International Journal of Science, Engineering and TechnologyAuthors: Emmanuel Uzochukwu Mordi, Chibuzo Joseph Attah, Chiamaka Sandra Ezugwu, Christian Onyemaechi Asogwa, David Chinonso Anih, Samuel Daniel Ejiga, Omobolanle Omotayo Solaja
Abstract: Edge AI is reshaping where and how artificial intelligence runs, promising lower latency and reduced network use by moving computation from centralized cloud data centers to distributed devices. This systematic review examines whether that promise translates into net environmental benefit or whether a rebound effect emerges that shifts and potentially amplifies overall energy and lifecycle impacts. We synthesized 40 qualitative studies and 38 quantitative analyses published between 2016 and 2025, comparing energy per inference, carbon intensity, lifecycle burdens, network scaling, and socio technical outcomes across cloud, edge, and hybrid deployments. Our findings show a nuanced landscape: for lightweight inference tasks, localized execution on specialized edge accelerators often reduces per inference energy and transmission emissions, while cloud processing retains advantages for heavy or batch workloads due to economies of scale and optimized cooling. However, cumulative effects matter. Millions of short lived or redundant edge devices can yield substantial aggregated energy demand, resource depletion, and e waste that offset per device gains. Hybrid strategies that combine edge preprocessing with cloud consolidation frequently offer the best tradeoffs, improving efficiency ratios and lowering carbon intensity when workloads are partitioned intelligently. We also document important non-technical tradeoffs. Edge deployment strengthens data privacy and responsiveness but increases attack surface and exacerbates unequal access where infrastructure or device availability is limited. Network effects are critical: pure edge scaling can surge local network load and create bottlenecks, while cloud centric models concentrate backbone traffic but remain easier to optimize at scale. Policy and governance emerge as decisive enablers: standardized energy reporting, lifecycle transparency, and harmonized ethical and sustainability criteria can steer deployments toward net benefit. We identify methodological heterogeneity across life cycle boundaries and geographic energy mixes as sources of uncertainty and recommend clearer reporting standards to improve comparability. In conclusion, Edge AI is neither inherently greener nor intrinsically harmful.
