Ekansh Agrawal | Shiv Nadar School, Noida, India |
Abstract
This paper presents a comprehensive review of the energy consumption by Artificial Intelligence technologies and consequently their impact on the environment. Datacenter energy consumption in 2021 was 220–320 TWh (Terawatt-hours), ~ 0.9–1.3% of global final electricity demand; data transmission networks consumed 260–340 TWh, ~ 1.1–1.4% of global electricity use. This paper reviews the energy consumption by different types of AI models, studying reasons behind why such models leave a significant carbon footprint. Various parameters and tools that are used to estimate the energy consumption are discussed. Furthermore, new solutions that scientists are exploring to reduce this energy consumption are analyzed. Lastly, upcoming developments in AI and various initiatives being taken for sustainable technological development are explored. This review sheds necessary input and insight into the heavy energy utilized by AI, making researchers aware of the need to develop and adopt techniques for energy-efficient AI.