Next to animal agriculture, all the world’s data centers use practically nothing.

Our World in Data · Poore & Nemecek (2018)

Farm animals use about 200 times more freshwater than AI is projected to use

Wisser and colleagues (2024) estimated that animal agri-food systems consume about 228 km³ of freshwater a year[1], with feed production accounting for more than 90% of those withdrawals[1]. Morgan Stanley’s September 2025 analysis projected water use linked to AI data centers at about 1,068 billion litres a year by 2028, roughly 1 km³[2]. 228 divided by 1.07 is about 213. Lawrence Berkeley National Laboratory’s 2024 report on U.S. data centers lands in the same neighborhood at national scale[3]. Farm animals still dwarf the machines even after you add the water used to generate the electricity[3]. CBS News, using LBNL and University of Twente numbers, noted that Americans use more water to grow food, especially meat, than all U.S. data centers use combined[3][4].

OpenAI query water · University of Michigan Beyond Burger LCA

One cow-flesh burger is about 1.4 million AI prompts

Going by OpenAI’s figure for the water in a single prompt[1] and the University of Michigan study of how much water a quarter-pound non-vegan patty takes[3], you would need about 1.4 million prompts to equal the water in one cow-flesh burger[1][3].

OpenAI put an average ChatGPT query at about 0.32 millilitres[1]. Michigan put that patty at 437.7 litres[3]. Divide one by the other and you get about 1.4 million[1][3].

Depending on the source and the calculation, the figure runs from about 8,800 to 5.5 million prompts[1][2][4]. The low end, about 8,800, comes from Li et al. 2023[2], which put a GPT-3 reply at about 50 millilitres[2]. The high end, about 5.5 million, is OpenAI’s 0.32 millilitres[1] against Mekonnen and Hoekstra, who put a quarter pound of beef near 1,750 litres (about 460 gallons)[4].

The 437.7 litres figure counts only blue water, meaning water from rivers, lakes, and wells. It leaves out green water, the rain that falls on pasture and feed, so the full amount is higher. Many people wrongly assume these water figures are based on green water.

You would have to live in the chatbot for years to drink a burger. The feed-and-trough loop runs every day, for tens of billions of animals, whether or not anyone opens a chat window.

Why we don’t use the bottle-per-email number

A 2024 Washington Post article said ChatGPT used about 519 millilitres of water to write a 100-word email, a little more than a bottle, and the number spread widely[1].

It modeled GPT-4 writing a full email, which is neither an average prompt nor the figure OpenAI later published (about 0.32 millilitres)[3]. The paper underneath those headlines, Li et al. 2023, put a medium-length GPT-3 reply at about 10 to 50 millilitres[2], not a bottle each[2]. Headlines flattened that into a bottle per email.

OpenAI’s published average[3] and the Michigan burger study give about 1.4 million prompts per burger[3], and other studies set the range around it[2].

University of Michigan LCA · cited by Beyond Meat, A Burger With Benefits

Heller & Keoleian, University of Michigan, CSS18-10

Vegan burgers already use about 99% less water than beef

The same Michigan study, the one Beyond Meat cites, compared a ¼ lb Beyond Burger with a ¼ lb U.S. beef burger[1][2]. The plant patty used 99% less water, 93% less land, 90% fewer greenhouse gases, and 46% less energy[1][2]. Table ES1 reports scarcity-weighted water: 1.1 liter-eq for Beyond versus 218.4 liter-eq for beef[2]. Tables 7 and 8 report the actual consumptive volume, 3.27 litres for Beyond versus 437.7 litres for beef, which is still more than 99% less[2]. 437.7 litres is about 116 gallons. Once rain on pasture and feed is counted, Mekonnen and Hoekstra put beef near 15,400 litres per kilogram globally, about 460 gallons for a quarter pound[4]. You can keep your burger and still stop paying a cow’s water bill, because the vegan version is already on the shelf.

Our World in Data · How is the world’s land used?

Farm animals take up continents; data centers take up industrial parks

According to Our World in Data, drawing on FAO and on Poore and Nemecek, about 44% of habitable land is already farmed, about 48 million km²[1][2]. Grazing and feed for farm animals take about 80% of that farmland[1][2], an area on the order of the Americas. A worldwide vegan diet would shrink farmland from about 4.1 billion hectares to about 1 billion[1][2]. Industry tallies put the world near 10,000 data centers. Give every one of those sites 200 acres and the total is about 8,000 km², a fraction of one percent of the land used for farm animals. Some new halls are being poured on former fields, but that is a local planning fight, not a global land budget to set beside cows.

Our World in Data · Poore & Nemecek (2018)

Farm animals’ climate footprint is about 40 times all data-center emissions

The IEA put electricity use by the world’s data centers at about 415 TWh in 2024[1], about 1.5% of global electricity[1], and the CO₂ from that power at about 180 million tonnes[1], on the order of 0.5% of world CO₂[1]. Its 2035 base case is about 300 million tonnes, still under 1.5% of energy-sector emissions. FAO’s assessment of farm animals is in a different league: about 14.5% of all human greenhouse gases[3], or about 7.1 billion tonnes of CO₂-equivalent[3], mostly from methane, manure, and the land cleared to grow feed[3]. 7.1 billion divided by 180 million is about 40[1][3]. Poore and Nemecek put the whole food system at 26% of global GHGs[4], and Xu and colleagues found animal foods, including feed, account for about 57% of food-system emissions[5]. Data centers are a real and growing electricity load, but breeding tens of billions of animals is a far bigger climate problem[1].

Our World in Data / FAO slaughter statistics

Animal agriculture kills about 88 billion land animals a year; AI has no slaughter line

Faunalytics, working from FAO slaughter data, counts about 87.9 billion land animals killed for food in 2024[1], about 241 million a day[1], and chickens alone were about 78.5 billion[1]. Add farmed finfish (about 130 billion a year) and a wild catch estimated at 1.1 to 2.2 trillion fish, and the food system is the largest killing on the planet. AI does not raise, crate, or stun animals for output. Construction and power lines can harm wildlife the way any factory can, but that incidental harm is a different category from breeding tens of billions of animals in order to kill them. Ask which industry kills the most animals and the answer is animal agriculture.

Local water stress is real, but it doesn’t change the global ranking

A hyperscale hall can pull millions of gallons a day, and more than 40% of U.S. data centers sit in high water-stress regions. Those facts matter to the town next door, but they do not make AI larger than farm animals. Farm animals push a share of the world’s rivers past freshwater limits, and most of that water is feed irrigation spread across basins. Both things are true: don’t put a cooling tower on a failing aquifer, and don’t pretend a chat window is the main drain. The main drain is the feedlot.