【New Update : Mar. 3】 Press Releases on AI-RAN Development

SoftBank announced the concept of "AI-RAN" in 2023, envisioning it as the cornerstone of next-generation mobile networks supporting the AI-driven society. AI-RAN is an innovative architecture that integrates AI applications and Radio Access Network (RAN) on the same computing platform. In November 2024, as part of efforts toward the commercialization of AI-RAN, SoftBank announced its research and development achievements related to the establishment of AI-RAN’s technological foundation, focusing primarily on the orchestrator, RAN infrastructure, and AI infrastructure.

【New Update : Nov. 28】 Press Releases on AI-RAN Development：https://www.softbank.jp/en/corp/technology/research/news/052/

In March 2025, SoftBank released six press releases regarding significant new achievements towards the commercialization of its AI-RAN solution, "AITRAS." These announcements include substantial enhancements to AITRAS's functionality and technological developments aimed at accelerating the adoption of the AI-RAN infrastructure. This strategic initiative encompasses numerous innovative approaches, including collaborations with industry-leading companies such as Nvidia, Nokia, RedHat, Ericsson, and Fujitsu, as part of a broader strategic effort. This page explains the relevance of each press release.

Three Concepts of "AI-RAN"

Structure of AITRAS and Positioning of Each Press Release

1. SoftBank Corp. Demonstrates RAN Performance Enhancement with AI Technology

Following the November 2024 announcement about SoftBank’s “AI-on-RAN" and “AI-and-RAN" initiatives, this press release focuses on the third key initiative, “AI-for-RAN," and the latest performance enhancements that have been achieved.

SoftBank demonstrated the enhancement of RAN performance using AI technology in three use cases as part of its initiatives aimed at enhancing RAN performance through AI.

In the initiative jointly conducted with NVIDIA and Fujitsu to enhance channel estimation accuracy for signals received at base stations, known as Uplink (“UL”) Channel Interpolation, SoftBank confirmed that the UL user throughput improved by approximately 20% on smartphones. Additionally, in predicting the timing of Sounding Reference Signals (“SRS”) transmitted from devices, SoftBank achieved an approximate 13% improvement in downlink (“DL”) throughput in high-mobility scenarios. Furthermore, by considering factors such as correlation between users, wireless quality, and priority in MAC Scheduling, SoftBank was able to improve average user throughput by approximately 8%. Moreover, SoftBank developed a system-level simulator as a verification platform to efficiently evaluate the enhancement of RAN performance through AI technology. With the realization of these "AI-for-RAN" technologies, the provision of even higher-quality communication services is expected.

2. SoftBank Corp.’s “AITRAS” Implements Centralized and Distributed AI-RAN Architectures Using Arm-based NVIDIA Platform

AITRAS is a product with a flexible architecture that can be tailored to customers and use cases. This initiative realizes the diversification of hardware configurations when deploying AITRAS.

Focused on the commercialization of AITRAS, SoftBank, in collaboration with Fujitsu, completed the implementation of CU (Central Unit) functions on the NVIDIA Grace CPU Superchip platform. Through this development, the number of RUs (Radio Units) that can be accommodated by a single CU is doubled, which helps reduce the number of servers required for CU processing.

Additionally, SoftBank realized a configuration pattern where both DU and CU functions are deployed on a single NVIDIA GH200 Grace Hopper Superchip. This allows for AITRAS to be realized not only in a C-RAN architecture (where DU/CU is installed in data centers and connected to multiple site RUs), but also in a D-RAN architecture (where DU/CU is installed at the same site as the RU). This makes it possible to deploy AITRAS at the optimal cost in areas with localized high AI demand, such as enterprise facilities. SoftBank plans to progressively deploy these AITRAS configurations into outdoor testing environments starting in fiscal year 2025 and to proceed with performance verifications.

3. SoftBank Corp. Develops Secure Access Feature to Edge AI Servers on “AITRAS” Utilizing Local Breakout Technology

"AI-on-RAN," one of the key concepts of AI-RAN, is a technology for deploying AI applications on the network edge. This announcement is about the technology for realizing secure operation of edge AI applications.

SoftBank developed a feature utilizing Local Breakout technology, one of the features of 5G SA, enabling secure access to edge AI servers on AITRAS. This was achieved through edge routing using Local Breakout technologies such as URSP (User Equipment Route Selection Policy) and LADN (Local Area Data Network). By leveraging URSP, PDU (Protocol Data Unit) sessions can be assigned to the network for standard Internet connections and to the edge AI servers on AITRAS, respectively. This enabled the development of applications that can access both public and private LLMs and use them appropriately. Additionally, by utilizing LADN, SoftBank developed applications that restrict access to private LLMs within specific geographical areas (e.g., a factory). This development is expected to promote the utilization of AI, including the handling of corporate confidential data such as that from RAG.

4. SoftBank Corp. and Red Hat Develop Solution to Optimize Power Consumption in AI-RAN Data Centers

AITRAS features orchestration capabilities that dynamically optimize resources for RAN and AI. This announcement pertains to the addition of power consumption optimization functions to the AITRAS Orchestrator capabilities through collaboration with Red Hat.

SoftBank and Red Hat enabled the monitoring of power usage for AI applications running on AITRAS by implementing Red Hat's OSS (open source software), Kepler, into the virtualization infrastructure of AITRAS. With this implementation, the AITRAS Orchestrator can dynamically allocate optimal resources by considering power consumption, in addition to AI application priority and server availability. Moving forward, this advanced AITRAS Orchestrator will facilitate cooperation between AITRAS data centers planned to be deployed across Japan, contributing to regional power load distribution and improved utilization rates of renewable energy.

5. SoftBank Corp. and Nokia Achieve AI and vRAN Coexistence with Automated Optimal Resource Allocation on a Single Server

This announcement showcases a new implementation example of "AI-and-RAN," one of the three main concepts of AI-RAN, which involves the coexistence of AI applications and vRAN on the same server.

SoftBank and Nokia developed a solution that enables the coexistence of AI applications and Nokia's vRAN software on a single server. For this solution, Nokia extended the functionality of the SmartNIC (Smart Network Interface Card), which performs part of the processing for the vDU (virtualized Distributed Unit), to operate on a single server, making it possible to share resources between vRAN and AI applications. Additionally, the two companies integrated Nokia's EMS (Element Management System), MantaRay NM, with the AITRAS Orchestrator to allow for the optimal allocation of server resources based on traffic behavior predictions derived from monitoring data, such as the number of connected vRAN users. This enables network operators to optimize their equipment utilization efficiency.

6. SoftBank Corp. and Ericsson Sign MoU for Joint Development to Advance AI-RAN

SoftBank and Ericsson signed a Memorandum of Understanding to expand their collaboration to further advance AI-RAN and conduct joint research and development in applied fields. Specifically, they aim to implement and verify Ericsson's Cloud RAN software using GPU technology. Additionally, they will work on integrating SoftBank's AI-RAN integrated solution "AITRAS" orchestrator with Ericsson's Open Network Management and Automation Platform to achieve optimal coexistence method for AI and vRAN applications. Furthermore, they will challenge to optimize placement of control functionalities for robotic AI applications using Ericsson's AI compute offloading framework.

This collaboration is expected to advance the fusion of SoftBank's "AITRAS" and Ericsson's Cloud RAN, contributing not only to the evolution of RAN technologies and mobile networks but also significantly to the development of the robotics industry.

Through these six press releases, SoftBank announced significant achievements in the establishment and commercialization of the AI-RAN technology infrastructure, centered around orchestrators, RAN infrastructure, and AI infrastructure. Furthermore, to accelerate the evolution of AI-RAN, a new announcement are planned at NVIDIA GTC, which will be held in San Jose, USA, from March 17 to 21, 2025. Moving forward, SoftBank will continue to lead innovations in AI-RAN technology and work on creating new value through the integration of communication technologies and AI.

Related Press Releases
・【New Update : Nov. 28】 Press Releases on AI-RAN Development
https://www.softbank.jp/en/corp/technology/research/news/052/

Related Websites
・AI-RAN Summary
https://www.softbank.jp/en/corp/technology/research/story-event/057/

・AI-RAN Introduction Video