A Water Efficiency Dataset for African Data Centers

AI computing and data centers consume a large amount of freshwater, both directly for cooling and indirectly for electricity generation. While most attention has been paid to developed countries such as the U.S., this paper presents the first-of-its-kind dataset that estimates water footprints for data centers in 11 African countries across five different climate regions. The dataset combines national-level weather and electricity generation data to estimate water usage efficiency in L/KWh. We use our dataset to evaluate and estimate the water consumption of inference on several popular large language models (including Llama-3-70B and GPT-4) in various countries across Africa. Our findings show that writing a 5-page report using Llama-3-70B could consume more than 0.5 liters of water while the water consumption by GPT-4 for the same task may go up to nearly 50 liters of water. Interestingly, given the same AI model, 8 African countries consume less water than the global average, mainly because of lower water intensities for electricity generation. However, the water consumption can be substantially higher in some countries with a rainforest climate than in the U.S. and global averages, prompting more attention when deploying AI computing in these countries that are already grappling with water scarcity challenges. Our dataset will be released at https://github.com/anonymous