Agentic AI for Energy-Efficient Telecommunications: A Comparative Performance Analysis

Abstract

The integration of Agentic Artificial Intelligence (AI) into telecommunications networks presents a transformative opportunity for autonomous network management, yet the energy consumption implications of these systems remain insufficiently characterized. This article presents a comparative analysis of two leading agentic AI frameworks which includes the Strategist Agent architecture for Radio Access Network (RAN) management and the AGORA (Agentic Green Orchestration Architecture) framework with a focus on energy efficiency performance in telecommunications applications. Drawing on recent empirical studies and experimental evaluations from 2025–2026, this study examines how architectural design choices fundamentally determine energy consumption patterns. The analysis reveals four critical findings: (1) framework architecture drives energy consumption with up to 9.4× differences, significantly outweighing model size or task complexity; (2) current frameworks achieve near-zero task resolution rates (0-4%) with SLMs, wasting energy on unproductive reasoning loops; (3) memory access dominates computational energy (70%) with DRAM access costing 640 pJ/access, while token transmission dominates communication energy (40%); and (4) the optimal model-framework pairing depends on the specific task, with llama3.2:3b + LangChain achieving the highest performance with the lowest energy consumption. For telecommunications applications, AGORA's direct intent-to-tool translation achieves superior energy efficiency, while the Strategist Agent provides better reasoning depth. Projected net energy impact reveals 85% infrastructure savings against 15% agentic cost, yielding 70% net energy reduction in optimized RAN configurations. This study presents a framework efficiency metric (η = Task Resolution Rate / Energy Consumed) revealing η ≈ 0 for current SLM deployments, indicating urgent need for architectural redesign. The findings from this study provide actionable insights for deploying energy-efficient agentic AI, emphasizing that sustainability must be a first-class design principle in next-generation wireless networks, requiring memory-optimized inference, adaptive reasoning depth, and carbon-aware decision-making to achieve both intelligence and environmental sustainability.