Agentic AI-RAN Demonstrated at MWC Barcelona
Northeastern University, SoftBank, Keysight, and zTouch Networks demonstrated an autonomous, agentic AI-native Radio Access Network (RAN) at MWC Barcelona 2026. The system is powered by a Large Telecom Model (LTM), showcasing a new approach to managing wireless networks using intent-driven AI agents.
- The Large Telecom Model (LTM) from SoftBank is a generative AI trained on proprietary network data to automate operations; in tests, it reduced base station configuration time from days to minutes with over 90% accuracy. - This demonstration highlights a major industry split on AI acceleration hardware: while SoftBank and Nokia are leveraging NVIDIA GPUs, competitors like Ericsson are doubling down on proprietary, purpose-built silicon (ASICs) with integrated Neural Network Accelerators for their AI-RAN offerings. - The partner zTouch Networks is a 2022 startup spun out of Northeastern University's wireless research programs, focused on commercializing AI-driven orchestration and automation tools for Open RAN systems. - Keysight's role is providing the critical testing and validation workflow, using high-fidelity emulation and digital twins to benchmark the performance of AI models and RAN modules together before deployment. - The "agentic" approach moves beyond simple automation by using AI agents that can reason, analyze root causes, predict future issues, and autonomously execute corrective actions to optimize the network. - Northeastern University provides the foundational research and open-source ecosystem for this effort through its Open6G+AI initiative, which uses large-scale testbeds like the Colosseum network emulator to validate new AI-RAN concepts. - SoftBank's live AI-RAN field trials, which form the basis for this type of technology, have used a software-defined 5G stack running on NVIDIA GH200 Grace Hopper Superchips and BlueField-3 DPUs. - The AI-RAN market is projected by Dell'Oro to reach $10 billion by 2029, with SoftBank aiming to commercially release its own AI-RAN product for worldwide deployment in 2026.
