Lesson 27 of 28 · Using AI well

Is AI bad for the environment?

AI does have an environmental cost, mainly the electricity and water used by data centres to train and run models, plus the emissions from making the chips and servers. How large the impact is depends on how much AI is used and how clean the local electricity is. AI can also cut energy waste elsewhere, so the net effect depends on what it is used for.

Where AI's footprint comes from

AI models run on specialised chips in data centres, large buildings full of servers. Their footprint has four main sources:

Training: building a large model can mean running thousands of chips for weeks, a big one-off energy cost.
Inference, meaning using a trained model: every question runs the model again. One answer uses little energy, but billions of answers add up, so for popular services everyday use can outweigh training over time.
Cooling: servers produce heat, and many data centres use water to cool them; power stations also use water to generate the electricity.
Hardware: making chips and servers takes energy, water and raw materials, and retired equipment becomes electronic waste.

How big is the impact?

The International Energy Agency's Energy and AI report (April 2025) estimated that data centres used around 1.5% of the world's electricity in 2024, about 415 terawatt-hours, and projected in its base case that this would roughly double to around 945 terawatt-hours by 2030, with AI the main driver of the growth. Data centres also run email, streaming and cloud storage, not just AI.

As a share of global electricity that is still small, but the impact is local: data centres cluster in particular regions, where they can strain power grids and water supplies. Estimates of the energy per AI question vary widely and are often disputed, because they depend on the model, the hardware, the length of the answer and the data centre's power source.

The other side: efficiency and benefits

Chips and models keep getting more efficient, so the energy needed for a given task tends to fall over time. Smaller models now handle many everyday tasks that once needed much larger ones, and some run directly on phones and laptops. Many data-centre operators also buy renewable electricity.

AI is also used to save energy: forecasting wind and solar output, managing power grids, tuning heating and cooling in buildings and speeding up research into new materials. The catch is the rebound effect: when something becomes cheaper and more efficient, people often use much more of it, so total consumption can still rise. Whether AI is "bad" overall depends on the balance between these forces.

What you can do

A single question is small, but habits multiply across millions of users. These habits cut energy use and, on token-billed plans, cost as well:

Use the right-size tool: a small, fast model is enough for simple questions; save large reasoning models for hard problems.
Write clear prompts so you get a good answer first time, instead of regenerating again and again.
Start a new chat for a new topic, because long conversations resend their whole history with every message.
Ask for shorter answers when you do not need long ones, and skip image or video generation you will not use.
If you buy AI for an organisation, ask vendors about their energy sources and sustainability reporting.

Example

Why small habits add up

Worked example

Call the energy of one typical AI answer E. The real value varies, so this example uses no real figures. One person, 20 answers a day: 20 E One million people, 20 answers a day: 20,000,000 E If each person avoids 5 needless regenerations a day: 1,000,000 × 5 = 5,000,000 E saved per day, a quarter of the total, without anyone giving up AI.

The same arithmetic explains why AI's footprint grows with adoption: each answer is small, but the number of answers is enormous.

Key points

AI's footprint comes from electricity and water in data centres, plus the hardware itself.
The IEA projected in 2025 that data-centre electricity use would roughly double by 2030, driven mainly by AI.
The impact is felt most locally, where data centres strain grids and water supplies.
Efficiency keeps improving and AI can save energy elsewhere, but rising use can cancel the gains.
Right-size models, prompt clearly and avoid needless regeneration to cut energy and cost.

Try it on Mindber

Compare models to find the right size

See how token volume adds up

Find on-device AI tools

Glossary: inference

AI fundamentals

Frequently asked questions

How much energy does one AI question use?

There is no single answer. Published estimates differ widely because the energy depends on the model's size, the hardware, the length of the prompt and answer, whether reasoning or image generation is involved, and how efficiently the data centre runs. A short text answer from a small model uses far less than a long reasoning chain or a generated video.

Does AI use a lot of water?

Data centres use water mainly for cooling, and power plants use water to produce the electricity they draw. The amount varies greatly with location, climate, season and cooling design: some sites use little water, others a lot. Research has drawn attention to this hidden water footprint, and some operators now publish water figures in their sustainability reports.

Can AI help fight climate change?

It can help. AI is used to forecast wind and solar output, balance electricity grids, cut energy waste in buildings and factories, and speed up research into batteries and materials. These gains are real but not automatic: they depend on AI actually being applied to such problems, and they have to outweigh the extra energy AI itself consumes.