The Numbers
Reliable information on how much energy each AI model requires is hard to come by and often outdated, and big AI companies often aren't willing to provide estimates. Past estimates by independent researchers have gone up to around 1.86 watt-hours per text response depending on query complexity and AI model quality. This number increases significantly for image and video generation.
Google's Estimate
In August of 2025, Google was the first of the large AI companies to provide an educated estimate based on their models' functionality. They calculated that the average Gemini AI text prompt required about 0.24 watt-hours of energy: enough to ride an e-bike a little over 100 feet. This was significantly lower than many previous estimates. This number, Google also revealed, has fallen significantly and will continue to do so.
MIT's Response
MIT and numerous other researchers were thrilled by the new information this estimate offered, but pointed out several potential problems that still remain. The 0.24 watt-hour estimate only represented text-only prompts, not image and video generation. Also, and perhaps most significantly, Google still hasn't disclosed the total number of Gemini queries they receive every day.
Trump Administration's 2025 report titled "Winning the Race: America's AI Action Plan"
We need to build and maintain vast AI infrastructure and the energy to power it. To do that, we will continue to reject radical climate dogma and bureaucratic red tape, as the Administration has done since Inauguration Day. Simply put, we need to “Build, Baby, Build!”
Trump Administration's 2025 report titled "Winning the Race: America's AI Action Plan"
Trump Administration's 2025 report titled "Winning the Race: America's AI Action Plan"
AI Energy Use
AI computers need vastly different hardware than traditional computers in order to operate, so big tech companies have begun to invest billions in the construction of enormous data centers specifically designed for AI infrastructure. This "arms race" toward the most sophisticated language model has resulted in an almost complete disregard for the environmental impacts of these new data centers—and the majority of AI companies have not been very transparent about the amount of energy their models consume per query, forcing climate scientists to take their best guesses.
AI Energy Use
AI computers need vastly different hardware than traditional computers in order to operate, so big tech companies have begun to invest billions in the construction of enormous data centers specifically designed for AI infrastructure. This "arms race" toward the most sophisticated language model has resulted in an almost complete disregard for the environmental impacts of these new data centers—and the majority of AI companies have not been very transparent about the amount of energy their models consume per query, forcing climate scientists to take their best guesses.
Environmental Impacts of Data Farms
Before AI, big tech companies were hopeful about making all their data centers 100% renewable within the decade. However, supporting the new data centers requires enormous amounts of energy, and several fossil fuel plants slated to be taken offline were left online in order to provide them with the necessary electricity.
The high computing power requirements of the data centers generate a lot of heat, and water is necessary to cool down the computers. Many of the data centers are built in arid areas where water is scarce to begin with.
The production of a single 2kg computer uses 800kg of raw materials, and often requires rare earth metals. The mining of these materials has an enormous environmental impact.
The data center itself has an enormous land footprint as well. Meta, a prominent AI company, is planning a 4 million square foot data center—an area roughly equal to the size of 70 football fields. Although areas selected are generally uninhabited by humans, any development of this magnitude causes some measure of habitat destruction.
We acknowledge that AI has its uses, and is already making breakthroughs in science that haven't been made in centuries of human science, in fields from neurodegenerative disease to quantum physics. In addition, AI models are rapidly becoming more and more efficient, consuming less energy per query than ever before. However, the consequences of everyday AI use becoming the norm should be examined, as this is an area in which most countries' governments are extremely lax.
