Nvidia Spectrum XGS Ethernet links connect different data centers, turning them into one large AI super factory. This setup allows separate facilities to work together as a single computing system under a common network control system. This approach removes the limits of power space and cooling that constrain single buildings and lets operators expand capacity across cities and countries. NVIDIA built the Spectrum XGS Ethernet technology to enable predictable long-distance performance for demanding AI workloads.
Scale Across
The platform adds scale across capabilities to existing scale-up and scale-out strategies. It uses distance-aware congestion control, precision latency management, and end-to-end telemetry to tune communication between sites. The result is more stable transfers and faster multi-node work that spans data centers. CoreWeave will deploy Spectrum XGS Ethernet to connect its sites and to offer unified gigascale services to customers.
Network Design
Spectrum XGS Ethernet pairs Spectrum X switches with ConnectX 8 SuperNICs and leverages silicon photonics and co-packaged optics for higher bandwidth density and lower energy use. This combination supports large GPU counts and multi-tenant deployments that require steady performance. NVIDIA integrates hardware with software libraries to speed distributed training and inference.
Market Effects
Organizations can now plan compute across multiple facilities rather than build single larger sites. This will help manage energy and space constraints and accelerate very large training and inference jobs. The platform delivers nearly doubled collective communications performance compared with typical Ethernet, making multi-site AI more routine. Early hyperscale adoption shows the approach is ready for production.
The Spectrum XGS Ethernet is available now as part of the Spectrum X platform. NVIDIA positions the offering as a core element in the future of AI infrastructure. NVIDIA Spectrum XGS Ethernet is shipping now to partners and customers, and it gives cloud providers a clear path to assemble unified AI factories across many sites.