Felix Pinkston
Oct 13, 2025 14:36
Meta and Oracle are enhancing their AI data centers with NVIDIA Spectrum-X Ethernet switches, aiming to improve AI training efficiency and network performance.
Meta and Oracle are set to revolutionize their AI data centers by integrating NVIDIA Spectrum-X™ Ethernet networking switches, as announced by NVIDIA. This strategic move aims to enhance AI training efficiency and network performance, positioning both companies at the forefront of AI innovation.
Meta’s Integration with Spectrum-X
Meta plans to incorporate the Spectrum-X Ethernet switches into its Facebook Open Switching System (FBOSS) platform. This integration is expected to streamline large-scale deployments and significantly improve AI training efficiency. According to Gaya Nagarajan, Vice President of Networking Engineering at Meta, the integration is crucial for managing the next-generation AI infrastructure, which demands open and efficient networking at unprecedented scales.
Oracle’s Giga-Scale AI Supercomputers
Oracle is set to construct giga-scale AI supercomputers powered by the Spectrum-X Ethernet switches. The company aims to leverage the NVIDIA Vera Rubin architecture to accelerate AI factories, enhancing interconnectivity and efficiency. Mahesh Thiagarajan, Executive Vice President of Oracle Cloud Infrastructure, emphasized the partnership’s potential to expedite AI training and deployment, benefiting customers with advanced AI capabilities.
NVIDIA Spectrum-X: The AI Networking Backbone
The Spectrum-X Ethernet platform is engineered for the era of trillion-parameter models, providing hyperscalers the ability to interconnect millions of GPUs with unparalleled efficiency. It has already set records in efficiency, achieving 95% data throughput with its congestion-control technology, compared to the 60% throughput of conventional Ethernet at scale. This performance leap is pivotal for developing the world’s most advanced AI infrastructures.
Industry Impact and Future Prospects
NVIDIA’s Spectrum-X Ethernet switches are not just faster Ethernet; they are the nervous system of AI factories, allowing hyperscalers like Meta and Oracle to connect vast numbers of GPUs into a unified computing powerhouse. This technological advancement is expected to redefine data center economics and performance, facilitating the construction of giga-scale AI super-factories that span cities, nations, and continents.
For more information, visit the NVIDIA Newsroom.
Image source: Shutterstock
Credit: Source link
