SoftBank’s sustained drive to realize the AI-RAN Alliance

1. Two Years of AI-RAN: Vision and Momentum

Mobile World Congress in Barcelona is always an exciting time in our industry. Two years ago, we announced the launch of the AI-RAN Alliance. As we approach the 2nd year anniversary 130+ members strong, and continue to build on the potential of an AI-native RAN, I have the honor of sharing a closer look at the ground we covered. At the AI-RAN Alliance, we are revolutionizing telecom by embedding AI across the RAN to enable smarter, more efficient, and more adaptable mobile networks.

As a founding member of the Alliance, the SoftBank team has had the distinct honor of overseeing the rapid growth of the alliance- my friend and mentor Alex Choi has served as Chairman since inception, and I was honored with taking the baton from Rajeev Koodli and Ryuji Wakikawa on the Technical Steering Committee and Board of Directors respectively. Through active participation in the governing bodies and technical contribution to working groups and task groups, we help define objectives and tangible outcomes.

2. Advancing the Technical Foundation Across the Working Groups

The Alliance is organized in workgroups and task groups. AI-for-RAN (WG1) to inject intelligence into the physical layer for spectral efficiency, AI-and-RAN (WG2) to unify computing and communications on a single platform, and AI-on-RAN (WG3) to turn the network edge into a powerhouse for mobile-native, AI-native services. This technical foundation is accelerated by agile strike teams, one of them the Agentic AI Task Group.

During the past year, SoftBank has been very active across the three working groups, in line with our taste for realizing technology through code that runs.

In WG1, some of the main contributions were across these 2 areas:

・AI switching mechanism - for this newly approved work item, SoftBank has provided the key insight on the need for AI switching. The work item is intended to be expandable, with UL channel estimation as the first use case. This use case has been selected for demo at MWC’26, and there is more detail below.

・Sounding Reference Signal (SRS) prediction - demonstrating performance gains from SRS prediction through system-level simulations. Back in Sep’25, SoftBank showcased that a unified transformer architecture is also applicable to this use case with the presentation titled “Unified Transformer Architecture for Low-Latency and Scalable Wireless Signal Processing”.

SoftBank has also been very active in AI-RAN Alliance’s Work Group 2 (WG2).

Leading AI-RAN orchestration discussion - SoftBank proposed to create a WG document capturing the intended roles and capabilities of the AI-RAN Orchestrator. SoftBank also served as the rapporteur of the document which was released just before the MWC. The details on this “AI-RAN Platform and Infrastructure (P&I) Orchestrator” can be found here.

Enhancing AITRAS Orchestrator for closer alignment with AI-RAN P&I Orchestrator - SoftBank has refined the design and implementation of AITRAS Orchestrator, which is built on Open Cluster Management (OCM). As a key enabler for AI-RAN orchestration, SoftBank developed the Dynamic Scoring Framework and upstreamed it to OCM as an Addon.

As part of WG3, SoftBank reviewed the hot off the press whitepaper on “Differentiated Connectivity”, which aligns very well with our vision for what is coming up on Physical AI.

3. MWC’26 Demonstrations: Running Code in Action

At MWC’26, SoftBank is bringing 3 demos under the AI-RAN Alliance umbrella-

・As part of WG1, SoftBank has put together an AI switching mechanism demo with Northeastern University. This demo showcases dynamic switching between an AI-based channel estimator and a non-AI-based channel estimator using a machine learning (ML)-based switching logic to ensure optimal performance under changing conditions. This ML-based switching logic monitors live RAN performance metrics and channel conditions in the current operating environment, and selects the optimal channel estimator in real time without interrupting the processing pipeline—even under rapidly changing conditions. This demo provides a pragmatic blueprint for a hybrid AI/non-AI RAN design that selects the best approach in real time from multiple AI models and non-AI methods, optimizing not only RAN performance but also operational factors such as power consumption. The demo can be found at the AI-RAN Alliance booth (Hall 2 Booth #2E45) or through this link.

・As part of WG2, SoftBank is showcasing the AITRAS Orchestrator as a concrete implementation of the AI-RAN Platform and Infrastructure Orchestrator. The demo covers 2 use cases focusing on performance-based optimization and energy-based optimization. The Dynamic Scoring Framework runs under the hood throughout the demo and is applied in both use cases. The demo shows that the framework can handle diverse evaluation criteria and that, based on the resulting scores, the orchestrator can optimize resources accordingly. With this framework in place, developers have the flexibility to design their own scoring models as well as optimization logic that leverages the calculated scores. We believe this will further strengthen the AI-RAN ecosystem and accelerate innovation across the industry. This demo can also be found at the AI-RAN Alliance booth (Hall 2 Booth #2E45) and through this link.

・Additionally, with the Agentic AI Task Group, SoftBank has put together “The Agent RAN - how to create and use Large Telecom Model (LTM)” demo in association with Northeastern University, KeySight, and zTouch. This demo showcases intent-based, AI-controlled RAN slicing in an emergency robotics scenario, where LLM-powered agents dynamically adjust uplink power control and PRB allocation to prioritize critical video traffic while optimizing energy efficiency across slices. This demo showcases the LTM operating outside of SoftBank infrastructure, underscoring the potential of this technology when realized across the whole industry. The demo can be found at both the AI-RAN Alliance booth (Hall 2 Booth #2E45) and the Keysight booth (Hall 5 - 5F41), and through this link.

4. From Innovation to Commercialization

Technical milestones and demos are important, but they only gain true meaning when they deliver benefits to society. To help bridge the gap between innovation and realization, SoftBank is spearheading the new Commercialization Task Group within the AI-RAN Alliance. Having recently ratified the charter, we are now focused on building consensus around a pragmatic commercial blueprint and value chain.

It is a necessary step to ensure our collective work moves beyond the lab and into the real world. Rough consensus got us here; the running code—and a clear path to commercialization—will deliver on the promise of AI-RAN.

See you in Barcelona!

About the Author: Mauro Goncalves Filho
In February 2025, Mauro Goncalves Filho was appointed SVP, AI-RAN of SB Telecom America Corp., a wholly owned subsidiary of SoftBank Corp. He has over 25 years of experience in management, strategic planning, and business development in the technology, media, and telecommunications (TMT) sector.
He previously served as Senior Director at Google X and Chief Strategy Officer at Loon, an Alphabet company. He also held management roles at Telefonica Vivo and Gradus Consultants, leading global expansion efforts. Currently, he serves on the boards of Altave (computer vision) and Onovolab (innovation ecosystem), and is active as a startup mentor at The Alchemist Accelerator and a venture partner at Niu Ventures.