Cooling the Cloud - A Study on Innovative Approaches to Cooling AI Data Centers

Data centers that provide computational power for Artificial Intelligence struggle with resource intensive practices. The cooling systems used for large-scale and commercial AI are mostly liquid-based and utilize large portions of water daily. When paired against a growing demand for water provided by a rising population and inequity crisis, the deficiency causes a substantial issue for general water accessibility. The following paper presents examples and explanations of simple cooling techniques (fan, heat sink, heat pump, and vapor chamber), primarily limited to those used in PCs as well as a comparative literature-based evaluation framework to modify these techniques to be more widely applicable. The information presented here is not a result of a concrete experiment, instead a comparative evaluation framework and process. Although it is true that up-scalling PC-systems for industrial use is novel, the following paper uses pre-existing knowledge to present a process through which such a breakthrough could occur by interpreting and novelizing existing works. The ideal goal of this paper is to present a research plan based on existing data that would yield a product capable of reducing the environmental footprint of AI data centers. While the data in this paper is not original (all sources cited), it is a compilation and summation of preexisting data in a novel and condensed format that presents a new use for it.