AI, climate and environment: energy, consumption and solutions

Understanding the topic

How much AI consumes

Training and running models requires large data centres, buildings full of computers that draw electricity and need cooling, often with water. International energy agencies expect data-centre electricity use to grow strongly in the coming years, and big technology companies have acknowledged that AI makes it harder to meet their own emissions targets. How much a single request to a chatbot weighs depends on the model and on how it is measured: the figures in circulation should be read with the source next to them.

How AI can help

On the other side, AI models are improving weather forecasts, in some cases computing far faster than traditional methods. They are used to predict solar and wind output, balance power grids, cut waste in buildings and factories, spot wildfires and methane leaks from satellites, and search for new materials for batteries and panels.

The final bill

Whether the balance turns out positive or negative is not settled: it depends on how much clean energy powers the data centres, on how efficient chips and models become, and on what we choose to do with all this computing power. It is also a local issue, because new sites ask the communities that host them for electricity, water and land, and protests and legal challenges are not lacking. Here we also follow climate and environment news in general, when it helps to understand the context.

What we follow

• electricity and water use of data centres
• energy for AI: renewables, nuclear, grids
• AI for weather, climate, agriculture and natural disasters
• local protests, rules and companies' emissions pledges

Latest news on AI, climate and environment

• For Black voters, climate action means little without relief from rising costsGrist · 24 Oct
• OpenAI now trains its frontier AI models with no internet accessThe Verge AI · 23 Oct
• The planet’s vital signs are getting worse. One of the easiest fixes is your diet.Grist · 23 Oct
• Chart: A record two-thirds of car sales in China are now electric vehiclesCarbon Brief · 23 Oct
• Distorted — October 28, 2026Grist · 23 Oct
• Securing our frontier model training environmentOpenAI News · 23 Oct
• Europe’s hottest summer ever was also its deadliest — by farGrist · 23 Oct
• Google's Agent Compute Engine gives every Gemini agent its own hardware-isolated micro virtual machineThe Decoder · 22 Oct
• Nirva raises $33M for a pendant that only listens to youTechCrunch AI · 22 Oct
• Experts: How are climate scientists using artificial intelligence?Carbon Brief · 22 Oct

Videos

Why we shouldn’t recycle old EV batteries… yet.Just Have a Think · 18 Oct Squeezing easy extra juice out of our existing energy systems.Just Have a Think · 11 Oct How China’s Next Five Years Will Transform Global Energy GenerationJust Have a Think · 4 Oct

Timeline

The most widely covered story of each day, newest first. The list grows over time; each headline links to the original article.

• 24 Oct For Black voters, climate action means little without relief from rising costs (Grist)
• 23 Oct OpenAI now trains its frontier AI models with no internet access (The Verge AI)
• 22 Oct Google's Agent Compute Engine gives every Gemini agent its own hardware-isolated micro virtual machine (The Decoder)
• 21 Oct Revealed: Expensive gas behind ‘nearly 90%’ of rise in UK electricity bills (Carbon Brief)
• 20 Oct Q&A: Why a pivotal IPCC meeting in Ethiopia ended in ‘profound disappointment’ (Carbon Brief)
• 19 Oct The world is crumbling in ‘The History of Concrete’ (Grist)
• 18 Oct A new gas line or cash to electrify? California lets households choose. (Grist)
• 17 Oct The fire retardant in the seafood chain (Grist)
• 16 Oct Congress has a plan to stop Trump’s attack on offshore wind. It may not be enough. (Grist)
• 15 Oct COP17: Tracking progress towards the UN’s 2030 nature targets (Carbon Brief)
• 14 Oct After a decade of wildfires, an overlooked election could shape California’s climate future (Grist)
• 13 Oct Why Google is paying to upgrade nuclear power plants (Grist)
• 12 Oct Interactive: Tracking negotiating texts at the COP17 biodiversity summit (Carbon Brief)
• 11 Oct In New Mexico, a river runs through ‘the middle of a hellscape’ (Grist)
• 10 Oct Fifty billion seedlings: Can Ethiopia replant its ravaged forests? (Grist)
• 9 Oct Factcheck: UK Conservatives double the ‘cost of net-zero’ after spreadsheet blunder (Carbon Brief)
• 8 Oct The bipartisan energy bill tearing the climate movement apart (Grist)
• 7 Oct Powering data centers on nuclear? That’s a tough sell in Utah’s coal country. (Grist)
• 6 Oct Supreme Court wrestles with who should pay for climate change (Grist)
• 5 Oct Extreme heat left this student athlete on life support. He won’t be the last. (Grist)
• 4 Oct To win over locals, data centers pump $340M into rural Texas projects (Grist)
• 3 Oct New Colorado neighborhood could offer glimpse into the future of home energy use (Grist)
• 1 Oct Under Trump, California’s clean car rules are essential — and vulnerable (Grist)
• 30 Sept Analysis: Wind and solar save UK from gas imports worth £5.9bn during Hormuz crisis (Carbon Brief)
• 29 Sept ‘Digger’ is the year’s big climate movie. This climate reporter hated it. (Grist)
• 28 Sept Analysis: Lula presidency saved at least 20,000km2 of Brazilian Amazon since 2022 (Carbon Brief)
• 27 Sept Why Germany is building an ark for US climate data (Grist)
• 26 Sept A rural Idaho county banned renewables. It’s having second thoughts. (Grist)
• 25 Sept ‘Concerning’ low sea ice persists as Antarctic hits third-lowest winter peak (Carbon Brief)
• 24 Sept Interactive: What China’s dozens of ‘five-year plans’ say about climate and energy (Carbon Brief)
• 22 Sept Explainer: How sea level rise poses an ‘existential threat’ to humans, heritage and nature (Carbon Brief)
• 21 Sept Analysis: ‘Super El Niño’ breaks ‘remarkable’ all-time record (Carbon Brief)

Frequently asked questions

How much energy does artificial intelligence use?

It depends on the model and on the use. Most of the consumption happens in data centres, both to train models and to answer everyday requests. International estimates point to strong growth in data-centre electricity demand in the coming years, driven in part by AI.

Why do data centres use water?

Because computers produce heat and many sites cool them with systems that use water, either directly or through the generation of the electricity they consume. The amount varies widely with the cooling technology and the local climate.

Can AI help fight climate change?

It can contribute: it improves weather forecasts, helps run power grids with a lot of renewables, cuts energy waste, speeds up research into new materials and makes it possible to monitor forests, ice and emissions from satellites. It does not replace political and economic choices on cutting emissions.

Does using a chatbot pollute?

Each request uses a small amount of energy, which becomes significant when multiplied by billions of uses. The real impact depends on the model, the efficiency of the data centre and the energy source that powers it.

How are AI data centres powered?

With electricity from the grid, which has a different mix of sources in each country. Big companies are signing contracts for solar, wind and nuclear power and in some cases building dedicated plants; where that is not enough, gas is used as well.

Other FAIND topics