NVIDIA's AI Ethernet Platform Secures Major Customer! Supporting Meta and Oracle's Large-Scale AI Infrastructure

According to reports on October 13th, at the OCP Global Summit in San Jose, NVIDIA showcased the future of gigawatt AI factories and announced that Meta and Oracle will use NVIDIA Spectrum-X Ethernet network switches to enhance their AI data center networks, accelerate large-scale deployments, and achieve exponential growth in AI training efficiency.

Nvidia founder and CEO Jensen Huang said Spectrum-X is more than just faster Ethernet; it's the "nervous system of the AI ​​factory," enabling hyperscale computing companies to connect millions of GPUs into a single giant computer to train the largest models ever.

Designed for the era of trillion-parameter models, the NVIDIA Spectrum-X Ethernet platform, comprised of Spectrum-X Ethernet switches and Spectrum-X Ethernet SuperNICs, is the first Ethernet platform purpose-built for AI.

Spectrum-X Ethernet has demonstrated record efficiency, with its congestion control technology enabling the world's largest AI supercomputers to achieve 95% data throughput. In comparison, existing large-scale Ethernet networks experience thousands of traffic collisions, limiting throughput to around 60%.

This leap in efficiency marks a breakthrough in the economics and performance of scalable AI networks. NVIDIA Spectrum-XGS Ethernet technology, part of the Spectrum-X Ethernet networking platform, enables cross-scale scalability, connecting data centers across cities, countries, and continents into massive, billion-scale AI superfactories.

Based on NVIDIA's full-stack platform (including GPUs, CPUs, NVLink, and software), Spectrum-X delivers seamless performance from compute to networking. Its advanced congestion control, adaptive routing, and AI-driven telemetry ensure efficiency and predictability for large-scale AI training and inference clusters.

Oracle will build a billion-scale AI factory interconnected by Spectrum-X Ethernet.

"By adopting Spectrum-X Ethernet, we can interconnect millions of GPUs with breakthrough efficiency, enabling our customers to more quickly train, deploy, and benefit from the next wave of generative and inference AI," said Mahesh Thiagarajan, Executive Vice President of Oracle Cloud Infrastructure.

Meta will integrate Spectrum Ethernet switches into its Facebook Open Switching System (FBOSS) network infrastructure. FBOSS is a software platform designed for managing and controlling network switches at scale. This integration will accelerate large-scale deployments, improving AI training efficiency and reducing time to insight.

Gaya Nagarajan, Vice President of Engineering at Meta Networks, shared, "Meta's next-generation AI infrastructure requires open, efficient networking at a scale never before seen in the industry. By integrating NVIDIA Spectrum Ethernet into Minipack3N switches and FBOSS, we can extend our open networking approach while unlocking the efficiency and predictability needed to train larger models and bring generative AI applications to billions of people."

NVIDIA NVLink Fusion is also rapidly gaining momentum, enabling enterprises to seamlessly integrate their semi-custom chips into highly optimized and widely deployed data center architectures, reducing complexity and accelerating time to market.

Intel and Samsung foundry companies are joining the NVLink Fusion ecosystem. This ecosystem includes custom chip designers, CPU, and IP partners, enabling AI factories to rapidly scale to handle the demanding workloads of model training and Agentic AI inference.

New NVLink Fusion ecosystem partners include:

• CPU partners: Fujitsu, Intel, Qualcomm.
• Custom chip partners: AIchip, Astera Labs, GUC, Marvell, MediaTek, Samsung.
• Technology Partners: Cadence, Synopsys.

As part of the recently announced collaboration between NVIDIA and Intel, Intel will build x86 CPUs that integrate into NVIDIA infrastructure platforms using NVLink Fusion.

Samsung Foundry has partnered with NVIDIA to meet the growing demand for custom CPUs and custom XPUs, providing a comprehensive design-to-manufacturing experience for custom chips.

Gigawatt AI clusters are now a common goal for leading AI chip companies. NVIDIA's Spectrum-X Ethernet and Spectrum-XGS, combining the openness of the Open Compute Project (OCP) with leading performance, have been selected for deployment by two tech giants, Meta and Oracle.

These technologies, designed to achieve cross-scalability, will deliver significant efficiency improvements for large-scale AI training and inference, accelerating the development and adoption of generative AI applications.