Artificial Intelligence: A Sustainable Future or An Unsustainable Burden
By Erika Pietrzak, February 20, 2025
AI could be used in many ways to benefit us all, but without regulation and a focus on sustainable improvement, the damage AI will create will be detrimental.
Artificial Intelligence (AI) has swept across the globe at supersonic speed as generative AI models like ChatGPT, Apple AI Assistant, and other AI software become a common part of many people’s lives. AI is a catch-all term for technologies that can partially mimic human thinking and has been around in rudimentary ways since the 1950s. However, the advent of generative AI in recent years is unprecedented. As of August 2024, 82 percent of large enterprises were experimenting or were considering experimenting with AI.
With interest in AI skyrocketing, the demand for high-performance computing hardware grows daily, increasing investments in generative AI. Generative AI is the type of AI that creates new, unique content, such as images, videos, and text. This increased demand has caused a dramatic rise in the construction of massive data centers to hold all of this hardware and computing infrastructure. The world’s AI data centers have boomed from 500 thousand in 2012 to 8 million today. The current demand for these new centers can, quite simply, not be met in sustainable ways. Because these centers are being built so fast, fossil fuels are being used recklessly.
Energy
Companies investing in AI have seen a dramatic increase in their overall energy consumption as they develop this emerging technology. Google’s carbon emissions have risen 48 percent in the last 5 years, as a result of integrating AI into their data centers, while Microsoft’s carbon emissions have increased 30 percent within the past 4 years. A computer engineer at the University of Illinois Urbana-Champaign expressed that these increased emissions come because “[u]sually people don’t care about energy efficiency when you’re training the world’s largest model.”
A single ChatGPT query consumes ten times the energy of a Google search. North American data centers have increased from 2,688 megawatts by the end of 2022 to an estimated 5,341 megawatts by the end of 2023. The 460 terawatts consumed by data centers around the world in 2022 placed themas the 11th largest consumer of electricity in the world. This already staggering rate of electricity consumption is expected to rise to the 5th largest in the world by 2026. By 2026, the “agency projects that data centers’ energy consumption will have increased by between 35 percent and 128 percent,” the same amount of energy consumed by the entire country of Sweden on the estimated minimum end and Germany on the estimated maximum end.
The training process for a model like OpenAI’s GPT-3, which includes the input of mass quantities of training data, consumes approximately 1,287 megawatts of electricity and generates roughly 552 tons of carbon dioxide. To train an AI model, up to 90 percent of consumed energy comes from accessing its memory, and is used for seemingly simple tasks such as identifying apples in a picture. More specifically, Generative AI has significant fluctuations in energy use that power grid operators, and their diesel-backed generators, are not accustomed to. In Ireland, a major technological hub, 35 percent of the country’s energy use is projected to come from AI by 2026. This significant use of electricity is worrisome for many experts who caution that without finding ways to decrease AI’s energy consumption, the software may become, a Roy Schwartz of the Hebrew University of Jerusalem, Israel stated, “a rich man’s tool.”
Water
Beyond the high-level use of energy involved in accessing system memory and maintaining data centers, every electron’s movement through AI data chips releases heat. In order to absorb heat from computing equipment in data centers, chilled water is used in abundance as a coolant. For each kilowatt hour (kWh) of energy data centers consume, about two liters of water are needed for cooling purposes. In the United States, a typical large data center consumes as much water on a daily basis as 4,200 people. Before microchips are even integrated into the technology in data centers, however, AI requires 2.1-2.6 gallons of water for a single microchip to be produced. Between 2021 and 2022, Microsoft increased its water consumption by 34 percent, using over a staggering 1.6 billion gallons. Google also increased its water consumption in the same period by 22 percent, or a mind-blowing 5.56 billion gallons, “about 800 times as big as The Lincoln Memorial reflecting pool.”
Globally, AI data centers consume six times more water than the country of Denmark by some estimates, equating to around 5 billion cubic meters. In the United States, 20 percent of data centers draw their water from moderately to highly stressed watersheds in the western US. This water could be used to help an estimated 66 percent of the world’s population likely to face water scarcity this year. Without regulation, money-hungry businesses will continue to waste and hoard water resources to help someone write an email they don’t want to write themselves rather than provide life-saving water for millions. This dilemma is likely to cause serious human rights crises when companies will inevitably be forced to choose “between shutting down operations or overtaxing municipal water supplies that communities rely on.”
Beyond cooling, AI technology also indirectly consumes even more water through their power supplies sourced from thermoelectric or hydroelectric plants, which demand abundant water to run. The average amount of water evaporated during the energy generation process from these plants is two gallons per kWh of electricity consumed. Exact figures for water consumption for data centers are incomplete because of a lack of transparency from these centers. However, researchers believe that each AI prompt generated uses about 16 ounces of water.
Mining and Waste
The hardware for AI technology includes microchips made out of rare earth materials that are often mined in “environmentally destructive ways.” Mining these materials often results in extensive soil erosion and increased pollutants released into the surrounding communities. Once submitted, AI queries require significant energy, which most cheaply and commonly comes from fossil fuels for these businesses. The data centers required for our modern use of AI also produce notable hazardous waste, including mercury and lead. Mercury can infect water systems and be “taken up by microorganisms that undergo biomagnification thus causing detrimental impacts of various degrees on aquatic lives.” Many electronics are not properly recycled, resulting in these hazardous waste materials contaminating soil and water. The World Health Organization states that electronic waste, like that caused by generative AI, can “disrupt the development of the central nervous system during pregnancy, infancy, childhood and adolescence” as well as be detrimental to the structure of and functionality of lungs.
AI requires the mining of precious minerals and metals that are already scarce. Four minerals central to semiconductor production are gallium, germanium, palladium, and silicon. Gallium, a requirement for maintaining speed and capacity of AI models, is exceedingly rare, as it represents less than 19 parts per million of the minerals and compounds present in the earth’s crust. Similarly, there are no viable alternatives for germanium in AI hardware, and germanium “is sourced as a byproduct of the processing of zinc ores; it does not occur as a natural metal.” Germanium is so rare that it only compromises 1.5 parts per million of the minerals and compounds present in the earth’s crust. Once mined, germanium must be significantly purified for usage, compounding energy costs. Improper mining techniques of zinc have also frequently resulted in environmental disasters that threaten thousands of lives and significantly disrupt agricultural practices.
Trump’s Executive Order 14179
Through Executive Order 14179, President Donald Trump seeks for the United States to dominate AI fields by means of a massive AI technological expansion across industry sectors. This expansion will require hundreds of new data centers, microchips, and other processing aids that need significant water, electricity, and fossil fuels to function. The Order revoked former President Joseph Biden’s Executive Order 14110, also called the Safe, Secure, and Trustworthy Development and Use of Artificial Intelligence Order. One technology expert in London is quoted as saying Executive Order 14110 represented “the US’ commitment to ensuring that there are some guardrails in place around AI use,” something that President Trump has revoked in the first few weeks of his presidency.
The increased energy required for this AI tech boom will almost certainly come from fossil fuels as the Trump administration encourages an “energy exploration and production on Federal lands and waters,” that endangers and disregards protected environmental areas. Fossil fuel demands will increase, putting more stress on our national grid while increased emissions, energy use, and waste from AI centers will exacerbate the environmental crisis.
The Trump administration promises in an executive order to “further revoke a series of ‘burdensome and ideologically motivated regulations’ designed to support the decarbonization of the US energy sector, as well as other environment regulations.” This initiative will create barriers to furthering renewable energy development, increase America’s contributions to climate change, and disproportionately impact marginalized groups, particularly Indigenous communities. Trump’s “America First” approach heavily favors American technological companies by monopolizing energy and water sources for their exclusive use.
Current Efforts
MIT and other institutions are leading the way in finding more sustainable ways to improve AI, though the future of AI is unknown, and unprecedented complications may arise. Such advancements include placing memory panels inside a computer’s core to reduce the amount of distance data must travel inside the computer. Moreover, moving AI training and deployment to more water-abundant regions would make it possible to reduce the large water footprint of AI. Companies like Veolia are working diligently to help data centers reduce their water use through remote monitoring, which has reduced AI water usage in some instances by 50 percent.
Lawmakers are attempting to tackle gaps in AI legislation at the federal level through the Artificial Intelligence Environmental Impacts Act of 2024, introduced by Massachusetts Senator Edward Markey. The Act would, amongst other things, require a thorough study of the environmental impacts of AI, create a reporting system for data centers to transparently disclosetheir environmental impacts, and direct Congress to issue an environmental impact report within four years of its findings. The Act, however, remains stagnant in committee and will likely not progress further as it has yet to be reintroduced to this Congress. On a more positive note, the EU recently adopted a similar proposal that would require AI systems to “report their energy consumption, resource use, and other impacts throughout their lifecycle,” which will take effect this year.
These kinds of changes, however, require a “comprehensive consideration of all the environmental and societal costs of generative AI,” along with detailed assessments of the perceived value in benefits. Since generative AI “can detect patterns in data, such as anomalies and similarities, and use historic knowledge to accurately predict future outcomes,” such technologies could be “invaluable for monitoring the environment, and in helping governments, businesses and individuals make more climate-friendly choices.”
Moving Forward
There are many other issues with AI than just the ones mentioned in this piece. AI could be used in many ways to benefit us all, but without regulation and a focus on sustainable improvement, the damage AI will create will be detrimental. Next time you decide to use AI, think about the specific program you chose to use. Programs like ChatGPT require significant water and energy, while applications like Permutable and Claude, while still causing significant environmental impacts, dedicate more of their resources towards sustainable building AI products.
Before you submit a query in an AI program, ask yourself three simple questions: 1) Can my question be answered with a simple web search?, 2) Can my question be answered using other types of software (ex: use Desmos for graphing or DoMyEssay.com for writing)?, and 3) Is my query worth the fossil fuel emissions, water consumption, energy use, and unethical mining needed to sustain the program? In order for AI to usher in the technological revolution many envision, it must first integrate sustainable practices that ensure a better future for all.
Change The Chamber is a nonpartisan coalition of over 100 student groups, including undergraduates, graduate students and recent graduates.