The race to power the world’s most advanced artificial intelligence systems is driving a revolution in data centre design, and networking technology is now at the forefront of that transformation. NVIDIA’s Spectrum-X Ethernet platform, built specifically for AI workloads, is emerging as the backbone of this new era, with global technology leaders Meta and Oracle standardising on the technology to accelerate their giga-scale AI infrastructures.
Transforming data centres into AI factories
As trillion-parameter models become the new standard in AI, data centres are evolving into vast AI factories, interconnected environments designed to train, deploy, and scale generative and reasoning models at unprecedented speed. NVIDIA’s Spectrum-X Ethernet switches provide the open, accelerated networking architecture that underpins this transformation, enabling hyperscalers to interconnect millions of GPUs with exceptional efficiency.
Jensen Huang, founder and CEO of NVIDIA, describes Spectrum-X as “the nervous system of the AI factory”. Unlike traditional Ethernet, Spectrum-X has been engineered specifically for AI workloads, offering the speed, scale, and reliability needed to manage exponential data growth. By deploying Spectrum-X Ethernet, both Meta and Oracle can connect large clusters of GPUs into unified, high-performance systems, dramatically reducing AI training times and accelerating time to insight.
Oracle builds giga-scale AI factories
Oracle’s next-generation AI infrastructure is set to be accelerated by the NVIDIA Vera Rubin architecture and interconnected by Spectrum-X Ethernet. This partnership strengthens Oracle Cloud Infrastructure’s (OCI) position as one of the leading environments optimised for AI workloads.
“Oracle Cloud Infrastructure is designed from the ground up for AI,” said Mahesh Thiagarajan, Executive Vice President of Oracle Cloud Infrastructure. “By adopting Spectrum-X Ethernet, we can interconnect millions of GPUs with breakthrough efficiency, allowing customers to train, deploy, and benefit from generative AI faster than ever before.”
This move signifies Oracle’s strategic investment in AI-optimised cloud networking, an increasingly critical capability as enterprises demand more scalable and responsive environments for training large language models (LLMs) and other advanced AI systems. By leveraging NVIDIA’s full-stack platform, including GPUs, CPUs, NVLink™ and Spectrum-X Ethernet, Oracle aims to deliver consistent high performance from compute to network, ensuring its infrastructure can meet the evolving needs of AI-driven enterprises.
Meta advances open networking at scale
Meanwhile, Meta is integrating NVIDIA Spectrum Ethernet switches into its Facebook Open Switching System (FBOSS), a software platform designed to manage and control network switches at massive scale. This integration will accelerate the deployment of Meta’s AI infrastructure, enabling faster training of large-scale models while maintaining its commitment to open networking.
“Meta’s next-generation AI infrastructure requires open and efficient networking at a scale the industry has never seen before,” explained Gaya Nagarajan, Vice President of Networking Engineering at Meta. “By integrating NVIDIA Spectrum Ethernet into the Minipack3N switch and FBOSS, we can extend our open networking approach while unlocking the efficiency and predictability needed to train ever-larger models and bring generative AI applications to billions of people.”
The combination of open architecture and AI-optimised Ethernet aligns perfectly with Meta’s long-standing strategy of designing scalable, modular data centre systems. Spectrum-X Ethernet allows Meta to handle the immense bandwidth requirements of next-generation AI workloads while maintaining flexibility and interoperability across its global infrastructure.
The NVIDIA Spectrum-X Ethernet Platform: Built for the Trillion-Parameter Era
The NVIDIA Spectrum-X Ethernet platform, encompassing both Spectrum-X switches and Spectrum-X SuperNICs, is the first Ethernet platform designed specifically for AI. It addresses one of the most critical challenges in AI-scale computing, achieving predictable, high-throughput performance across millions of simultaneous data flows.
In large-scale deployments, conventional Ethernet typically achieves around 60% throughput due to flow collisions and congestion. Spectrum-X, however, delivers record-setting efficiency of up to 95% data throughput, thanks to its advanced congestion-control technology. This leap in performance not only boosts the speed of AI training but also improves the overall economics of large-scale infrastructure deployment.
The Spectrum-XGS Ethernet technology extends these capabilities even further, enabling “scale-across” functionality that connects data centres across cities, nations, and continents into vast, giga-scale AI super-factories. This allows organisations like Meta and Oracle to create unified global AI infrastructures capable of sharing workloads, data, and insights seamlessly.
Efficiency, predictability, and scalability
Spectrum-X’s architecture leverages NVIDIA’s end-to-end stack, from GPUs and CPUs to NVLink interconnects and AI-driven telemetry, to ensure efficient communication between compute and network layers. Its adaptive routing algorithms dynamically manage traffic to avoid congestion, while telemetry provides real-time insights into network performance.
For digital construction, engineering, and infrastructure companies operating in the AI-driven digital transformation era, the implications are significant. As generative AI models become embedded in design, project management, and digital twin applications, the ability to train and deploy models faster and at greater scale will redefine industry competitiveness. Technologies such as Spectrum-X Ethernet are laying the groundwork for these AI-powered ecosystems, where data moves seamlessly from model to machine.
Enabling the next wave of AI innovation
NVIDIA’s Spectrum-X Ethernet marks a major milestone in the evolution of data centre networking. By standardising on this platform, Meta and Oracle are setting the benchmark for AI infrastructure design, open, efficient, and ready for the trillion-parameter model era.
As the boundaries between computing, networking, and AI continue to blur, Spectrum-X provides the foundation for an interconnected AI future, one where networks don’t just move data, but actively enable intelligence at global scale.
Read next: Human-centred collaboration in BIM projects – Build in Digital
Are you a building professional? Sign up for a FREE MEMBERSHIP to upload news stories, post job vacancies, and connect with colleagues on our secure social feed.
