This article was compiled by the editorial desk based on the findings published in the journal Joule and statements reported by the New York Times, with additional context from public regulatory announcements.

The rapid expansion of artificial intelligence is driving a surge in electricity consumption that, by 2027, could rival the annual power use of entire nations. A new analysis published in the journal Joule estimates that the server farms underpinning AI systems may consume between 85 and 134 terawatt hours of energy each year, a range comparable to the electricity demands of Argentina, the Netherlands, or Sweden.

Alex de Vries, a data scientist at Vrije Universiteit Amsterdam who conducted the study, based his projections on sales of Nvidia A100 servers, which account for an estimated 95 percent of the AI industry's computing infrastructure. Each of these units, de Vries told the New York Times, are “power-hungry beasts.” The figures represent about 0.5 percent of global electricity demand, a share that some experts argue warrants a reassessment of the pace and scale of AI investment.

The energy footprint of AI has drawn comparisons to the cryptocurrency mining boom, which similarly saw rapid spikes in power consumption. But the scale of AI's growth, driven by products like OpenAI's ChatGPT and Google's Bard, has intensified concerns among researchers and environmental advocates.

Why the Numbers Matter

The study underscores a growing tension between technological advancement and environmental sustainability. While AI companies are often secretive about their energy usage, de Vries's methodology offers a transparent way to estimate consumption based on hardware sales. The findings suggest that without significant improvements in efficiency or a shift to renewable energy sources, the carbon footprint of AI could become a major obstacle to climate goals.

Roberto Verdecchia, an assistant professor at the University of Florence, told the New York Times that the industry should pause and consider the environmental resources being consumed. “Maybe we need to ideally slow a bit down to start applying solutions that we have,” he said. “Let’s not make a new model to improve only its accuracy and speed. But also, let’s take a big breath and look at how much are we burning in terms of environmental resources.”

Regulatory pressure is beginning to mount. Over the weekend, California Governor Gavin Newsom signed two climate disclosure laws that will require companies—including OpenAI, Google, and roughly 10,000 other firms—to report their carbon emissions starting in 2026. These measures aim to increase transparency and could force tech giants to confront the environmental costs of their operations.

Despite the looming regulations, the AI industry remains largely self-regulated. Companies continue to invest heavily in expanding their models, with little public accountability for the energy required to train and run them. However, the financial burden of high electricity costs may provide an incentive for innovation in energy-efficient computing.

The study's authors and other experts emphasize that the window for action is narrow. As AI becomes more embedded in daily life, the demand for computing power is likely to grow, making it imperative to develop sustainable solutions before the environmental impact becomes irreversible.