Autonomous Agentic AI-RAN Demoed at MWC
Northeastern University, SoftBank, Keysight, and zTouch Networks demonstrated an agentic, AI-native Radio Access Network (RAN) at MWC Barcelona. The system is powered by a Large Telecom Model (LTM) and is designed to be intent-driven and autonomous. The demonstration is the result of a long-standing collaboration between the industry and academic partners.
- The "Large Telecom Model" (LTM) is a generative AI foundation model developed by SoftBank, trained on their extensive network data, including base station configurations and performance data. In trials in Tokyo, it achieved over 90% accuracy in predicting optimal base station configurations, reducing the time required from days to minutes. - Northeastern University's Institute for the Wireless Internet of Things (WIoT) is a key academic partner and leads the Platforms for Advanced Wireless Research (PAWR), a $100 million public-private partnership. The institute holds 68 patents and collaborates with over 30 industry partners, focusing on AI for wireless systems. - zTouch Networks is a startup spun out of Northeastern University that focuses on automating the network intelligence lifecycle in Open RAN systems. Their zTouch.OS provides an O-RAN-compliant automation platform that integrates RAN Intelligent Controllers (RICs) to manage the coexistence of RAN and AI workloads. - The AI-RAN Alliance, of which SoftBank and Keysight are members, is a collaborative initiative to integrate AI into cellular technology to improve RAN performance. SoftBank is contributing to working groups focused on "AI-for-RAN" (using AI to optimize the network) and "AI-and-RAN" (running AI applications on network hardware). - The demonstration's goal is to enable the RAN to function as a distributed AI infrastructure, allowing AI and RAN workloads to coexist on the same hardware. This approach, termed "AI-RAN," aims to support low-latency edge AI services like autonomous driving and remote robotic surgery. - Keysight provides the testing and validation solutions for the AI-RAN ecosystem, including digital twins and high-fidelity emulation to train AI/ML models and benchmark performance before deployment. Their collaboration with NVIDIA and Samsung has shown up to a 30% better cell edge throughput by using an AI model for channel estimation. - The concept of "agentic AI" is a major theme at MWC Barcelona 2026, representing a shift from generative AI for customer-facing functions to autonomous agents managing network operations based on high-level business intent. This is seen as a way for telecom companies to move beyond Level 2 of the TM Forum's Autonomous Networks framework. - SoftBank plans a commercial release of its AI-RAN product for global deployment in 2026, built on NVIDIA Grace Blackwell superchips. They are also collaborating with Yaskawa Electric to develop "physical AI" robots for office environments that run on an AI-RAN-based Multi-access Edge Computing (MEC) platform.
