Peer-reviewed literature establishes that generative AI systems like ChatGPT and their underlying data centers consume significant amounts of freshwater, primarily for cooling purposes and associated electrical power generation.
Artificial intelligence (AI) computing and data centers consume large amounts 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 combines nation-level weather and electricity generation data to estimate water usage efficiency for data centers in 41 African countries across five different climate regions. We also use our dataset to evaluate and estimate the water consumption of inference on two large language models (i.e., Llama-3-70B and GPT-4) in 11 selected African countries. Our estimates suggest that writing a 10-page report using Llama-3-70B could consume as much as 0.6 liters of water, while the water consumption by GPT-4 for the same task may go up to about 53 liters. For writing a medium-length email of 120-200 words, Llama-3-70B and GPT-4 could consume about 0.12 liters and 2.6 liters of water, respectively. Interestingly, given the same AI model, 8 out of the 11 selected African countries consume less water than the global average, mainly because of lower water intensities for electricity generation. However, water consumption can be substantially higher in some African countries with a steppe climate than the U.S. and global averages, prompting more attention when deploying AI computing in these countries. Our dataset is publicly available.1
The rapid rise of generative AI systems, such as ChatGPT, has sparked important conversations around their hidden environmental costs—particularly in terms of electricity usage, carbon emissions, and water consumption. These systems, while revolutionizing fields like education, customer support, and creative production, require significant computational infrastructure that exerts pressure on global energy and ecological resources. For instance, a single AI query can consume nearly ten times more electricity than a basic internet search. Moreover, large-scale data centers, often located in water-stressed regions, rely heavily on intensive cooling systems, further amplifying their environmental footprint. The purpose of this research is to critically examine and quantify these environmental impacts, and to propose practical, evidence-based strategies that enable the continued advancement of AI without compromising ecological sustainability. Drawing from a wide range of empirical studies and industry reports, this paper compares the energy and water use of generative AI to traditional digital technologies and evaluates their lifecycle emissions. It also explores promising mitigation solutions, including model compression, renewable energy integration, and next-generation cooling technologies. Ultimately, the study advocates for a more transparent and accountable approach to AI development—where performance and innovation are pursued in tandem with environmental responsibility. By translating complex technical findings into actionable strategies, this research offers a roadmap for developers, researchers, and policymakers seeking to build a more sustainable future for artificial intelligence
Data centres require energy to power their computing equipment as well as to maintain proper environmental conditions through their extensive cooling systems. Data centres are a key part of digital infrastructure, but use a lot of energy, especially for cooling and computing. This paper explores energy trends and reviews solutions like CRAC systems, chilled water cooling, and free cooling. It also discusses energy efficiency using measures like Power Usage Effectiveness (PUE) and shows real examples of improvement. The goal is to help design greener, more efficient data centres. The research investigates the main elements that determine energy consumption in data centres. The paper examines two emerging technologies and strategies for decreasing energy usage, which include CRAC systems and free cooling, and chilled water systems. It is estimated that annually in the U.S, the data centres consume roughly 50% of electricity, mainly by the equipment. The cooling needs for Heating, Ventilation, and Air Conditioning (HVAC) are estimated to be up to 40% using computer room air-conditioners to cool down the equipment, such as servers, and other IT equipment in the data centres. This paper discusses the high energy needs of data centres and also reduces energy use by making systems as efficient as possible. Providing sustainable data centres is the energy goal so as to maximise energy from renewable systems. Data centres need a comprehensive strategy that combines operational excellence with environmental responsibility and financial sustainability to enhance their energy efficiency. Future research must create regionally adaptable solutions that reduce data centre environmental impact because digital performance expectations will continue to grow.
Generative AI technologies, such as ChatGPT, have measurable environmental impacts primarily from electricity consumption and freshwater usage at data centers. An individual AI query emits roughly 4.3 grams of CO₂ and uses around 10 milliliters of freshwater. In comparison to common everyday tasks, AI's carbon footprint is small, significantly lower than driving or showering but higher than simple digital activities like web browsing.At scale, however, generative AI contributes notably to global energy and water demands. AI data centers consume tens of terawatt-hours (TWh) of electricity annually, with major companies reporting rapid increases in energy use due to expanding AI capabilities. Freshwater use at these centers is similarly substantial, reaching billions of gallons annually.Though concerns regarding AI's environmental impacts are supported by data, significant mitigation is achievable through energy-efficient designs, renewable energy sourcing, and enhanced operational transparency.
Everything we examined (5) — 4 independent sources
This check searched the claim as stated. It did not run a separate search for evidence against it.