AI and the Environment

AI can feel weightless, just text on a screen, but it runs on very physical infrastructure. Behind every chatbot reply are data centers full of computer chips that draw electricity and use water to stay cool. As AI use has exploded, so has attention to its environmental cost. This page looks at what we know, what is still debated, and how to think about it.

Why AI Has an Environmental Cost

Two stages use resources. Training a large model is a one-time but enormous computation across thousands of specialized chips running for weeks. Inference, the everyday use when millions of people send prompts, adds up across billions of requests. Both run in data centers that:

• Draw electricity to power the chips, much of it still generated from fossil fuels in many regions.
• Use water to cool the equipment, and indirectly through the power plants that supply the electricity.
• Require hardware whose manufacturing and eventual disposal carry their own footprint.

The Big Picture: A Fast-Growing Footprint

The clearest part of the story is the aggregate trend. As AI scales up, its share of global energy and water use is rising quickly:

The exact figures vary by study, but the direction does not: training and running AI at global scale uses a large and growing amount of power and water, enough that some new data centers are straining local electricity grids and water supplies.

What About a Single Prompt?

Here is where the numbers get noisy, and where your AI literacy really matters. You may have seen the claim that one AI prompt "drinks a bottle of water." Before you read any numbers, make a prediction and commit to it.

It Is Not Shared Equally

Environmental costs do not land on everyone the same way. Data centers are often built where land and power are cheap, and some draw water from regions already facing drought. The communities living near a data center may carry the local cost of its energy and water use without seeing much of the benefit. This is part of why AI's environmental impact is also an equity question, connecting back to the fairness themes in this module.

How to Think About It

You do not need to feel guilty about a single question to an AI tool; one prompt is a small thing. But the totals are real, and worth keeping in view:

• Be aware, not paralyzed. Use AI where it genuinely helps, and skip it where it does not add much. Choosing not to use AI remains a legitimate position.
• Bigger tasks cost more. Generating long videos or many images uses far more than a short text question.
• Efficiency is improving, and providers are reporting more openly than they used to, partly because students, researchers, and the public have pushed them to.
• Read claims critically. When you see a dramatic "AI uses X water" headline, ask what was measured and who is reporting it.

Knowledge Check

Select an answer to see feedback. Each option explains why it is or is not correct.

Sources