NVIDIA has revealed that Meta and Oracle will deploy its Spectrum-X Ethernet networking switches to strengthen their AI data center infrastructures, positioning both companies at the forefront of large-scale AI computing.
The Spectrum-X Ethernet platform, designed specifically for the trillion-parameter model era, allows hyperscalers to interconnect millions of GPUs efficiently. This move reflects a broader trend among cloud and social media giants to standardize on purpose-built AI networking to accelerate training, inference, and deployment of next-generation AI systems.
For Oracle, Spectrum-X will serve as the backbone for new AI factories built on NVIDIA’s Vera Rubin architecture while Meta integrates Spectrum Ethernet switches into its Facebook Open Switching System (FBOSS), which manages and scales its data center networks. By embedding Spectrum Ethernet into its Minipack3N switch, Meta aims to expand its open networking ecosystem while achieving the predictability and scalability necessary for massive AI model training.
At the technical core, NVIDIA Spectrum-X combines advanced congestion control, adaptive routing, and AI-driven telemetry to deliver unprecedented throughput. In tests, the platform achieved 95% network efficiency, a dramatic improvement over standard Ethernet’s approximate 60% performance at scale.
The Spectrum-XGS technology further enables cross-data center connectivity, linking facilities across continents into unified AI super-factories. Together with NVIDIA’s broader hardware and software stack spanning GPUs, CPUs, and NVLink interconnects, the Spectrum-X Ethernet platform redefines how hyperscalers build and operate global AI infrastructure.