AI is often described as a black box. This is typically a reference to how little even its creators understand about its inner workings. But there is another big mystery around AI: its energy use and climate impacts. Even as leading AI firms have advocated for reduced regulations and increased government funding to remove barriers between them and unprecedented levels of energy, they’ve kept their own energy figures closely guarded secrets.
MIT Technology Review cracked open this particular box in an investigation, “We did the math on AI’s energy footprint. Here’s the story you haven’t heard.”
Over six months, senior reporters Casey Crownhart and James O’Donnell teamed up to pursue the truth that the likes of OpenAI and Google had kept from the public. They traced the previously invisible currents connecting the chips that AI models run on to power plants—finally zeroing in on just how much energy (and what kind) is used when someone asks a single question to a chatbot. They spoke to dozens of experts, pored over hundreds of pages of reports, ran the numbers on different AI models and prompts, and made use of open-source intelligence on energy consumption. Their groundbreaking work helped reveal how big AI’s energy footprint is, where that energy comes from, how much it’s set to grow, and who will ultimately pay for it.
This dogged reporting enabled MIT Technology Review to calculate the energy and emissions cost of an individual text, image, or video prompt—putting this vital information on the record for the first time. These conclusions were later validated when, after publication, OpenAI and Google released energy numbers that were aligned with the project’s estimates.
To make their work as accessible as possible, they published a technical brief about their methodology and summed up their findings in an animated explainer video. Together with our Editor in Chief Mat Honan, they hosted a live online event for subscribers that dove into the details of their reporting. Nearly 1,000 subscribers registered, and the several hundred attendees were incredibly engaged in asking questions and providing feedback. Crownhart also created a social video describing the high-level takeaways from their work, which was shared with MIT Technology Review’s 1.5 million LinkedIn followers, 1.1 million Instagram followers, and 100,000 YouTube subscribers.
This investigation was part of a larger project, “Power Hungry: AI and Our Energy Future,” which also explored the real-world climate and energy consequences of the AI boom. For months following publication, Crownhart and O’Donnell reported on related developments from Google and on grid impacts in a series of news stories.
