Blog
Wireless, Network, Computing

From Proof-of-Concept to Commercial Deployment of AI-RAN ~SoftBank's Initiatives to Build Distributed Infrastructure for the AI Era, Including Collaboration with NVIDIA~

#AI-RAN #AI #AITRAS #PhysicalAI

Jul 16, 2026
SoftBank Corp.

SoftBank is actively working on technology development, verification, and ecosystem formation toward the commercial deployment of AI-RAN. This article reviews SoftBank's past initiatives for outdoor trials and commercialization regarding "AITRAS," an integrated AI-RAN solution being developed by SoftBank, including collaborative efforts with NVIDIA.

AITRAS is SoftBank's integrated AI-RAN solution that enables both AI and RAN (Radio Access Network) to run on the same accelerated computing platform. It aims to provide high-capacity, high-performance, and carrier-grade RAN while simultaneously and efficiently operating a wide range of AI applications, including generative, agentic and physical AI.

AI-RAN is more than just a next-generation mobile network technology. It extends AI infrastructure close to the actual sites where robots, cameras, sensors, and industrial equipment operate, serving as a vital foundation to support AI services like Physical AI that require real-time processing. By integrating AI and telecommunications, SoftBank aims to realize a low-latency, high-capacity, and secure social infrastructure for the AI era.

Contents

01.

The History of AI-RAN / AITRAS Technological Development
02.

Progress Toward Commercialization Through Field Trials
03.

Expanding the AI-RAN Ecosystem
04.

Broadening Use Cases for AI-RAN
05.

Future Direction: Toward Commercialization and Social Implementation

1. The History of AI-RAN / AITRAS Technological Development

SoftBank's initiatives regarding AI-RAN have evolved in stages. In March 2023, at the "AI-on-5G Lab," jointly established with NVIDIA, SoftBank successfully verified end-to-end (E2E) communication using an actual machine, spanning from GPU-based vRAN to MEC applications. In this verification, the RAN's DU (Distributed Unit) and the MEC's AI application ran on a hardware infrastructure of the identical configuration, and real-time human detection utilizing AI was successfully achieved. This confirmed the potential of utilizing GPUs for both RAN and AI processing.

In May of the same year, SoftBank and NVIDIA announced a collaboration to construct a next-generation platform for generative AI and 5G/6G. This platform is based on the NVIDIA GH200 Grace Hopper Superchip and is planned for deployment across SoftBank's new distributed AI data centers throughout Japan. The concept of delivering generative AI and wireless communication applications on a multi-tenant, common server platform was a crucial step leading to today's AI-RAN and distributed AI infrastructure initiatives.

In November 2024, SoftBank announced the development of "AITRAS," an integrated AI-RAN solution. AITRAS is the commercialized manifestation of AI-RAN.

With AITRAS, SoftBank aims to efficiently execute both AI and RAN processing on the same infrastructure utilizing its proprietary RAN software and orchestration technology. For L1 software development, SoftBank leverages the NVIDIA AI Aerial platform to achieve the high stability and performance essential for carrier-grade networks. Furthermore, SoftBank has developed an orchestrator that dynamically allocates computing resources according to the distinct application characteristics of AI and RAN, aiming to simultaneously enhance network performance and efficiently execute AI applications.

2. Progress Toward Commercialization Through Field Trials

The establishment of verification environments is also moving forward. At Keio University's Shonan Fujisawa Campus (SFC), SoftBank built an AI-RAN verification environment designed for urban area operations, testing AI and RAN processing using servers equipped with the NVIDIA GH200. In this environment, a single AITRAS server successfully processed 20 cells with a 100MHz 4T4R profile, demonstrating simultaneous video streaming across 100 devices, quadrupedal robots utilizing Large Language Models (LLMs), and AI applications for autonomous driving.

Related Press Release: please refer to the "Overview of AI-RAN Outdoor Test" section.

Another critical domain of AI-RAN is "AI for RAN," which enhances the capabilities of the RAN itself using AI. In August 2025, SoftBank developed a new AI architecture utilizing the high-performance AI model "Transformer" for wireless signal processing, successfully improving 5G communication speeds—namely, throughput—by approximately 30%. This trial confirmed real-time operation in an actual wireless environment compliant with 3GPP 5G standard specifications, demonstrating that AI-RAN is advancing from the conceptual stage to practical application.

Furthermore, in a Massive MIMO demonstration for AITRAS conducted at NVIDIA's headquarters in Santa Clara, California, USA, SoftBank built an AI-RAN system that fully executes wireless signal processing via software utilizing NVIDIA GPUs, and successfully demonstrated a downlink 16-layer MU-MIMO in an outdoor environment. Compared to conventional 4-layer configurations, both spectral efficiency and throughput improved by approximately three times, marking a major milestone toward the commercialization of AI-RAN.

Related Press Release: SoftBank Corp.'s AI-RAN Realizes "Software-only Massive MIMO" on GPU

* NVIDIA’s Technical Blog also features the improvement of spectral efficiency through AI-native RAN and NVIDIA AI Aerial, referencing the outdoor 16-layer Massive MU-MIMO trial conducted by SoftBank and NVIDIA.（NVIDIA Developer）

3. Expanding the AI-RAN Ecosystem

To advance the social implementation of AI-RAN, fostering an industry-wide ecosystem is just as vital as technological development and trials. In February 2024, leading telecom and AI companies including SoftBank and NVIDIA, along with universities, established the AI-RAN Alliance. Embracing three core themes—"AI for RAN," "AI and RAN," and "AI on RAN"—the AI-RAN Alliance drives innovations to redefine existing global RAN infrastructure into AI infrastructure through the fusion of AI and RAN.

Related Press Release: AI-RAN Alliance Launch

As of July 2026, the AI-RAN Alliance has expanded globally to include over 140 organizations, comprising participating companies, universities, and research institutions, demonstrating that the fusion of AI and telecommunications is becoming a global trend. SoftBank participated as a founding member and leads the Alliance's activities, with SoftBank's Alex Jinsung Choi serving as Chair. Furthermore, SoftBank joined the OCUDU Ecosystem Foundation, launched by the Linux Foundation, as a founding member to drive the formation of an open and global ecosystem for the advancement of distributed AI-RAN infrastructure.

SoftBank itself is taking steps to expand the AI-RAN ecosystem. It open-sourced the Dynamic Scoring Framework (DSF), a core function of the AITRAS orchestrator. The DSF enables optimization based on resource status across multi-cluster environments. By doing so, SoftBank aims to foster collaboration with partner companies and the open-source software (OSS) community, lower the entry barriers to AI-RAN, and accelerate commercial deployment.

Additionally, the AITRAS orchestrator is expanding its capabilities through partnerships. Collaborating with Ericsson, SoftBank integrated the AITRAS orchestrator with the Ericsson Intelligent Automation Platform, an O-RAN compliant Service Management and Orchestration (SMO) framework, enabling dynamic computing resource allocation across both AI and RAN. In cooperation with Nokia, SoftBank linked the Nokia AI-RAN External Compute Engine with the AITRAS orchestrator, allowing computing resources to be provisioned for external AI workloads within an AI-RAN environment. This enables dynamic allocation between RAN computing resources from Ericsson and Nokia—which currently support SoftBank's mobile network—and AI computing resources, advancing initiatives to create new value by utilizing telecommunications infrastructure in the AI-RAN era.

4. Broadening Use Cases for AI-RAN

Beyond advanced communication networks, AI-RAN is highly anticipated as a foundational base to support AI utilization for enterprises. SoftBank is working on developing Physical AI, Edge AI, and industrial AI applications leveraging AI-RAN and AI data centers.

In partnership with Yaskawa Electric Corporation, SoftBank combined AI-RAN with AI robotics to jointly develop a use case for an office-oriented Physical AI robot that leverages AI operating on MEC.

Furthermore, using the "AI Data Center GPU Cloud" as a development foundation for Physical AI, and in collaboration with NVIDIA, SoftBank is conducting trials for a deformable object manipulation system. This system streamlines everything from robot motion data collection to AI training, evaluation, and deployment on physical robots.

These initiatives utilize AI infrastructure as a foundation for robots to operate flexibly in the physical world.

With Mitsubishi Heavy Industries, Ltd. (MHI), SoftBank launched a collaboration in the edge data center domain leveraging AITRAS. SoftBank established an AI-RAN environment using AITRAS at an edge data center located within MHI's Yokohama Hardtech Hub, and is conducting a field trial for edge AI applications in an on-premises environment. This trial verifies the realization of stable communication, high-speed AI processing, and advanced security by executing AI inference within an environment isolated from external networks.

With Ericsson, SoftBank successfully verified Physical AI over a low-latency, highly reliable network utilizing AI-RAN. This trial validates network controls designed to stably operate Physical AI applications, such as robots, through the dynamic offloading of AI processing and the optimization of the telecommunications network.

Through these initiatives, SoftBank is evolving AI-RAN beyond a mere upgrade to communication networks into a distributed infrastructure for utilizing AI across diverse fields including industrial sites, robotics, and edge AI.

5. Future Direction: Toward Commercialization and Social Implementation

SoftBank is advancing the development of AITRAS with the goal of introducing it into its commercial network from fiscal year 2026 onward. Additionally, as part of its efforts toward full AI-RAN commercial deployments, SoftBank is evaluating the NVIDIA Aerial RAN computer platforms, NVIDIA RTX PRO 4500 and NVIDIA RTX PRO 6000.

SoftBank will continue to push forward with development and verification toward the commercialization of AITRAS and the social implementation of AI-RAN, collaborating with various partner companies and organizations, including NVIDIA. By combining nationwide telecommunications infrastructure with AI computing, software, and AI applications, SoftBank aims to evolve communication networks into a distributed infrastructure for the AI era, establishing a societal foundation that supports diverse AI applications.