NVIDIA Spectrum-6 SPX
Type: Platform Tags: NVIDIA, Spectrum-6 SPX, SPX, networking rack, Spectrum-X, Quantum-X800, silicon photonics, co-packaged optics, Vera Rubin, AI factory Related: NVIDIA-Vera-Rubin-POD, NVIDIA-Vera-Rubin, NVIDIA-Spectrum-X, NVIDIA-Quantum-X800-InfiniBand, NVIDIA-Silicon-Photonics, NVIDIA-ConnectX-9, NVIDIA-BlueField-4, NVIDIA-STX, NVIDIA-CMX, NVIDIA-Enterprise-AI-Factory, NVIDIA-DGX-SuperPOD Sources: https://developer.nvidia.com/blog/nvidia-vera-rubin-pod-seven-chips-five-rack-scale-systems-one-ai-supercomputer/, https://nvidianews.nvidia.com/news/nvidia-vera-rubin-platform Last Updated: 2026-05-09
Summary
NVIDIA Spectrum-6 SPX is the networking rack in the NVIDIA-Vera-Rubin-POD architecture. NVIDIA positions SPX as the rack-to-rack networking layer that connects the Vera Rubin POD into a single AI supercomputer, accelerating east-west and north-south traffic across AI factories with either NVIDIA-Spectrum-X Ethernet or NVIDIA-Quantum-X800-InfiniBand switching.
Detail
Purpose
SPX is the POD-scale networking building block for next-generation NVIDIA AI factories. It exists because agentic AI workloads need low-latency, high-throughput, resilient connectivity among compute, storage, inference accelerator, and CPU racks.
Architecture notes
- Spectrum-6 SPX networking racks connect the entire Vera Rubin POD.
- The SPX rack can be configured with Spectrum-X Ethernet or Quantum-X800 InfiniBand switches.
- The rack includes the 102.4 Tb/s Spectrum-6 switch.
- The Spectrum-6 switch has 512 lanes and 200 Gb/s co-packaged optics in single-chip and multi-chip switch offerings.
- NVIDIA positions the silicon photonics/CPO integration as replacing pluggable transceivers to improve power efficiency, resiliency, latency, jitter, and effective bandwidth.
- SPX keeps AI workloads synchronized across compute and storage environments, including traffic patterns created by training, inference, KV-cache movement, and data services.
NVIDIA context
Use this page for “SPX” search and for questions about Vera Rubin-era AI factory networking racks. Use NVIDIA-Spectrum-X for current Ethernet AI networking and NVIDIA-Quantum-X800-InfiniBand for the InfiniBand option.
Connections
- NVIDIA-Vera-Rubin-POD - SPX is one of the five rack-scale systems in the POD.
- NVIDIA-Vera-Rubin - next-generation platform context.
- NVIDIA-Spectrum-X - Ethernet option for SPX configurations.
- NVIDIA-Quantum-X800-InfiniBand - InfiniBand option for SPX configurations.
- NVIDIA-Silicon-Photonics - co-packaged optics and optical networking direction.
- NVIDIA-ConnectX-9 - SuperNIC direction adjacent to Spectrum-6 SPX fabrics.
- NVIDIA-BlueField-4 - DPU used by adjacent Vera Rubin POD storage/data racks.
- NVIDIA-STX and NVIDIA-CMX - storage and context memory traffic that SPX networking must carry.
- NVIDIA-Enterprise-AI-Factory - SPX is a future AI factory networking building block.
- NVIDIA-DGX-SuperPOD - current SuperPOD patterns lead into Vera Rubin POD-scale networking.
Source Excerpts
- NVIDIA says Spectrum-6 SPX networking racks connect the entire Vera Rubin POD into a single supercomputer.
- NVIDIA describes the rack as configurable with Spectrum-X Ethernet or Quantum-X800 InfiniBand switching.
- NVIDIA says the Spectrum-6 switch includes 102.4 Tb/s bandwidth, 512 lanes, and 200 Gb/s co-packaged optics.