Environmental impact of AI

basicwidespread

Environmental impact of AI refers to the electricity, water, and raw material demand, as well as carbon emissions and waste associated with building and using artificial intelligence.

In plain terms

When you type a question into a chatbot, the answer is not generated on your phone but in a distant data center where thousands of machines work. These consume electricity, heat up, need cooling, and are replaced every few years. The burden of all this on the environment is the environmental impact of AI.

It is like tap water: you only see the tap, but behind it there is a reservoir, pumps, a pipe network, and a treatment plant. A single glass of water hardly matters, but the consumption of an entire city does. The same is true for AI: the impact of one query is small, but billions of requests run daily.

The concept is useful for comparing the benefits of AI with its costs. AI can indeed help the environment, for example through smarter control of energy grids or more accurate weather forecasts. The question is therefore what we use it for and how much it costs the planet.

How it works

Manufacturing: models require special chips, whose production demands significant energy, water, and hard-to-extract raw materials. This impact exists even before the machine is turned on.
Training: training a large model takes weeks or months, during which thousands of chips run non-stop. This is a one-time but large energy cost per model, repeated with experiments and every new model variant.
Usage: every query, image generation, or task performed by an agent involves further computation. Since there are so many, cumulative usage consumption typically exceeds training consumption over time.
Cooling and power source: the heat generated by machines must be dissipated, often using water. Emissions depend on whether the data center receives power from coal, gas, nuclear, or renewable sources.
Obsolescence: hardware becomes outdated quickly, and decommissioned machines become electronic waste, only part of which is recycled.

An everyday example

You ask the AI to draw ten variations for a birthday invitation, then another ten because none of them are satisfactory. Generating one image requires far more computation than a short text response, so twenty images together use noticeably more electricity than asking for an answer to a single question. The consumption does not appear on your electricity bill, but on that of a distant data center.

Who is affected

everyoneThe growing electricity and water demand of data centers affects energy prices, network load, and climate goals, even if you do not use AI yourself.

entrepreneur, executiveA company's sustainability report and costs increasingly factor in how much and what kind of AI services are used.

developerThe scale of consumption may depend on which model size you choose and how many steps you run the agent through. → Code

parent, student, teacherConscious AI use includes understanding that digital services also have a physical, environmental cost. → Education & Family

A common misconception

MythA single question posed to an AI wastes enough electricity and water to cause serious harm to the planet.

In realityThe consumption of an average text question is low: according to providers' own 2025 statements, it is a few tenths of a watt-hour, roughly equivalent to ten seconds of TV viewing. Long, multi-step tasks, and especially image and video generation, consume much more. The issue is the total volume, the rapid expansion of data centers, and where the electricity and water come from locally.

Limits and risks

The exact figures are uncertain: companies share little data and use different measurement methods, so estimates in the news can differ by multiples.
More efficient models do not necessarily reduce overall consumption, because cheaper AI is used by more people and more often (this is the rebound effect).
The claim 'we operate on 100% renewable energy' often refers to annual accounting, not that the data center uses green power every hour.
The burden falls locally: a data center built in a water-scarce area or a region with a weak grid can cause problems there, while the benefits are enjoyed elsewhere.

Similar, but not the same

Green AI – Green AI is the effort to design models more efficiently, while environmental impact is the problem it tries to reduce.
Carbon footprint – Carbon footprint measures only greenhouse gas emissions, but AI's environmental impact also includes water consumption, raw material demand, and electronic waste.
AI in climate protection – This means using AI to solve environmental problems, while environmental impact examines how much the operation of AI costs the environment.

Related concepts

Broader

AI ethics, Responsible AI, Sustainability

Narrower

AI energy consumption, AI water footprint, AI carbon footprint, Electronic waste

On the edge

Data center, Green AI, GPU, Model training, Inference

A brief history

The topic came into the spotlight in 2019, when Emma Strubell and her co-authors calculated the carbon dioxide emissions associated with training a large language model. After the mass spread of generative AI at the end of 2022, the issue moved from professional debates into energy policy and daily news.

Where things stand

In late 2026, the electricity and water demand of data centers is a constant topic in energy policy and local permitting debates. According to the International Energy Agency's 2025 forecast, data center electricity consumption could roughly double by 2030 compared with 2024 levels. Measurement and disclosure are not yet standardized, although the EU AI Act has required providers of general-purpose models to document the known or estimated energy consumption of the model since August 2025.

Status: 2026 November

In the news

Appears in 1 Agent Age stories; the curve shows weekly occurrence in the news magazine.

New procurement guide urges hospitals and payers to weigh AI's environmental impactSep 23, 2026 · Health

Also known as

AI environmental impact · environmental impact of artificial intelligence · AI ecological footprint · AI environmental footprint · environmental impact of AI · environmental impact of artificial intelligence

Sources

This entry was machine-edited and has undergone machine expert review; the “In the news” section updates automatically from Agent Age stories.