How much energy does AI use?

When you type a question into a chatbot, the answer doesn't come from your phone. It comes from a data center: a large building full of computers, cooling systems, and power lines, often far from where you live. Training and running artificial intelligence (AI) happens in places like these. They use electricity, and often water for cooling. So it's fair to ask how much, and the answer comes with real numbers.

The International Energy Agency, which tracks energy for governments around the world, estimates that data centers used about 485 terawatt-hours of electricity in 2025. That's about one and a half percent of all the electricity used worldwide. Data centers run everything online, from email to streaming to banking, but AI is now the biggest driver of their growth.¹²⁶

The same agency expects data center electricity use to roughly double by 2030, to around 950 terawatt-hours. That's a little more than Japan uses in a year today. In some places the effect is large. In the United States, data centers are on course to account for almost half of the growth in electricity demand between now and 2030. That puts real pressure on power grids, and it's why energy has become a big part of the AI conversation.¹²⁶

What about a single question? Estimates vary with the model and the task. Google has published one measurement for its Gemini app. The median text prompt used about a quarter of a watt-hour, which Google compares to watching television (TV) for less than nine seconds. That figure comes from the company itself, and it covers ordinary text prompts, not images, video, or long reasoning tasks, which use more. It does show that a typical question is small, and that the large total comes from billions of them, plus the training of the models themselves.³⁴

Data centers also use water, mostly to keep the computers cool, and researchers at the Massachusetts Institute of Technology (MIT) note that this can strain local water supplies. Google's same report put its median prompt at about five drops of water. As with electricity, the amount for one question is small, and what matters is the total in each place.³⁵

There's more to the story than demand. The International Energy Agency (IEA) expects renewable energy to meet nearly half of the growth in data center electricity through 2030. Google reports that the energy used by its median prompt fell by a factor of thirty-three over one year, as its chips and software got more efficient. And AI is also being used in the energy system itself, to run grids and find savings. None of that cancels the growth, but it means the picture moves in more than one direction.¹³

You don't need to feel guilty about asking a chatbot a question. If you'd like to make a difference, the useful choices are practical. Use a smaller, faster model when it's good enough for the job. Skip the long thinking mode for simple questions. And value companies that publish real numbers, because that's how everyone gets a clearer picture.