{"id":373501,"date":"2026-07-10T12:06:31","date_gmt":"2026-07-10T12:06:31","guid":{"rendered":"https:\/\/wolfscientific.com\/?p=373501"},"modified":"2026-07-10T12:06:31","modified_gmt":"2026-07-10T12:06:31","slug":"ai-email-creation-requires-substantial-water-resources-expected-to-equal-fifty-percent-of-the-uks-yearly-water-consumption-by-2027-due-to-drought-worries","status":"publish","type":"post","link":"https:\/\/wolfscientific.com\/?p=373501","title":{"rendered":"“AI Email Creation Requires Substantial Water Resources, Expected to Equal Fifty Percent of the UK’s Yearly Water Consumption by 2027 Due to Drought Worries.”"},"content":{"rendered":"
<\/div>\n

The disconcerting aspect of AI’s water usage isn’t that a single chatbot response depletes a reservoir. It doesn’t. The troubling part is that a minor, concealed expense escalates when it is repeated across millions or billions of queries, subsequently contributing to the water required to construct and operate the data centers supporting them.<\/p>\n

A 2024 Washington Post study<\/a>, conducted collaboratively with researchers from the University of California, Riverside, estimated the water and electricity utilized by ChatGPT employing GPT-4 to generate an average 100-word email at a typical American data center. The Post represented the outcome as approximately a bottle of water per email.<\/p>\n

This figure is informative but only if interpreted with care. It\u2019s not a constant measurement linked to each ChatGPT response. Water usage varies based on the model, the data center, the weather conditions, the cooling system, the electricity source, and the accounting method. The same prompt can have a different ecological impact depending on where and when it is processed.<\/p>\n

The extensive scientific research behind many of these estimates is credited to Pengfei Li, Jianyi Yang, Mohammad A. Islam, and Shaolei Ren. Their paper, Making AI Less \u201cThirsty\u201d<\/em><\/a>, was initially posted on arXiv in 2023 and later accepted by Communications of the ACM<\/em>. It estimated that a model like GPT-3 could utilize approximately 500 milliliters of water for about 10 to 50 medium-length responses, depending on the location and timing of its use.<\/p>\n

This variability is significant. It highlights the difference between considering water use as a uniform fact and perceiving it as an infrastructural issue influenced by geography.<\/p>\n

What qualifies as AI water use?<\/h2>\n

Data centers consume water in two primary manners. The first is direct water usage, typically for cooling purposes. Servers produce heat. In many facilities, evaporative cooling systems utilize water to dissipate that heat. Some of this water is consumed as it evaporates into the atmosphere.<\/p>\n

The second is indirect water usage. Generating electricity may necessitate water, particularly in thermal power plants that utilize water for steam cycles or cooling. A data center that appears water-efficient on site could still be connected to water consumption elsewhere via the electrical grid that powers it.<\/p>\n

This is why the terminology in these studies is precise. Water withdrawal refers to water taken from a source, such as a river, aquifer, or municipal supply. Water consumption usually indicates water removed from immediate reuse, often through evaporation. Both aspects are important, but they are not identical figures.<\/p>\n

Li and colleagues forecasted that global AI demand could represent 4.2 to 6.6 billion cubic meters of water withdrawal in 2027, based on the scenarios they analyzed. In the abstract, they compared this range to the total annual water withdrawal of several Denmarks or about half of the United Kingdom.<\/p>\n

This is a model-based estimation, not a measurement of all AI systems currently. It should not be regarded as an inevitable outcome. However, it provides perspective to a challenge that is otherwise easy to overlook within the polished interface of a chat window.<\/p>\n

Why estimates vary<\/h2>\n

Public estimates of AI water usage vary significantly. In 2025, a Google-authored arXiv paper, Measuring the environmental impact of delivering AI at Google Scale<\/em><\/a>, indicated that the median Gemini Apps text prompt consumed 0.24 watt-hours of energy and 0.26 milliliters of water according to Google\u2019s accounting framework. The authors noted this was roughly equivalent to five drops of water.<\/p>\n

This figure is considerably smaller than the bottle-scale estimates frequently referenced for ChatGPT. The disparity does not necessarily imply that one number is true and the other is false. The studies measure different aspects. They encompass various systems, time frames, assumptions, workloads, and boundaries.<\/p>\n

Google\u2019s paper focused on servicing Gemini text prompts within Google’s production environment. The Li and Ren research attempted to estimate a wider AI water footprint, including both direct and external water use. The Washington Post calculation zeroed in on GPT-4 generating a 100-word email at a standard U.S. data center.<\/p>\n

The takeaway is not that AI consumes either five drops or one bottle. The takeaway is that without clear, comparable reporting, the public is left to juxtapose dissimilar figures.<\/p>\n

The local issue<\/h2>\n

Water is<\/p>\n","protected":false},"excerpt":{"rendered":"

The disconcerting aspect of AI’s water usage isn’t that a single chatbot response depletes a reservoir. It doesn’t. The troubling part is that a minor, concealed expense escalates when it is repeated across millions or billions of queries, subsequently contributing to the water required to construct and operate the data centers supporting them. A 2024 […]<\/p>\n","protected":false},"author":2,"featured_media":373502,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"Default","format":"standard","meta":{"footnotes":""},"categories":[1],"tags":[179],"class_list":["post-373501","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-uncategorized","tag-source-scienceblog-com"],"_links":{"self":[{"href":"https:\/\/wolfscientific.com\/index.php?rest_route=\/wp\/v2\/posts\/373501","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/wolfscientific.com\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/wolfscientific.com\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/wolfscientific.com\/index.php?rest_route=\/wp\/v2\/users\/2"}],"replies":[{"embeddable":true,"href":"https:\/\/wolfscientific.com\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=373501"}],"version-history":[{"count":0,"href":"https:\/\/wolfscientific.com\/index.php?rest_route=\/wp\/v2\/posts\/373501\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/wolfscientific.com\/index.php?rest_route=\/wp\/v2\/media\/373502"}],"wp:attachment":[{"href":"https:\/\/wolfscientific.com\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=373501"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/wolfscientific.com\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=373501"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/wolfscientific.com\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=373501"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}