Since the release of ChatGPT in late 2022, the rise of Artificial Intelligence, or AI, has been at the forefront of the public’s attention. It’s exploded within the tech scene, with all of the major tech companies developing their own versions. In addition to OpenAI’s ChatGPT software, there is now Google Gemini, Microsoft Azure and Amazon Web Services, among others. The rise of this technology has warranted both praise and scrutiny, and ethical concerns continue to plague its ever-increasing prevalence. AI use has brought forth issues ranging from exacerbating structural biases and discrimination to increasing the lack of transparency within the private sector. Therefore, many have found a reason to be concerned about AI’s rapid expansion into our lives. 

But there’s one pressing issue that has not been talked about nearly enough — the vast and unequally distributed toll that AI data centers have on the environment. On the consumer side, it’s understandably difficult to imagine that typing a couple questions into some website on the internet would have any tangible impact on the world outside of the cloud. But the data centers where AI models are “trained” are huge consumers of electricity and water; by 2027, it is estimated that data centers will emit more carbon than a small country in addition to using up millions of liters of dwindling freshwater supplies. Other concerns include the inherent profit-driven goals of many big tech companies, which lead to selling their AI products to questionable consumers, such as fast fashion producers and major oil companies. Here are some of the main concerns — as well as some promising solutions — regarding AI and its impact on the environment. 

Carbon and climate impacts

It’s not necessarily the act of asking ChatGPT to write your next cover letter that causes the industry’s huge carbon impacts. The source is what makes AI so powerful — its intensive training and associated environmental costs. According to the Harvard Business Review, the “training process for a single AI model, such as a large language model, can consume thousands of megawatt hours of electricity and emit hundreds of tons of carbon. This is roughly equivalent to the annual carbon emissions of hundreds of households in America.” Moreover, one study by UMass Amherst estimated that training one big language model causes the emission of 300,000 kilograms of carbon dioxide, the equivalent of 125 round trip flights from New York to Beijing. Furthermore, the models must be retrained multiple times per year. This process occurs in data centers throughout the United States and worldwide, requiring huge amounts of electricity from local power grids, many of which run on nonrenewable energy sources like coal and fossil fuels. Virginia, for example, has a data center considered to be one of the major centers in the world, and only 1% of the state’s electrical output comes from renewable resources. 

The amount of energy set to be used for AI purposes is not slowing down. The integration of AI into our lives is quite obvious with the incorporation of AI assistants into search engines such as Google and Bing. Using an AI search takes four times as much computing power as a typical search, having tangible impacts on the amount of energy consumed while only offering minor, if any, gains on search accuracy. Pure AI websites have gained high traction lately (ChatGPT has about 13 million users daily), but it still doesn’t compare to the incredibly wide usage of traditional search engines, such as Microsoft’s Bing’s half a billion users daily. 

To combat detrimental effects, big tech companies could choose to build their data centers in areas that mostly rely on renewable energy. This would stop the release of carbon at its source, rather than undertaking net-zero projects such as planting trees (which is not a permanent solution) or buying carbon credits. Advancements in technology could also pave the way for a more climate-friendly AI system. More efficient models of AI training and the creation of energy-efficient graphics processing units and accelerators yield promising results. 

But the climate impact of AI goes beyond its direct power consumption. The ways consumers use it and the industries it supports are just as important when evaluating its holistic impact. Microsoft came under fire when it sold its AI Azure to ExxonMobil in 2019, who hoped to use the software to optimize oil production. Microsoft defended their choice after receiving backlash, stating that they sell to every customer. 

Another potentially environmentally hazardous use of AI is its utilization for marketing techniques. AI-controlled personal ad algorithms push consumerism and fast fashion in a way never seen before. Fast fashion, which already accounts for 8% of global carbon emissions, has a significantly negative impact on the environment, including pollution, waste production and carbon usage. The rise of AI in targeted advertising may exacerbate companies’ reach within an already chronically consumerist society. 

Water usage

Climate impacts alone cannot properly describe the overall environmental impact of AI, especially on a local scale. The cooling of data centers requires huge amounts of freshwater, often in areas already experiencing drought, such as Arizona and Chile. Data centers draw millions of liters of water either for energy generation or for cooling servers. Freshwater usage is a rapidly expanding global problem, and severe water scarcity already impacts ⅔ of the world’s population for at least one month per year. Google’s self-owned data centers alone (not even counting their third party colocation facilities) “directly withdrew 25 million liters and consumed nearly 20 billion liters of scope-1 water for onsite cooling in 2022, the majority of which was potable water” (Li, et al., “Making AI Less Thirsty”). 

And, like carbon emissions, this number is only growing. Google’s data center water usage increased by 20% when compared to 2021, and Microsoft’s by 34%. By 2027, it is estimated that water withdrawal for AI will equal the total annual withdrawal of four to six Denmarks, or about half of the United Kingdom’s annual withdrawal. 

Even if big tech companies are able to reduce the total amount of carbon released through their operations, this is only half of a very important battle. To create sustainable AI, we must address the question of freshwater depletion in an urgent manner. As stated in a joint study by professors from UC Riverside and UT Arlington, “AI models” water footprint can no longer stay under the radar — water footprints must be addressed as a priority in the collective efforts to combat global water challenges” (Li, et al., “Making AI Less Thirsty”).

Distribution of impacts and resulting inequalities

AI’s impact on the environment tends to exacerbate preexisting climate inequalities. As previously mentioned, pollution generated by data centers’ electricity consumption is worse in areas where energy production is run on nonrenewable resources, which is often in areas already impacted by heavy pollution. For example, Google runs its Finland data centers on 97% carbon-free energy, but data centers across Asia only run on 4-18% carbon-free energy. The same logic goes for water — water consumption is higher in hotter, more drought-stressed regions because more water is required to cool servers in the heat. 

Big tech companies are incredibly wealthy and run the risk of imposing negative impacts on underdeveloped countries. Current plans to improve AI’s environmental sustainability “often prioritize easily measurable environmental metrics such as the total amount of carbon emissions and water consumption. They do not give enough attention to environmental equity — the imperative that AI’s environmental costs be equitably distributed across different regions and communities” (Ren, Wierman, “The Uneven Distribution of AI’s Environmental Impacts”). Continuing to run AI at its current rate will worsen the environmental inequities that are globally the status quo — subjecting already vulnerable countries to increased air, water and hazardous waste pollution at the hands of a few U.S. tech giants. 

Geographical load balancing: a possible solution

Geographical load balancing, shortened as GLB, is a well-established technique within AI that reroutes computing load across worldwide data centers in order to align to real time grid conditions and carbon intensities. Some major AI networks use GLB, including Google’s carbon intelligent computing platform. Current GLB systems have been shown to increase environmental inequities, but models that are refitted to place higher weight on disproportionately affected areas provide some hope for balancing future environmental impacts, along with decreasing air pollution and increasing water efficiency. 

AI as a force for good

While the current impact of AI on the environment can realistically only be negative, the computing power of AI may ironically be an important tool for combating environmental problems. There have been some positive models developed, such as a program that combines machine learning with satellite imagery to identify buildings damaged in climate change-fueled natural disasters. There are also models that can identify sources of pollution and monitor emissions. It could also be used to reduce the effects of the climate crisis by developing smart grid designs and low-emission infrastructure while effectively modeling future impacts of climate change. It is difficult to predict what industries will make use of AI in the future, but if tech companies continue to chase profit incentives, it is unlikely that climate change research and mitigation will be at the forefront of their research, even though it could produce positive results. 

Like any other form of technology, AI is a tool. It is a tool that comes across as misleading in its apparent immateriality, despite its very real impacts. It is a tool designed by humans based on data collected by humans, and can therefore never be truly unbiased. AI itself can never generate new knowledge, instead relying on pattern recognition to create the most likely outcome. But in a world full of inequalities and a never-ending push towards market consumerism, the most likely outcome is not always the best one. 

As big tech marches on, it is important to keep one question in mind: Who will benefit? There are undoubtedly positive uses for AI. But as we move closer and closer into a looming climate disaster, society must reflect on what values we reinforce by integrating AI into our lives. While I believe that AI could ultimately be a force for good, even through an environmental lens, the tool of AI ultimately depends on how we use it. But unless the leaders of the big tech giants experience a change of heart, it seems unlikely that AI will yield more benefits than the pollution it creates.