Press Releases 2024

SoftBank Corp. Implements
NVIDIA AI Enterprise on Edge AI Server of
“AITRAS” Converged AI-RAN Solution

Enabling the deployment of ultra-low latency, highly secure AI application on AI-RAN

November 13, 2024
SoftBank Corp.

SoftBank Corp. (“SoftBank”) announced it implemented NVIDIA AI Enterprise software on its edge AI servers as part of “AITRAS,” a converged solution currently under development by SoftBank that enables the operation of AI and RAN (Radio Access Network) on the same network infrastructure. This initiative aims to accelerate the deployment of AI applications on AI-RAN. Consequently, it will become easier to develop and deploy various AI applications on edge AI servers by utilizing Open Source Software (OSS) and high-performance AI foundation models from partner companies.

AITRAS

NVIDIA AI Enterprise is a comprehensive software platform equipped with the essential features required for enterprise large language models (LLMs) development and deployment. It includes NVIDIA NeMo for building and customizing foundation models and NVIDIA NIM inference microservice for optimizing model performance at scale. By leveraging NVIDIA AI Enterprise, it is possible to efficiently develop and deploy AI applications such as custom LLMs trained with industry- or sector-specific knowledge, and Retrieval-Augmented Generation (RAG), which combines LLMs with external information retrieval, such as a company's proprietary data.

SoftBank has developed three AI applications utilizing NVIDIA AI Enterprise implemented on AITRAS.

(1) Cloud Robot Utilizing Ultra-low-latency LLMs

By applying LLMs to robot motion generation, it is expected that robots will be able to respond appropriately even in unfamiliar situations. In this case, running LLMs on AITRAS with access to greater computational resources, rather than relying on the limited computational resources of the robot itself, enables more flexible and sophisticated decision-making capabilities. SoftBank has developed an ultra-low latency LLM capable of generating robot motions in real-time based on sensor information from the robot. The developed model operates on edge AI servers within AITRAS, allowing for the low-latency input of sensor data from the robot and output of control information from the LLM. This enables real-time control of robots by LLMs running on external computing machines.

Cloud Robot Utilizing Ultra-low-latency LLMs

(2) RAG Menu @Edge

As a means for companies to leverage their proprietary data in generative AI, a technology called RAG, which searches external databases and extracts information and enables LLMs to respond based on that information, has gained attention. SoftBank has developed enterprise-grade RAG applications on AITRAS, incorporating multiple technologies that enhance the convenience and accuracy of RAG. By integrating corporate data, it becomes possible to generate responses based on the latest internal information, allowing generative AI to handle tasks specific to a company's operations. Since all data is processed exclusively on AITRAS's edge AI servers in a closed environment, it can be executed in a more secure environment compared to using cloud services over the Internet.

RAG Menu @Edge

(3) Traffic Understanding Multimodal AI for Autonomous Driving

In the implementation of autonomous driving in society, improving the safety of autonomous vehicles and reducing operational costs are key challenges. To address these challenges, SoftBank has developed “Traffic Understanding Multimodal AI,” which is a comprehensive AI foundational model trained not only with general traffic knowledge from manuals and Japanese traffic regulations, but also with risk scenarios and countermeasures for common driving scenarios and unpredictable situations. By operating Traffic Understanding Multimodal AI in real-time with low latency and high security on the edge AI servers of AITRAS, autonomous vehicles can receive external support when faced with unforeseen circumstances. The AI analyzes traffic conditions and risks and generates appropriate instructions for the vehicle, enabling remote support for autonomous driving. For more details, please refer to the press release dated November 5, 2024, titled “SoftBank Corp. Develops Traffic Understanding Multimodal AI for Autonomous Driving that Operates on Low-latency Edge AI Servers”.

Ronnie Vasishta, Senior Vice President, Telecommunications, NVIDIA said, “As AI applications flourish across industries, enterprises seek innovative ways to deploy these services with speed, security, and efficiency at the edge. NVIDIA AI Enterprise on SoftBank's AITRAS addresses this by enabling easier development and deployment of high-performance AI models, offering ultra-low latency and enhanced security for applications ranging from autonomous driving to cloud robotics.”

Makoto Noda, Senior Vice President at SoftBank said, “SoftBank is actively promoting various initiatives aimed at addressing the challenges faced by businesses and society through digital transformation (DX). With the implementation of NVIDIA AI Enterprise within our AI-RAN converged solution, we anticipate the accelerated development of AI applications tailored to a variety of use cases, significantly broadening the potential of our customers' businesses.”

Going forward, SoftBank will continue to contribute to solving societal challenges through AI-RAN by not only developing the foundational technologies necessary for the deployment of AI applications on AI-RAN but also by promoting the development of AI applications that leverage these technologies.

  • SoftBank, the SoftBank name and logo are registered trademarks or trademarks of SoftBank Group Corp. in Japan and other countries.
  • Other company, product and service names in this press release are registered trademarks or trademarks of the respective companies.