGS technology, which links geographically distributed data centers into a single AI supercomputer. Together, the technologies are designed to enable organizations to construct AI factories that span multiple facilities while operating as a unified computing environment.