AI and the Environment: Considerations and Concerns

Developments in generative AI and the promise of further innovation gesture toward a near future full of computational possibilities and solutions to problems across many sectors. But at what environmental cost do these advances come? With the rise of popular AI programs like Gemini and ChatGPT and their release to the general public, awareness of the technology's environmental costs is growing. AI can arguably help us find solutions to environmental problems, but very real costs accompany the creation and application of this technology. Among the impacts to consider: infrastructure needs, energy use, e-waste and resource extraction, carbon emissions, and effects on local communities.

Aerial View of the Amazon CC BY-SA 2.0

Infrastructure Needs: E-Waste and Extraction

The hardware and software that support AI across its lifecycle can be called AI infrastructure. As AI technologies advance, the need for sophisticated infrastructure grows with them. Hardware and components that enable massive data storage, computation, and data processing are critical to keeping AI running. Producing these specialized parts—GPUs, TPUs, and custom chips—ultimately generates a higher volume of e-waste, and manufacturing them requires continued extraction of rare earth and other minerals (for example, lithium and cobalt).

The environmental costs to the places where these minerals are mined vary in degree but can include deforestation, pollution of local resources, and habitat destruction. The communities most affected by extraction also tend to benefit the least from AI's application—raising a larger question of social inequity and access.

Energy Consumption: Electricity and Water

Training and running large AI models requires enormous computational power, which requires a great deal of energy. Data centers consume high volumes of electricity, which can strain already-burdened grids. As an early, illustrative benchmark, training the original GPT-3 reportedly required about 1,287 MWh of electricity and emitted roughly 552 tons of CO₂. That figure is now dated: today's flagship models are far larger, though their exact training footprints are generally not disclosed by developers. Notably, the per-task energy cost of AI has been falling rapidly as efficiency improves—but total demand keeps climbing because far more people are using AI, and for more energy-intensive purposes.

The aggregate trend is striking. In its 2025 Energy and AI report, the International Energy Agency projected that global data-center electricity use would roughly double from about 460 terawatt-hours (TWh) in 2024 to around 945 TWh by 2030—slightly more than Japan's total electricity consumption today—reaching just under 3% of global electricity demand, with electricity from AI-optimized data centers growing several times faster than the data-center sector overall. The IEA later reported that data-center electricity use surged about 17% in 2025 alone, far outpacing the roughly 3% growth in overall global electricity demand.

Water is the other major input. Vast amounts are needed to cool data centers and prevent overheating. Google, for instance, reported using roughly 6.1 billion gallons (about 23 billion liters) of water to cool its data centers in 2023—a figure it attributed in part to a 17% jump driven by AI—and its most recent environmental report put 2024 global data-center water use at around 7.7 billion gallons. When a data center sits in a water-scarce region (the American Southwest, for example), this can strain local water budgets; in some communities, data centers have drawn a significant share of the local supply. Some operators are responding with closed-loop and "zero-water" cooling designs, though how widely these are deployed in practice remains unclear.

Carbon Footprint

AI-driven cloud computing generates significant CO₂ emissions. Estimates vary, but data centers and data transmission are generally put at roughly 2–3% of global energy-related greenhouse gas emissions—and on the IEA's projections, data centers alone are set to account for close to 3% of global electricity demand by 2030. As AI continues to grow in capability and popularity, and as energy-intensive uses like AI agents spread, these numbers can be expected to rise—though grid, hardware, and permitting bottlenecks add real uncertainty to how fast.

The Other Side of the Ledger

AI could potentially help address environmental problems—through precision agriculture and more responsible land, water, and crop use; smarter urban planning; and more efficient expansion of public transit, to name a few. But these possibilities won't materialize simply because AI devises them. Like any AI-proposed solution, implementation depends on organizations and companies—that is, on people—and can be strengthened or stymied by social, political, and economic factors. The environmental case for AI, in other words, is not self-executing.

AI and the Environment: Considerations and Concerns

Infrastructure Needs: E-Waste and Extraction

Energy Consumption: Electricity and Water

Carbon Footprint

The Other Side of the Ledger

Further Reading