If you’ve been holding off on AI because of the environmental impact, the data tells a very different story than what’s been spreading online. Here are the real numbers, properly sourced.
I started this account because I don’t want women to get left behind in this AI era. The people who use AI the most are getting promoted, building careers, starting side hustles and making more money.
One of the biggest reasons I hear from people who haven’t started is guilt about the environment. So I pulled the actual data from primary sources, not viral posts. Here it is.
The International Energy Agency is the go-to source on global energy. Its Energy and AI report and its April 2026 update give the big numbers.
01
1.5% of world electricity
Data centres used about 415 TWh in 2024, roughly 1.5% of global electricity. That covers everything, from streaming and cloud storage to AI.
02
17% growth in 2025
Data centre demand jumped 17% in 2025, and AI-focused data centres grew even faster. The IEA calls AI the most important driver of data centre growth.
03
About 3% by 2030
The IEA expects data centre use to roughly double to around 945 TWh by 2030, about 3% of world electricity. It expects renewables to meet about half of that extra demand.
So yes, data centres use real energy and the number is growing. The question most people actually have is different: what does my use add up to?
01
About 0.3 Wh per text question
Epoch AI, an independent research group, estimated a typical ChatGPT text query at about 0.3 watt-hours. Google measured its median Gemini text prompt at 0.24 Wh, “equivalent to watching TV for less than nine seconds.”
02
One second of microwave
A 1,000-watt microwave running for one second uses about 0.28 Wh. That’s where the viral comparison comes from. The IEA never said it. Researcher Hannah Ritchie used it with Google’s figure.
03
100 questions a day is about 90 seconds of your home
100 questions × 0.3 Wh = 30 Wh. The average US home uses over 28,000 Wh a day, so 100 questions is about 0.1% of it, roughly 90 seconds of normal household use.
04
The old 3 Wh number is outdated
The widely shared 3 Wh per question estimate dates from 2023. Epoch AI’s estimate is about 10 times lower, mostly because of newer chips and more realistic answer lengths. Google says its own per-prompt energy fell 33x in one year.
Not every use is a short text question. Very long documents, image generation and video use a lot more per request, and AI agents running for hours add up. Training new models is energy-intensive too. And while energy per AI task keeps falling, the IEA notes that total use keeps rising because so many more people and companies use AI.
My take: the guilt people feel about everyday personal use isn’t backed by the data. Use it on purpose, for things that matter.
01
Try Claude on the free plan
Go to claude.ai and sign up for the free plan. Have a real back-and-forth conversation and see what it’s good at.
02
Use it for one job task for a week
Pick the most boring, repetitive part of your work and use AI for only that, for 7 days. This prompt sets it up.
03
Go deeper from there
After a week you’ll know if it’s worth more of your time. If it is, my free guides walk you through every direction you can take it.