By Brittany Findley
At the core of artificial intelligence (AI) is data. Often referred to as the “training fuel” for AI, each AI form relies on vast quantities of assorted and comprehensive data for effective operation.[1] For instance, the latest version of ChatGPT, a text-based generative AI tool, leverages a dataset exceeding one petabyte, providing it with 1.8 trillion parameters to manipulate as it formulates its responses.[2]
These massive quantities of data are stored and managed at data centers. Data centers are spaces dedicated to housing the physical infrastructure necessary to store, process, and distribute data.[3] To meet the growing demand for AI and, accordingly, data, big technology companies such as Microsoft, Amazon Web Services, and Meta have increasingly allocated funds to AI development and data storage.[4] Microsoft, for instance, has over 300 data centers worldwide,[5] each designed to centralize the critical data that feed into their technological capabilities. In keeping up with AI advancements, Microsoft has dedicated $80 billion to build AI-focused datacenters to provide compute power to AI models and deploy them across the globe.[6]
How AI Quenches its Thirst
While our world rushes to expand AI, its environmental repercussions have raised alarm. Aside from its highly scrutinized carbon footprint, AI data centers are one of the top 10 freshwater-consuming industrial or commercial industries in the United States.[7]
AI’s total water consumption spans three main areas, or scopes. The first, scope-1, utilizes water as a cooling mechanism.[8] Servers within data centers convert their energy into immense amounts of heat.[9] To prevent overheating, water is either evaporated in a cooling tower by the servers’ heat or the water flows to a heat exchanger to directly absorb the heat.[10] The first method, known as evaporative cooling, uses anywhere from three to five million gallons of water per day per hyperscale data center.[11] Scope-2 considers the off-site water used to sustain electricity production.[12] The electricity required to power data centers comes from thermoelectric or hydroelectric power plants, both relying on water to cool their own facilities.[13] Finally, scope-3 usage pertains to AI supply chains including chip and server manufacturing.[14]
To put the water dependency of AI’s three scopes into perspective, consider OpenAI’s GPT-3. GPT-3 alone requires 16.9oz of water (approximately the size of one standard water bottle) for just 10 to 50 medium-length responses.[15] With over 400 million weekly active users, GPT-3 and its subsequent models underscore the imperative role that water plays to keep AI functioning.[16] By 2027, AI’s global annual water consumption is projected to hit between one and two trillion gallons.[17]
Implications
Polluted Water
Scope-1 water usage results in unusable polluted water. Over time, water used to cool data centers collects unwanted components including excess minerals, salt, and bacterial growth, increasing the risk of damage, such as corrosion, to the servers.[18] Because of this accumulation, the water used in cooling towers can only be recycled between three and ten times before it is replaced with fresh water.[19] This process occurs constantly where new water must be added to replace the water that was evaporated in the cooling process and the polluted water that was dispensed.[20]
Strain on Drought-Stricken Areas
AI data centers’ high demand for freshwater burdens local water supplies, especially in areas that already face water scarcity.[21] Watersheds in the Western United States show high levels of stress which is worsened by data centers.[22] Nationwide evaporative cooling only draws 20% of its water from the West, however the national water scarcity footprint (measured by the pressure exerted by consumptive water use from data centers on available freshwater) is predominantly influenced by Western states.[23] This exposes a disproportionate reliance on the West and, correspondingly, a reliance on an already scarce supply of water.[24] Concerned by the dependence on the West for water, the 2021 Vice Mayor of Mesa, Arizona, condemned the development of a $800 million data center that would necessitate 1.25 million gallons of water daily, further straining Mesa’s already limited water supply.[25]
Beyond the West, other communities have also expressed their distaste towards data centers being constructed. Since 2017, Goose Creek, South Carolina conservation groups have pushed back against Google’s request to supply a data center with 1.5 million gallons of water from an already depleted aquifer.[26] Similarly, residents of Peculiar, Missouri have fought against data center developments out of fear that servers would disrupt water flow to the city’s residents.[27]
Solutions
Citing water as a “precious resource,” various big technology companies operating data centers have pledged to become water positive within the upcoming years.[28] Amazon Web Services, with 44 data centers spread across North America, strives to replenish more water than their data centers utilize by 2030.[29] To accomplish this, their solution is to rely on more sustainable water sources, reuse as much water as possible, and return 7 billion liters of water annually to local communities.[30] Google, in a similar fashion, announced their goal to be water positive by 2030 by reducing freshwater consumption in exchange for wastewater and seawater.[31] While these efforts are a step in the right direction, the most crucial component to reducing AI’s water footprint is transparency. Less than one-third of data center operators currently track water use,[32] resulting in inadequate data to drive efforts to enhance water sustainability. Presently, AI model cards which delineate AI training methods include carbon emission data but exclude data on water use across all three scopes.[33] To achieve a complete understanding of AI’s water dependency, datacenters need to start producing this missing information. Only then can a fully accurate depiction of AI’s water footprint be drawn to understand the extent of its consequences and develop solutions that both foster AI development and protect the communities that data centers threaten to dr
[1] Rohit Sehgal, AI Needs Data More Than Data Needs AI, Forbes (Oct. 5, 2023, 8:15 AM), https://www.forbes.com/councils/forbestechcouncil/2023/10/05/ai-needs-data-more-than-data-needs-ai/.
[2] Erika Balla, Here’s How Much Data Gets Used By Generative AI Tools For Each Request, Data Sci. Cent. (Nov. 28, 2023, 2:23 PM), https://www.datasciencecentral.com/heres-how-much-data-gets-used-by-generative-ai-tools-for-each-request/.
[3] What is a Data Center?, Palo Alto Networks, https://www.paloaltonetworks.com/cyberpedia/what-is-a-data-center (last visited Feb. 18, 2025).
[4] Anna Tong, Aditya Soni & Deborah Mary Sophia, Big Tech’s AI splurge worries investors about returns, Reuters (Nov. 1, 2024, 7:24 AM), https://www.reuters.com/technology/artificial-intelligence/meta-microsoft-lift-ai-spending-worrying-wall-street-ahead-amazon-results-2024-10-31/.
[5] Microsoft datacenters, Microsoft, https://datacenters.microsoft.com/#:~:text=The%20Microsoft%20network%20connects%20more,global%20edge%20points%20of%20presence (last visited Feb. 18, 2025).
[6] Brad Smith, The Golden Opportunity for American AI, Microsoft (Jan. 3, 2025), https://blogs.microsoft.com/on-the-issues/2025/01/03/the-golden-opportunity-for-american-ai/.
[7] Md Abu Bakar Siddik, Arman Shehabi & Landon Marston, The environmental footprint of data centers in the
United States, 16 Env’t. Rsch. Letters 1, 6 (2021).
[8] Ana Pinheiro Privette, AI’s Challenging Waters, Univ. Ill. Urbana-Champaign Grainger Coll. Eng’g (Oct. 11, 2024), https://cee.illinois.edu/news/AIs-Challenging-Waters; Pengfei Li, Jianyi Yang, Mohammad A. Islam & Shaolei Ren, Making AI Less “Thirsty”: Uncovering and Addressing the Secret Water Footprint of AI Models, 1, 3 (Jan. 15, 2025).
[9] Pengfei, Yang, Islam & Ren, supra note 8.
[10] Id.
[11] Patrick K. Lin, The Cost of Teaching a Machine: Lighting the Way for a Climate-Aware Policy Framework That Addresses Artificial Intelligence’s Carbon Footprint Problem,34 Fordham Env’t L. Rev. 1, 16 (2023).
[12] Privette, supra note 8.
[13] Id.
[14] Privette, supra note 8; Pengfei, Yang, Islam & Ren, supra note 8.
[15] Pengfei, Yang, Islam & Ren, supra note 8.
[16] @bradlightcap, X (Feb. 20, 2025, 9:19 AM), https://x.com/bradlightcap/status/1892579908179882057; Kate Rooney, OpenAI tops 400 million users despite DeepSeek’s emergence, CNBC (Feb, 20, 2025, 10:57 AM), https://www.cnbc.com/2025/02/20/openai-tops-400-million-users-despite-deepseeks-emergence.html.
[17] Privette, supra note 8.
[18] Siddik, Shehabi & Marston, supra note 7 at 5; Pengfei, Yang, Islam & Ren, supra note 8.
[19] Pengfei, Yang, Islam & Ren, supra note 8.
[20] Id.
[21] Alesia Zhuk, Artificial Intelligence Impact on the Environment: Hidden Ecological Costs and Ethical-Legal Issues, 1 J. Digit. Techs. & L. 932, 944 (2023).
[22] Siddik, Shehabi & Marston, supra note 7 at 7.
[23] Id.
[24] Id.
[25] Olivia Solon, Drought-stricken communities push back against data centers, NBC News (June 19, 2021, 6:00 AM), https://www.nbcnews.com/tech/internet/drought-stricken-communities-push-back-against-data-centers-n1271344.
[26] Id.
[27] Caroline O’Donovan, Fighting back against data centers, one small town at a time, Wash. Post (Oct. 5, 2024), https://www.washingtonpost.com/technology/2024/10/05/data-center-protest-community-resistance/.
[28] Water stewardship, Amazon, https://sustainability.aboutamazon.com/natural-resources/water#:~:text=AWS%20water%20positive%20by%202030,in%20our%20data%20center%20operations (last visited Feb. 18, 2025).
[29] Id.; AWS Global Infrastructure, Amazon Web Servs. https://aws.amazon.com/about-aws/global-infrastructure/ (last visited Feb. 20, 2025).
[30] Id.
[31] Water stewardship, Google, https://sustainability.google/operating-sustainably/water-stewardship/ (last visited Feb. 18, 2025).
[32] Why circular water solutions are key to sustainable data centres, World Econ. F. (Nov. 7, 2024), https://www.weforum.org/stories/2024/11/circular-water-solutions-sustainable-data-centres/.
[33] Privette, supra note 8; Pengfei, Yang, Islam & Ren, supra note 8.
