As a climate scientist, Zeke Hausfather uses artificial intelligence every day for research and data analysis. But he kept wondering how much energy all that AI use was consuming. So over eight weeks this summer, he tracked the 1,138 prompts he typed into Claude Code and estimated how much electricity they consumed at data centers.
Tech companies have shared little data about the energy use of their AI models. One of the few public estimates comes from Google, which said last year that a typical text prompt on Gemini used just 0.24 watt-hours, less than watching TV for nine seconds. OpenAI’s CEO, Sam Altman, has said a ChatGPT query uses a similar amount, around 0.34 watt-hours. But Hausfather suspected those figures no longer reflected how people are actually using AI.
“In the AI world, a year is an eternity,” Hausfather says.
A year ago, most AI use was still simple conversations in chatbots. Now, agentic AI is increasingly dominant, “where you give a set of instructions to an agent to go spin up 20 different subagents, and do some massive process that is taking orders of magnitude more energy use,” he says. “I thought it would be interesting to use the actual history of my agent to get at what the impact of this is.”
He calculated that his median prompt used around 150 kilowatt-hour of energy—or 600 times more than a chatbot query.
That’s still a very rough estimate, based in part on looking at usage of tokens, the chunks of text that AI models process. “We can roughly assume that there’s a correlation between tokens and how they’re priced and the energy use of the system, because energy is a big part of the cost of running in terms for these models,” Hausfather says. “But it’s certainly not perfect.”
On one hand, that’s still a relatively small amount of electricity; Hausfather’s own heavy AI use amounted to a little more than running an electric dryer for a year. But in aggregate, AI is becoming a major new source of energy demand. By one estimate, AI data centers could consume a staggering 12% of all U.S. electricity by 2030.
At the moment, many tech companies are turning to fossil fuels to meet the energy demands of new data centers. Meta, for example, is paying to build seven natural gas plants for a massive data center in Louisiana.
But that’s a choice. Hausfather argues that the AI boom could instead help accelerate the transition to clean energy, if tech companies brought the same money and urgency to building a cleaner, more capable grid.
“If we’re going to reduce the impact of our energy use, we need to electrify almost everything,” he says. “We need to triple the size of our grid and our electricity generation in order to electrify vehicles, electrify home heating with heat pumps—all these other parts of the economy that we burn fossil fuels for now. That’s going to create a huge challenge to be able to build so much energy so quickly. With AI, in some ways, we’re speedrunning a little bit of the challenge that we’re going to face as we try to dramatically expand our grid to meet our climate goals.”
Hausfather doesn’t argue that we should necessarily feel guilty about using AI. As a scientist, he uses it for everything from analysis to quickly creating new tools like a map showing how much every point on the planet has heated up since the Industrial Revolution. But AI, like everything else, needs to run on the right kind of energy.
“Longer term, our goal is to decouple energy use and emissions,” he says. “Using energy itself is not inherently bad. How that energy is produced is what matters.”