[News] NVIDIA Expands AI Networking Push as Oracle and Meta Adopt Spectrum-X and OCP Technologies

According to Mydrivers, NVIDIA announced that Meta and Oracle will upgrade their AI data center networks using NVIDIA Spectrum-X Ethernet adapters and switches. Meta will integrate the Spectrum-X Ethernet switch into its Facebook Open Switching System (FBOSS) network infrastructure—a software platform designed to manage and control large-scale network switches. The combination of Spectrum-X and FBOSS will accelerate large-scale deployment and enhance AI training efficiency.

The report notes that Oracle plans to build a “Giga-Scale” AI factory, powered by NVIDIA Vera Rubin architecture for compute acceleration and interconnected through Spectrum-X Ethernet. A key component of NVIDIA’s full-stack AI platform, Spectrum-X is the industry’s first Ethernet networking solution purpose-built for AI and trillion-parameter models. It features the Spectrum-X Ethernet switch and Spectrum-X SuperNIC adapter, capable of linking millions of GPUs.

NVIDIA claims that Spectrum-X delivers record-breaking efficiency, enabling the world’s largest AI supercomputers to achieve 95% data throughput, compared to roughly 60% in conventional Ethernet systems. Notably, the report adds that the CX9 SuperNIC, also part of the next-generation platform, has completed tape-out and will offer 1,600 Gbps (1.6 Tbps) of bandwidth.

NVIDIA Expands Industry Partnerships to Strengthen Open and Scalable AI Infrastructure

As TechNews notes, NVIDIA has announced a series of strategic partnerships focused on interconnect and data center integration. Fujitsu’s Monaka processor will use the NVLink Fusion protocol to achieve tight coupling with NVIDIA GPUs. Intel will manufacture Fusion-compatible CPUs to connect with NVIDIA GPUs. NVIDIA is also adding Samsung to its existing list of partners—Alchip, Astera, and MediaTek—to help the community integrate their accelerators with NVIDIA CPUs or NVLink-based designs. TechNews further mentions that the Stargate data center, jointly developed by Oracle and OpenAI, was built using NVIDIA Spectrum and OCP technologies, achieving 95% effective bandwidth and zero application latency.

In addition, TechNews notes that in the second half of 2027, NVIDIA will introduce the Kyber design, aimed at connecting 500 GPUs within a single rack. To achieve this scale and power density, NVIDIA is working with industry partners to promote 800-volt DC infrastructure. The company emphasized that this architecture is being developed in collaboration with the OCP community, enabling the creation of data centers capable of delivering extraordinary AI performance.

Read more

• [News] NVIDIA Adds Samsung Foundry to NVLink Fusion Ecosystem for Custom Silicon Manufacturing
• ASIC Set to Outpace GPU? NVIDIA’s Scale-Up and Beyond

(Photo credit: TrendForce at 2025 OCP Global Summit)