The Invisible Cost of AI Usage: How Our Prompts Are Destroying Communities

Many high school students use artificial intelligence (AI) for convenience in studying, researching, and even for life advice, but we rarely think about the environmental effects of such large systems. When uncovered, the footprint of AI systems is hard to ignore. From water usage to sound pollution, the growth of AI is coming at the cost of the environment and our health. 

Water Usage

AI systems rely on huge amounts of water to cool down their processors. AI data centers are powered by stacks of servers that handle intense workloads. These servers include many high-performance graphics processing units (GPUs) that run constantly, converting enormous amounts of electricity into heat. To prevent overheating and GPU damage, water is used as a cooling agent. In evaporative cooling, water flows through pipes adjacent to hot computer chips and is then pumped into a cooling tower, where the water is sprayed over a plastic mesh and spread into a thin layer. About 80 percent of water used by data centers evaporates, meaning that it doesn’t return to local waterways. The remaining unevaporated water is collected in a basin and reused. This remaining water cannot enter back into the local water system because data centers often add chemicals to water used for cooling to prevent rust and bacterial growth, making it unfit for human consumption. These systems can only use freshwater because saltwater is corrosive. The ions in saltwater steal electrons from metal, which oxidizes the metal in a process commonly known as rusting. The ions continue to gradually break down the metal until it collapses, a process commonly referred to as corrosion. Pure freshwater has low amounts of ions and does not exhibit this corrosive effect. 

To put all this water usage by AI systems into perspective, each 100 word AI prompt uses approximately one bottle of water. When a person pastes an essay into an AI chatbot for editing, they consume more than 20 percent of the water they would typically drink in a day within a few seconds. By 2028, Forbes predicts that AI related data centers may consume up to 33 billion gallons of water annually in the U.S. alone, which is enough to support approximately 360,000 households. This amount of water usage can easily be the tipping point for communities facing a near-water crisis. Many of the communities are rural or low-income and minority communities without enough funding to build reliable water infrastructure, meaning that when an AI data center causes a water shortage, affected communities likely won’t have the resources to alleviate it.

AI water usage is an issue because it contributes to water shortages, or a lack of freshwater availability to meet local water consumption needs. Water shortages exist because only one percent of Earth’s water is freshwater that is accessible to humans, and when more water is consumed than renewed, our water infrastructure cannot keep up. These shortages affect areas in dry climates and areas with poor water infrastructure more than others. About 40 percent of large data centers in the U.S. are found in the most water-scarce regions of the country. Two thirds of data centers built since 2022 have been in water-stressed regions with hot and dry climates, like Arizona. AI companies value dry climates for data centers because dry climates allow the water in their cooling systems to evaporate easier. In addition, dry states like Arizona and Texas offer huge tax breaks to data centers. 

When AI data centers drain water pipes, nearby residents are the ones that feel the burden. The Potomac River Basin in Virginia has one of the world’s largest concentrations of data centers. A new study by the Interstate Commission on the Potomac River Basin reports that data centers can account for 12 percent of water usage in the summer, when water demand from local residents is at its highest and water supply at its lowest. This could push the region into a water shortage when river levels drop lower than usual, jeopardizing the water supplies of critical federal service buildings in Washington D.C. 

Sound Pollution

AI data centers also produce significant sound pollution. Cooling systems also employ large, industrial fans to help cool processors. These fanning systems emit an incessant, bone-jarring hum, making the surrounding areas nearly unlivable for residents. Google began building data centers in Prince William County, Virginia, in 2021. Before the site completed construction, the tonal noise emitted from the data center disrupted local residents’ sleep and daily lives. In the town of Dulles in Virginia, three Amazon data centers sit just 200 meters away from a Regal Theater parking lot, right in the middle of community infrastructure. The sound from the three data centers completely masks everything in the parking lot, making it impossible to hear daily, regular birdsong. Constant environmental noise exposure has been shown to lead to a variety of health problems, including arterial hypertension, myocardial infarction, and stroke.

Energy Usage 

To process so much data, AI data centers require massive amounts of electricity. A single data center can consume as much power as 100,000 homes, with larger centers expected to consume up to twenty times more electricity to power more processor units. The Lawrence Berkeley National Laboratory reports that data centers could represent up to 12 percent of total U.S. electricity consumption by 2028. It is important to note that this figure includes all U.S. data centers, not just AI data centers.

This electricity demand is putting severe strain on the U.S. energy grid. Because most of the U.S. energy grid still relies on fossil fuels to generate electricity, a transition to renewable energy would not be able to sustain the increasing electricity demand created by AI data centers. Thus, AI companies are increasingly backing the fossil fuel industry to ensure that data centers are able to increase energy consumption without facing issues with renewable energy transitions.

Data center energy consumption has put a strain on local utility systems as well. Electricity prices are skyrocketing more than two times faster than inflation, partly because of data centers that pressure local energy grids. When one in three American households struggle to pay energy bills, rising energy costs makes it difficult for families to make ends meet.

Further Problems 

To make matters worse, AI companies release little to no details about their data center activities. When local residents of Prince William County requested information on the Google data center from officials, the officials could not release substantial information because Google had them sign legally binding non-disclosure agreements. Large AI companies often hide behind “startup names” to acquire contracts without making their identities clear. In the U.S., as long as a Limited Liability Company (LLC) is registered with the Secretary of State and pays taxes, it is a valid legal entity and does not have to disclose its parent company. Many large AI corporations form LLCs to pretend to be small startups. In New Albany, Indiana, Meta acquired a contract for construction under the name “Sidecat LLC,” making it hard for the public to trace records back to Meta. AI companies are also not required to release much data, so the extent of the environmental effects of AI are still largely produced from estimates. 

Possible Solutions 

While it’s impossible to demand a stop to AI development, the federal government should mandate AI companies to switch to more environmentally friendly and less water-intensive methods for cooling, such as closed-loop cooling systems, air cooling, or immersion cooling. These methods can dramatically reduce water usage while still being incredibly effective for cooling processor units. However, methods like air cooling have their downsides—air can become very hot in the summer, and in order to operate air cooling systems, cold water is still necessary to cool the air. Immersion cooling systems are incredibly expensive, closed-loop cooling systems require much more electricity than evaporative cooling systems, and because electricity generally costs more than water, companies tend to choose the more cost-effective cooling method. 

In addition, the U.S. currently doesn’t have any comprehensive federal AI regulations. Data center placement needs to be regulated on the state and federal level to ensure that data centers are placed in areas with adequate water availability and in locations far away from residential communities. AI companies should be required to release comprehensive system data periodically to ensure public knowledge and allow legislators to pass effective legislation limiting the harmful effects of AI. If the U.S. federal government passes legislation similar to the 2024 EU AI Act, it will be able to ban AI systems that are clearly detrimental to society. AI development will inevitably continue, but implementing regulations makes it safer for everyone.