Next-Gen AI Takes Flight: Oracle Teams Up with NVIDIA for Million-GPU Supercomputing

In a groundbreaking collaboration, Oracle Cloud Infrastructure (OCI) has teamed up with NVIDIA to develop giga-scale AI supercomputers, leveraging NVIDIA's cutting-edge Spectrum-X Ethernet switches. Announced earlier this month, this partnership aims to revolutionize AI workloads by interconnecting millions of GPUs with unprecedented efficiency, marking a significant leap forward in the next generation of AI supercomputing. The partnership was spotlighted in a recent post by NVIDIA Networking on X, showcasing the integration of Oracle’s AI-optimized cloud infrastructure with NVIDIA’s Vera Rubin architecture.

According to Mahesh Thiagarajan, Executive Vice President of Oracle Cloud Infrastructure, “Oracle Cloud Infrastructure is designed from the ground up for AI workloads, and our partnership with NVIDIA extends that AI leadership. By adopting Spectrum-X Ethernet, we can interconnect millions of GPUs with breakthrough efficiency so customers can more quickly train and deploy the next wave of AI.” The collaboration, unveiled in October 2025, positions Spectrum-X Ethernet as a game-changer, achieving an impressive 95% data throughput compared to the 60% offered by standard Ethernet solutions, which is poised to accelerate the training and deployment of complex AI models, including trillion-parameter systems, catering to the growing demands of industries worldwide.

NVIDIA’s Spectrum-X Ethernet switches, combined with the Vera Rubin architecture, enable the creation of AI “factories” capable of linking data centers globally. The technology’s advanced congestion-control capabilities and high bandwidth density make it ideal for hyperscale environments, further supported by the NVIDIA Vera Rubin NVL144 MGX-generation rack servers, which feature energy-efficient liquid cooling and modular designs, as highlighted in NVIDIA’s recent blog posts. Oracle plans to deploy these supercomputers within its OCI Supercluster, which can support up to 131,072 GPUs, delivering performance at the zettascale level. This infrastructure is tailored for demanding AI tasks, including frontier model training, inference, agentic AI, scientific computing, and recommender systems, offering customers a competitive edge in cost-effectiveness and speed.

The NVIDIA-OCI partnership builds on a broader trend of hyperscale pioneers adopting Spectrum-X technology, with companies like Meta and CoreWeave also integrating these switches to enhance their AI data center networks. CoreWeave, for instance, is planning to connect distributed data centers into a unified supercomputer, addressing the limitations of traditional Ethernet networking, which struggles with high latency and unpredictable performance as individual data centers reach capacity constraints. Analysts predict that this collaboration will set a new standard for AI infrastructure, enabling faster time-to-insights and reducing energy costs. Oracle’s unique GPU bare metal instances, paired with ultrafast RDMA cluster networking, promise latency as low as 2.5 microseconds up to 220% better pricing than competing cloud providers.

As AI demand continues to surge, the Oracle-NVIDIA alliance is poised to redefine the landscape of computational power. With plans to contribute the Vera Rubin architecture as an open standard to the Open Compute Project (OCP) consortium, the partnership is expected to foster widespread adoption and innovation across the industry. This development, announced just a day ago on December 11, 2025, underscores the rapid pace of AI advancement and positions Oracle and NVIDIA at the forefront of the technological revolution. For more details on this transformative collaboration, visit the official NVIDIA-OCI partnership page.