CoreWeave will be among the first to adopt NVIDIA Spectrum-XGS Ethernet, a scale-across technology uniting multiple data centers into giga-scale AI factories.
NVIDIA has announced Spectrum-XGS Ethernet, a new networking breakthrough designed to combine distributed data centers into unified giga-scale AI super-factories.
As Artificial Intelligence demand surges, single data centers are hitting the limits of power and capacity within one facility. To expand, they must scale beyond a single building. Off-the-shelf Ethernet infrastructure, however, creates barriers due to high latency, jitter and unpredictable performance.
Spectrum-XGS Ethernet addresses these challenges by introducing scale-across infrastructure as part of the NVIDIA Spectrum-X platform. It becomes a third pillar of AI computing alongside scale-up and scale-out, designed to interconnect multiple distributed data centers into massive AI super-factories capable of giga-scale intelligence.
“The AI industrial revolution is here, and giant-scale AI factories are the essential infrastructure,” said Jensen Huang, Founder and CEO, NVIDIA. “With NVIDIA Spectrum-XGS Ethernet, we add scale-across to scale-up and scale-out capabilities to link data centers across cities, nations and continents into vast, giga-scale AI super-factories.”
Spectrum-XGS Ethernet is fully integrated into Spectrum-X with algorithms that dynamically adapt the network to the distance between facilities. Features include auto-adjusted distance congestion control, precision latency management and end-to-end telemetry. The result is nearly double the performance of the NVIDIA Collective Communications Library, accelerating multi-GPU and multi-node communication and enabling predictable performance across geographically distributed AI clusters.
This allows multiple data centers to function as a single AI super-factory, fully optimized for long-distance connectivity. Hyperscale pioneers such as CoreWeave are already adopting the technology.
“CoreWeave’s mission is to deliver the most powerful AI infrastructure to innovators everywhere,” said Peter Salanki, Cofounder and CTO, 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.”
The Spectrum-X Ethernet platform also delivers 1.6x greater bandwidth density than traditional Ethernet, underpinning multi-tenant hyperscale AI factories including the world’s largest AI supercomputer. It comprises NVIDIA Spectrum-X switches and NVIDIA ConnectX-8 SuperNICs, delivering seamless scalability, ultralow latency and breakthrough performance.
The launch follows a series of networking innovations from NVIDIA, including Spectrum-X and Quantum-X silicon photonics networking switches. These solutions integrate electronic circuits and optical communications at scale, enabling AI factories to connect millions of GPUs across sites while reducing power use and operational costs.
NVIDIA’s photonics switches deliver world-leading networking capabilities, combining optics innovations with 4x fewer lasers to achieve 3.5x higher power efficiency, 63x signal integrity, 10x greater resiliency at scale and 1.3x faster deployment than conventional methods.
“AI factories are a new class of data centers with extreme scale and networking infrastructure must be reinvented to keep pace,” said Huang. “By integrating silicon photonics directly into switches, NVIDIA is shattering the old limitations of hyperscale and enterprise networks and opening the gate to million-GPU AI factories.”
Spectrum-X Photonics and Quantum-X Photonics switches are available as part of NVIDIA’s Ethernet and InfiniBand platforms. These networking advances are already powering record-breaking systems. xAI’s Colossus supercomputer in Memphis, Tennessee — currently the world’s largest AI cluster — uses NVIDIA Spectrum-X Ethernet. Colossus integrates 100,000 NVIDIA Hopper GPUs to train the Grok family of large language models, offered as a feature for X Premium subscribers. The system is now being doubled in size to 200,000 GPUs.
Spectrum-X Photonics switches come in multiple configurations, including 128 ports of 800Gb/s or 512 ports of 200Gb/s, for a total of 100Tb/s bandwidth. At the high end, 512 ports of 800Gb/s or 2,048 ports of 200Gb/s deliver 400Tb/s throughput. Quantum-X Photonics switches provide 144 ports of 800Gb/s InfiniBand, based on 200Gb/s SerDes, with liquid cooling for greater efficiency.
“A new wave of AI factories requires efficiency and minimal maintenance to achieve the scale required for next-generation workloads,” said C. C. Wei, Chairman and CEO, TSMC. “TSMC’s silicon photonics solution combines our strengths in both cutting-edge chip manufacturing and TSMC-SoIC 3D chip stacking to help NVIDIA unlock an AI factory’s ability to scale to a million GPUs and beyond, pushing the boundaries of AI.”
NVIDIA has also unveiled the NVIDIA AI Data Platform, a customizable reference design for building AI-ready storage infrastructure. It enables storage providers to develop systems that accelerate AI reasoning workloads with query agents that extract insights in real time.
The platform combines NVIDIA AI Enterprise software with NIM microservices for NVIDIA Llama Nemotron models, as well as the NVIDIA AI-Q Blueprint for agentic AI systems. By integrating Blackwell GPUs, BlueField DPUs, Spectrum-X networking and the Dynamo open-source inference library, providers can deliver faster inference and more efficient storage pipelines.
“Data is the raw material powering industries in the age of AI,” said Huang. “With the world’s storage leaders, we’re building a new class of enterprise infrastructure that companies need to deploy and scale agentic AI across hybrid data centers.”
The AI Data Platform accelerates storage access by up to 48% compared with standard Ethernet, thanks to adaptive routing and congestion control. BlueField DPUs deliver 1.6x higher performance than CPU-based storage while halving power use, providing more than 3x efficiency per watt.
With these innovations across Ethernet, photonics and data platforms, NVIDIA is laying the foundations for giga- and million-scale AI factories — the core infrastructure of the AI era.
