The rapid adoption of artificial intelligence is increasing electricity demand, water consumption, and carbon emissions from data centers. Live Science reports that data centers currently consume about 414 terawatt-hours of electricity annually, representing roughly 1.5% of global electricity use. The International Energy Agency expects their electricity demand to more than double by 2030.
Generative AI contributes to this demand because large language models rely on energy-intensive graphics processing units and perform enormous numbers of calculations when generating responses. Individual queries use relatively little energy, but billions of daily requests can create a substantial cumulative demand. Companies remain primarily responsible for improving AI efficiency, but researchers say users can also reduce unnecessary consumption.
One approach is to avoid AI when simpler tools can handle a task. A conventional web search, for example, may require less computation than asking a large AI model for basic information.
Users who regularly perform specialized tasks such as translation or summarization can also choose smaller, task-specific models. A 2025 UNESCO study found that specialized models consumed 15–50 times less energy than Meta’s Llama 3.1 while producing better results for their intended tasks. Switching to smaller models reduced overall energy consumption by 90% in the study.
Another strategy is requesting shorter responses. Because language models perform calculations for every word generated, output length strongly influences energy use. Researchers found that halving Llama’s output reduced its energy consumption by about 50%, while shortening the prompt itself produced much smaller savings.
Image and video generation require considerably more computation than text. Experts recommend starting with lower resolutions, editing existing images instead of generating new ones, and creating multiple images in batches.
These individual actions cannot replace industry-wide improvements, but collectively they can help limit AI’s environmental footprint as its use continues to expand.
