AI: We Should Care a Whole Awful Lot

You’ve likely heard of it by now, if not used it most days without realizing it. There are many names for it: AI, Large Language Models, GenAI, ChatGPT, Claude, Grok, DeepSeek. Over the past few years, it has become capable of serving many purposes, such as writing letters, brainstorming home projects, making funny videos of animals jumping on trampolines, and even helping in writing our notes. To hear people talk about it, you would think it is the best thing since sliced bread. There are questions surrounding its use, however, that have not been answered in full—specifically, its utilization of resources and the ensuing costs involved. Is humanity drinking from a poisoned chalice in using AI so frequently? That’s what I want to explore here.

Before we talk about the costs of AI, let’s talk about how it works. For the uninitiated, when you utilize an AI model, it sends queries or prompts to a data center, which houses a host of interconnected servers, working in tandem to perform the tasks they are given. In FlOPs (Floating-point operations per second), these can scale in the quadrillions.

The racks that contain the servers and computers consume an amount of energy equivalent to three to six typical American homes—but take up about a dresser’s worth of space. The data centers that house these server racks are massive facilities. For reference, OpenAI’s Stargate campus in Abeline, Texas, which is slated for completion in mid-2026, will occupy as much space as Central Park in New York City. Other upcoming projects still plan to dwarf this in the coming years upon completion.

The appetite of data centers is tremendous, with large data centers demanding close to one gigawatt of energy (equal to one billion watts). This is comparable to the output of a single nuclear power plant. Some of the places where the data centers are built, however, lack the infrastructure and supply to meet this demand. Consider Memphis, Tennessee, circa 2024. That is where Colossus, a colossal data center, was built for the purpose of training Grok-4—which is, at the time of this writing, XAI’s largest language model and GenAI. Local supply was not enough to power the facility, so XAI brought in 35 gas turbines to make up the difference—a demand of energy to power about 80,000 homes. Residents in local Boxtown, Tennessee, complained at a public hearing that they had noticed an increasing need to use their inhalers. A sharp spike in nitrous oxide, a pollutant known to exacerbate respiratory illnesses such as asthma, was implicated as the cause of the residents’ problems. Colossus’ natural gas generators contributed to increasing the local concentration of NOx by as much as 79% in the surrounding area. This part of Memphis is now considered a “sacrifice zone”—an area facing disproportionate pollution, often inhabited by lower-income or minority populations.

As these data centers create an increased demand for energy, we can expect a proportionate increase in emissions from U.S. power plants and similar gas turbines. With all of the work these data centers put in, another problem will arise from the amount of energy they expend: heat.

If you have any experience with a heavily-used smartphone, you know it’s bound to generate substantial heat. Most computers and laptops use fans to help dissipate that heat through radiation and conduction. However, with the amount of energy and heat that server stacks in these data centers generate, liquid cooling is the primary solution to help dissipate that heat quickly—that is, freshwater.

This type of water usage is problematic from a supply standpoint. The UN has already declared a global era of “water bankruptcy” in which freshwater is being used at a rate that exceeds the recovery rate of its sources. The U.S. is no exception, as many aquifers that support farmland are depleting at accelerated rates, affecting crop yields. Data centers? They use about 300,000 gallons a day—as much as 1,000 American homes.

What happens when that water has been used?  Through a combination of conductive and evaporative processes, used water is either sent back to utilities for water treatment or put into the environment through water vapors. The catch is that it is not just water going back into the environment. The water that contacts the surfaces of the cooling systems contain biocides, anti-corrosive agents, and other contaminants. While smaller applications are not problematic, this changes at scale. There are still questions unanswered about the potential health risks to marine life, the ability of water treatment facilities to remove these contaminants from drinking water, and the potential for carcinogens to enter the environment around these data centers from discharge.

What is driving this demand for AI? Individual users are not the sole or even principal culprits. Businesses across multiple industries are looking to AI for solutions related to operations, identifying market trends, and developing new products for consumers. Healthcare is no exception. From being used as a scribe in our offices, or for inpatient encounters to enhance radiology imaging, AI is quickly finding many uses in healthcare. However, as the user of an AI scribe myself, there is little transparency in how many or what kind of resources these tools use.

What to do?  Awareness is a start. Undoubtedly, many go through their days not aware that using AI to make a five-second video uses the equivalent energy of running your microwave for over an hour, as was reported in MIT’s Technology Review. This awareness helps individuals to be more mindful of their use of AI before making that funny video, or asking Claude about new recipes to try. As the saying goes, “Google is your friend.” These large language models shouldn’t replace your web browser.

For those who wish to be more selective with their use of AI, disabling AI features on websites like your favorite search engine can also help cut down on use. More environmentally conscious solutions exist as well, from “greener” technologies like Tiny AI models, to models trained in data centers that feed off more renewable energies or use less resource-intensive cooling designs.

For clinicians, it's important to demand more transparency about the resources used in each AI tool and also to select the AI tools and language models that utilize resources responsibly. Ultimately, long-term solutions may lie in policy advocacy for more responsible use of resources and regulations related to the extra demand for electricity. The demand and continued search for new applications of this  is likely to proceed unabated, and the time is now to consider the implications this has on our health and the planet.

References 

Atherton, B. (2024, October 21). The square-foot race for AI space: Physical requirements for AI data centers. AFL - Hyperscale Solutions. https://www.aflhyperscale.com/articles/physical-requirements-for-ai-data-centers/

Bignami, E., Darhour, L. J., & Bellini, V. (2025). Sustainable AI in medicine: navigating innovation, challenges, and environmental impact. Health economics review, 15(1), 110. https://doi.org/10.1186/s13561-025-00704-w

Chow, A. R. (2025, August 13). “We Are the Last of the Forgotten:” Inside the Memphis Community Battling Elon Musk’s xAI. TIME; Time. https://time.com/7308925/elon-musk-memphis-ai-data-center/

Copley, M. (2022, August 30). Data centers, backbone of the digital economy, face water scarcity and climate risk. NPR. https://www.npr.org/2022/08/30/1119938708/data-centers-backbone-of-the-digital-economy-face-water-scarcity-and-climate-ris

Cottier, B. (2025, November 19). The largest AI data center campuses will soon be a fifth the size of Manhattan. Epoch AI. https://epoch.ai/data-insights/data-center-sizes

Data. (2025, August 18). The Science and Environmental Health Network. The Science and Environmental Health Network. https://www.sehn.org/sehn/2025/8/14/data-centers-and-the-water-crisis

Hegde, G. (2024, July 24). Discharge from AI Data Centers and How to Mitigate Contamination. KETOS. https://ketos.co/discharge-from-ai-data-centers-and-how-to-mitigate-contamination

Midhat Tilawat. (2025, September 23). Your Premier Source for AI Reviews, Guides, News, and Insights. All about AI. https://www.allaboutai.com/resources/ai-statistics/ai-data-centers/

O’Donnell, J., & Crownhart, C. (2025, May 20). We did the math on AI’s energy footprint. Here’s the story you haven’t heard. MIT Technology Review. https://www.technologyreview.com/2025/05/20/1116327/ai-energy-usage-climate-footprint-big-tech/

Wittenberg, A. (2025, May 6). “How come I can’t breathe?”: Musk’s data company draws a backlash in Memphis. POLITICO; Politico. https://www.politico.com/news/2025/05/06/elon-musk-xai-memphis-gas-turbines-air-pollution-permits-00317582