Why AI Is Bad for the Environment: Energy, Water, Waste

AI is bad for the environment because it consumes enormous amounts of electricity and water, generates significant greenhouse gas emissions, and produces toxic electronic waste. A single generative AI query can use seven to ten times more energy than a standard web search, and data centers are projected to consume 6% of U.S. electricity by 2026. The technology's rapid growth is straining power grids, depleting water resources, and adding hazardous e-waste to landfills.

While AI offers tools for climate modeling and efficiency, its environmental footprint is substantial and growing. This article breaks down the key impacts—energy, water, waste, and carbon emissions—and offers practical steps to reduce your own AI-related footprint.

Energy Consumption: AI's Insatiable Appetite

Data centers that power AI models are energy-intensive. Training a single large model like GPT-3 consumed an estimated 1,287 megawatt-hours of electricity—enough to power about 120 average U.S. homes for a year—and emitted roughly 552 tons of carbon dioxide, according to a 2021 study by Google and UC Berkeley researchers cited by MIT News. Deploying these models for everyday use, such as answering ChatGPT queries, draws even more power over time.

Global data center electricity consumption reached 460 terawatt-hours in 2022, which would rank them as the 11th largest electricity consumer in the world, between Saudi Arabia and France. By 2026, that figure is expected to nearly double to 1,050 terawatt-hours, pushing data centers to fifth place globally, according to the OECD. In North America alone, data center power requirements jumped from 2,688 megawatts at the end of 2022 to 5,341 megawatts at the end of 2023, driven largely by generative AI.

This surge is straining power grids. As Noman Bashir, a Computing and Climate Impact Fellow at MIT, told MIT News, "The demand for new data centers cannot be met in a sustainable way. The pace at which companies are building new data centers means the bulk of the electricity to power them must come from fossil fuel-based power plants." That means more coal and natural gas burning, increasing carbon emissions.

Water Footprint: Cooling the Machines

AI hardware generates immense heat, requiring water for cooling. In 2022, Google's data centers consumed about 5 billion gallons of fresh water for cooling, a 20% increase from the previous year. Microsoft's water use rose by 34% in the same period, according to The Elm. By 2027, AI's projected water usage could reach 4.2 to 6.6 billion cubic meters—four to six times the annual water consumption of Denmark.

This water is often drawn from municipal supplies, competing with residential and agricultural needs. In drought-prone regions, data centers can exacerbate water scarcity and disrupt local ecosystems. The water is used for evaporative cooling, meaning much of it is lost to the atmosphere rather than returned to the source.

Electronic Waste: A Toxic Legacy

AI hardware has a short lifespan, often replaced every few years as technology advances. This creates a growing stream of electronic waste (e-waste). Data center e-waste contains hazardous substances like mercury and lead, which can contaminate soil and water if not properly disposed of, according to The Elm. The United Nations reports that AI data centers generate toxic e-waste, adding to the 50 million tons of global e-waste produced annually.

The production of AI hardware also has environmental costs. Manufacturing a single 4-pound computer for a data center requires about 1,763 pounds of raw materials, including cobalt, silicon, and gold. Mining these materials causes soil erosion, water pollution, and often involves unjust labor conditions.

Carbon Emissions: The Climate Cost

AI's carbon footprint comes from both electricity use and hardware manufacturing. Training GPT-3 emitted about 552 tons of CO2, but that's just one model. With thousands of models being trained and deployed, the cumulative emissions are significant. Data centers currently account for about 2% of global greenhouse gas emissions, comparable to the aviation industry, and that share is rising.

Because many data centers rely on fossil fuel-powered grids, their emissions are not easily offset. Even when companies purchase renewable energy credits, the actual electricity may still come from coal or gas plants, especially during peak demand.

What You Can Do to Reduce AI's Environmental Impact

Individual actions can help mitigate AI's footprint. The University of Maryland, Baltimore's Office of Sustainability suggests asking yourself these questions before using AI, as reported by The Elm:

• Do I really need AI? A simple web search may suffice for many queries.
• Is my input sensitive? Avoid sharing private data with open-source AI tools that may use it for training.
• Can I turn off AI summaries? For example, you can type "-AI" at the end of a Google search to remove the AI summary.
• Am I making every inquiry count? Skip pleasantries like "thank you" to AI—each response consumes energy and water.

Additionally, you can support companies that use renewable energy for their data centers, advocate for greater transparency in AI energy reporting, and extend the life of your own devices to reduce e-waste.

AI's environmental impact is a complex issue with no single solution. While the technology holds promise for climate solutions, its current trajectory is unsustainable. By being mindful of our AI use and demanding greener practices, we can help steer the industry toward a more sustainable future.