AI doesn’t name a single technology, but a “vast toolbox” of systems and techniques. For climate change, it is going to matter how AI is designed and deployed in these critical early years. Understanding AI’s historic roots in military and colonial contexts, engaging with the technology skeptically and critically instead of mindlessly, will be crucial for navigating the ethical and environmental concerns. As Cornell engineering professor Fengqi You puts it, “This is the build-out moment. The AI infrastructure choices we make this decade will decide whether AI accelerates climate progress or becomes a new environmental burden.”
Generative AI is extremely resource-intensive. It is estimated that asking ChatGPT a question instead of doing a Google search consumes about ten times more electricity. Data centers, required to train and run tools like ChatGPT, have high electricity, water, and energy demands, increasing carbon emissions, intensifying grid pressures, and producing toxic chemicals in the processing, mining and transportation of hardware. In 2022, data centers were the 11th largest electricity consumer in the world, between Saudi Arabia and France; but by 2026, they are expected to be fifth in the world, between Japan and Russia.
By 2030, AI could produce 24-44 million metric tons of carbon dioxide, which is the equivalent of 5-10 million cars, and could use 731-1,125 million cubic meters of water per year. Data centers use water both directly (cooling equipment) and indirectly (producing electricity). In 2023, data centers in the U.S. consumed 17 billion gallons of water directly and 211 billion gallons indirectly - this could double or quadruple by 2028. AI infrastructure could soon consume 6 times more water than Denmark; and, within the next five years, data centers could exceed emissions of all cars in California.
The energy and water demands of AI will only increase. By 2030, electricity consumption from data centers will double, according to the IEA. In one extreme scenario, data centers could consume 12% of U.S. electricity. Because of the extremely fast-pace of construction, the majority of this demand will be met by burning fossil fuels, increasing global carbon emissions by 220 million tons.
Karen Hao’s lauded new book Empire of AI (2025) compares the AI industry to colonial powers: “The empires of AI are not engaged in the same overt violence and brutality that marked this history. But they, too, seize and extract precious resources to feed their vision of artificial intelligence.” These resources include personal data, copyrighted work by artists and writers, and land, energy and water. As Karen Hao argues, Silicon Valley operates with a “scale at all cost” approach in an AI race motivated by profit and dominance. This approach necessarily requires an extractive, intensive mode of development, relying on cheap and quick fossil fuels, not only exacerbating climate change, but in fact reversing progress for the clean energy transition.
While Silicon Valley tech companies used to market themselves as eco-friendly, with Microsoft making carbon negative pledges and Amazon investing in electric vehicles, there has been a recent shift to anti-science and climate denialism, fueled by the immense profits in AI. Data centers are responsible for Google’s greenhouse gas emissions rising by almost 50% in five years. Coal plants that were supposed to retire are being kept online for data centers. New methane gas plants are being built. And this new AI infrastructure will “lock us into burning fossil fuels for decades.”
As data centers proliferate across the United States, scientists are scrambling to quantify their impacts on nearby communities. Studies have shown that these facilities strain local power and water supplies, drive up utility costs, and emit harmful pollutants. Now, research suggests their waste heat can actually crank up the temperature in downwind neighborhoods.
Where data centers are located can also have a big impact on efficiency, and location is also a major environmental justice issue. While consuming massive amounts of water, data centers are most often located in water-scarce regions, and just as coal-fired power plants and toxic facilities have been disproportionately sited in low-income communities and communities of color, data centers appear to be following the same racist, exploitative pattern.
The NAACP has threatened to sue Elon Musk’s xAI, for the 35 unpermitted gas turbines fueling its new data center Colossus in Boxtown, Memphis, a historically Black and low-income community where cancer rates are four times the national average. Meanwhile, Meta’s largest data center is being built in Richland Parish, Louisiana, another low-income community, with a footprint the size of Manhattan. Three new methane gas plants will be built to power the data center, which will emit 5,862 tons of carbon dioxide annually - one of the new methane gas plants will be built in a region known as “Cancer Alley.” By 2030, data centers could cause 1,300 premature deaths and 600,000 asthma cases.
There are ways to reduce generative AI’s carbon footprint, but that requires slowing down the pace. As lead of MIT Climate Project’s Decarbonization Mission Elsa A. Olivetti explains, “Due to the speed at which there have been improvements, we haven’t had the chance to catch up with our abilities to measure and understand the tradeoffs.” Some options for mitigating data centers’ carbon footprint include using sustainable materials, less energy-intensive computing hardware, recycling water, reusing components, improving energy-efficiency, and long-duration renewable energy storage. Grid decarbonization, efficiency, and siting could reduce carbon 73%, and water usage 86%.
Yale Climate Connections has made a list of 13 individual actions to push for a climate-friendlier AI. These include weighing your use against the energy consumed, advocating for transparency from companies, and staying informed. However, as they conclude, the climate impacts of AI won’t be solved by individual users. The problem is “an industry that treats scale as success, training ever-larger models for increasingly trivial purposes while communities face water shortages and polluted air.”
Trump’s One Big Beautiful Bill originally included a moratorium on any AI regulation from states for a decade, but the moratorium was ultimately removed by the Senate before the bill was signed into law. The final version of the bill did include a requirement for states to integrate more AI into healthcare in order to receive funding.
On July 23, 2025, the Trump administration released an AI Action Plan along with three executive orders on the development of AI. Together, the Action Plan and the executive orders aim to rapidly accelerate the building of data centers by reducing or streamlining regulation under the Clean Air Act and Clean Water Act, waiving NEPA environmental impact assessment requirements, and making federal lands available for construction. Trump’s AI Action Plan also removed bias protections from federal guidelines, and eliminated references to diversity, climate science, and misinformation, all while Elon Musk’s chatbot Grok has started parroting climate denialism and spewing white supremacist views.
In an April executive order on “Reinvigorating America’s Beautiful Clean Coal Industry,” President Trump called for agencies to identify areas of the U.S. where coal infrastructure could support new data centers. As Mitch Jones of Food & Water Watch notes, the Trump administration’s AI plan is “nothing more than a thinly veiled invitation for the fossil fuel and corporate water industries to ramp up their exploitation of our environment and natural resources - all at the expense of everyday people. The expanding data center industry is being leveraged as an excuse to prolong the life of filthy, climate-killing fossil fuel power and dangerous nuclear plants, and even build new ones.”
On December 11, 2025, Trump signed another executive order relating to AI. The E.O. establishes an AI Litigation Task Force to challenge State AI laws inconsistent with the policy of enhancing U.S. global AI dominance. Some researchers say the E.O. is illegal, as the Trump administration can’t restrict state regulation without the approval of Congress.
Also in December 2025, over 200 environmental groups including Greenpeace and Food & Water Watch wrote an open letter to Congress demanding a moratorium on all new data center construction until regulations are established.
Climate change is not a problem of technology, but a political problem. We have all the tools at our disposal to stop the acceleration of global warming and burning of fossil fuels - AI is another one of these tools with potential to solve complex problems in mitigation and adaptation of climate change, if it is designed and deployed properly.
AI empowers climate scientists to process vast amounts of environmental data—far beyond what humans could handle alone. For example:
AI can give policymakers better data and smarter tools for decision-making, and can democratize access to sustainability tools:
