Nvidia, a leading chipmaker, and SoftBank Corp, the telecommunications division of SoftBank Group, have successfully piloted the world’s first AI and 5G telecom network. This innovative system is capable of running AI and 5G workloads simultaneously, a process called an artificial intelligence radio access network (AI-RAN). This breakthrough could open up billions in new revenue opportunities for telecom companies. The network utilizes Nvidia’s AI Aerial accelerated computing platform.
The companies say it can accelerate generative AI, deep learning, and machine learning applications for telecoms while providing the performance needed for AI-native radio access network (AI-RAN) deployments. This is accomplished by using NVIDIA’s accelerated computing platform with software-defined radio and virtualized antenna arrays (vRAN). The system is designed to allow AI and RAN to be deployed on the same accelerated infrastructure, enabling telcos to become the local AI compute provider for enterprise customers while meeting data sovereignty and security requirements.
During the trial, which took place in Japan, Nvidia and SoftBank demonstrated the benefits of this approach by running AI inferencing on a local AI-RAN server instead of sending it to a cloud-based solution. The results showed that the robot connected to the local AI-RAN performed a sequence of actions much faster and more accurately than the same robot connected to the cloud. The demonstration highlighted the importance of AI and robotics, enabling the robot to quickly respond to a human’s movements and perform tasks independently.
This trial is the latest milestone in Nvidia’s journey to create a complete software-defined, cloud-native 5G network. In this journey, Nvidia is working with partners across the ecosystem, from wireless telecommunications operators, like T-Mobile and AT&T, to silicon vendors, like Samsung, Huawei, and Nokia; networking stack providers, including Aarna Networks, Canonical, and Red Hat; server infrastructure providers, such as Dell Technologies and Hewlett Packard Enterprise; and edge solutions providers, including Vapor IO and World Wide Technology.
Nvidia’s new Grace CPU “super chip,” with its BlueField-3 data processing units, will accelerate vRAN workloads and support the software-defined 5G architecture. The company’s MGX reference architecture will enable manufacturers and hyperscalers to build server variations optimized for vRAN use cases, including L1 (Long Range Access) acceleration.
SoftBank will deploy NVIDIA DGX SuperPOD systems with Blackwell chips to power its new, distributed AI data centers. The new AI-related business will develop generative AI models for SoftBank’s businesses and other industries, such as telecommunications, finance, healthcare, and retail. The data center will also serve as the primary training platform for the SoftBank AI Supercomputer, the world’s most potent extensive language model development system, which is being built with NVIDIA DGX B200 systems. When completed, it will be Japan’s most advanced accelerated computing system and deliver breakthrough performances for various generative AI applications.
