NVIDIA announced that Meta and Oracle have adopted its Spectrum-X Ethernet networking platform to enhance AI data center performance, scalability, and efficiency.
The collaboration advances global efforts to build giga-scale AI factories capable of supporting trillion-parameter models and next-generation generative AI development.
Hyperscalers Embrace Spectrum-X Ethernet
NVIDIA’s new Spectrum-X Ethernet switches and SuperNICs are being integrated into the AI infrastructure of Meta and Oracle. These systems are designed to deliver high-performance networking for data centers handling enormous AI workloads, particularly those involving large-scale language and reasoning models.
The partnership reflects the growing trend among hyperscalers to standardize open and accelerated Ethernet architectures that enable faster training, reduced congestion, and efficient resource utilization across millions of GPUs.
NVIDIA’s Vision for Giga-Scale AI
NVIDIA founder and CEO Jensen Huang described Spectrum-X as the “nervous system of the AI factory,” connecting compute resources across vast infrastructures. He noted that the platform allows hyperscalers to train the largest AI models ever built while maintaining performance consistency and reliability.
The Spectrum-X platform combines networking, compute, and software layers into a unified architecture that supports the development of trillion-parameter models — a defining step toward more capable AI systems.
Meta’s Open Networking Integration
Meta will integrate Spectrum Ethernet switches into its Facebook Open Switching System (FBOSS) and Minipack3N architecture. This integration enhances the scalability and predictability of Meta’s next-generation AI infrastructure, which supports global research and application deployment.
According to Meta’s vice president of networking engineering, Gaya Nagarajan, Spectrum-X enables open and efficient networking at an unprecedented scale, aligning with Meta’s long-term vision for distributed AI systems.
Oracle’s AI Factories and Spectrum-X
Oracle Cloud Infrastructure (OCI) will deploy Spectrum-X Ethernet across its AI data centers to interconnect millions of GPUs with greater bandwidth and reduced latency.
Mahesh Thiagarajan, executive vice president of OCI, stated that the adoption of Spectrum-X supports Oracle’s commitment to faster AI training and deployment, helping customers build and manage complex generative AI workloads.
Performance and Technical Highlights
| Feature | Description |
|---|---|
| Platform | NVIDIA Spectrum-X Ethernet with SuperNICs |
| Throughput Efficiency | Up to 95% with congestion-control technology |
| Scalability | Supports interconnection of millions of GPUs |
| Architecture | Integrated with NVIDIA NVLink™, CPUs, and AI software stack |
| Use Case | AI model training, inference, and data center interconnection |
Spectrum-X and Global AI Infrastructure
The Spectrum-X Ethernet platform delivers predictable performance through advanced congestion control, adaptive routing, and AI-driven telemetry. This allows data centers to operate at higher efficiency than traditional Ethernet systems, which typically achieve around 60% throughput.
With its “scale-across” capability, Spectrum-X enables hyperscalers to link data centers across cities and continents, forming vast AI super-factories optimized for distributed workloads.
Industry Implications
The adoption of Spectrum-X by Meta and Oracle represents a pivotal step in expanding AI infrastructure capacity. These implementations demonstrate how open Ethernet-based networking can deliver cloud-scale performance while maintaining flexibility for diverse AI applications.
As AI models grow in complexity, efficient networking has become a defining factor in advancing both research and commercial deployments. Spectrum-X’s integration into leading global data centers underscores its role in enabling that evolution.
Building the AI Backbone
NVIDIA’s Spectrum-X Ethernet platform marks a new phase in high-performance networking for AI. By equipping Meta and Oracle with scalable, open, and efficient infrastructure, it strengthens the backbone of future AI systems and accelerates progress toward globally interconnected computing environments.
Sources: NVIDIA.
Prepared by Ivan Alexander Golden, Founder of THX News™ — an independent news organization delivering timely insights from global official sources. Combines AI-analyzed research with human-edited accuracy and context.
