NVIDIA powers SoftBank’s data centers using Grace Hopper Superchip

NVIDIA and SoftBank have joined forces to develop an innovative platform for generative AI and 5G/6G applications. This collaboration is built on the NVIDIA GH200 Grace Hopper Superchip and aims to establish distributed AI data centers across Japan, as part of SoftBank’s expansion plans.
SoftBank 5G networkThe partnership between SoftBank and NVIDIA will enable the deployment of generative AI applications and services on a shared server platform, reducing costs and improving energy efficiency. SoftBank will construct data centers capable of hosting generative AI and wireless applications, leveraging the expertise of NVIDIA. This initiative will facilitate the global adoption of generative AI while optimizing performance, scalability, and resource utilization for various application workloads.

Junichi Miyakawa, president and CEO of SoftBank, stated that the collaboration with NVIDIA would significantly enhance infrastructure performance through AI utilization, including RAN optimization. Furthermore, it aims to reduce energy consumption and establish an interconnected network of data centers capable of hosting diverse generative AI applications.

Jensen Huang, founder and CEO of NVIDIA, said NVIDIA Grace Hopper is a groundbreaking computing platform designed to process and scale generative AI services.

The upcoming data centers will be distributed more evenly across SoftBank’s infrastructure and will handle both AI and 5G workloads. This strategic approach will enable the centers to operate at peak capacity, ensuring low latency and significantly reduced overall energy costs.

SoftBank intends to explore the development of 5G/6G applications for various sectors, including autonomous driving, AI factories, augmented and virtual reality, computer vision, and digital twins.

The combination of NVIDIA Grace Hopper and NVIDIA BlueField-3 data processing units will enhance the software-defined 5G vRAN and generative AI applications without relying on specialized hardware accelerators or dedicated 5G CPUs. Additionally, the NVIDIA Spectrum Ethernet switch with BlueField-3 will provide precise timing protocols for 5G networks.

This solution achieves breakthrough 5G speeds using an NVIDIA-accelerated 1U MGX-based server design, achieving an industry-leading downlink capacity of 36Gbps. Previously, operators struggled to deliver such high downlink capacity using standard servers.

NVIDIA MGX serves as a modular reference architecture that empowers system manufacturers and hyperscale customers to swiftly and cost-effectively create a wide range of server variations for AI, HPC, and NVIDIA Omniverse applications.

By integrating NVIDIA Aerial software for high-performance, software-defined, cloud-native 5G networks, SoftBank’s 5G base stations will allow operators to dynamically allocate compute resources, delivering 2.5 times greater power efficiency compared to competing products.

Rene Haas, CEO of Arm, said the Arm Neoverse-based Grace Hopper Superchip from NVIDIA enables SoftBank’s new 5G data centers to handle demanding compute- and memory-intensive applications while delivering exponential efficiency gains to software-defined 5G and AI on Arm.