Networking-Aware Energy Efficiency in Agentic AI Inference: A Survey

2026-04-09

AI Summary

This study addresses the escalating energy challenges in Agentic AI systems, where iterative reasoning and continuous data interaction lead to compounded computational and communication energy consumption, rendering traditional FLOPs-centric efficiency metrics inadequate. The work presents the first holistic energy accounting framework tailored to the perception–reasoning–action loop, introducing a unified taxonomy encompassing model simplification, computation control, input and attention optimization, and hardware-aware inference. It advocates for cross-layer co-design across models, wireless transmission, and edge resources, establishing a new paradigm of joint communication-computation optimization. The paper further outlines forward-looking research directions—including federated green learning, carbon-aware agents, and 6G-native Agentic AI—providing both theoretical foundations and a technical roadmap for the scalable, sustainable deployment of autonomous intelligent systems.

Abstract

The rapid emergence of Large Language Models (LLMs) has catalyzed Agentic artificial intelligence (AI), autonomous systems integrating perception, reasoning, and action into closed-loop pipelines for continuous adaptation. While unlocking transformative applications in mobile edge computing, autonomous systems, and next-generation wireless networks, this paradigm creates fundamental energy challenges through iterative inference and persistent data exchange. Unlike traditional AI where bottlenecks are computational Floating Point Operations (FLOPs), Agentic AI faces compounding computational and communication energy costs. In this survey, we propose an energy accounting framework identifying computational and communication costs across the Perception-Reasoning-Action cycle. We establish a unified taxonomy spanning model simplification, computation control, input and attention optimization, and hardware-aware inference. We explore cross-layer co-design strategies jointly optimizing model parameters, wireless transmissions, and edge resources. Finally, we identify open challenges of federated green learning, carbon-aware agency, 6th generation mobile communication (6G)-native Agentic AI, and self-sustaining systems, providing a roadmap for scalable autonomous intelligence.

Problem

• Agentic AI
• energy efficiency
• communication cost
• iterative inference
• mobile edge computing

Innovation

• Networking-Aware Energy Efficiency
• Agentic AI
• Cross-Layer Co-Design
• Perception-Reasoning-Action Cycle
• Hardware-Aware Inference