AI’s Power Hunger Has a Price—Just Not a Ruinous One Given the Upsides

Artificial intelligence is ravenous for electricity, so much so that AI companies measure data-center growth in megawatts, the way manufacturers once counted factory floor space. A new paper by Nicholas Z. Muller of Carnegie Mellon University tries to put a price on that consumption and its consequences. The economist finds that US data centers now use about five to six percent of the country’s power—about 250 TWh (terawatt hours) for those of you keeping score at home—and the pollution tied to that electricity adds up to roughly $25 billion a year. That latter number, in particular, might seem sizable. And it’s not nothing. But it shouldn’t dominate the debate about the expansion of these windowless, warehouse-sized supercomputers at the heart of the AI revolution. Muller’s approach is straightforward enough. He does the math, adding up how much electricity each data center uses, identifies the kinds of power plants supplying it, estimates the pollution those plants produce, and then translates that pollution into health and climate damages in dollar terms. He then compares those damages to the economic output from the data processing and hosting sector, offering a rough sense of scale rather than a down-the-penny, full cost-benefit calculation. Perhaps to the surprise of some ardent data-center opponents, there are indeed benefits as well as costs! The damages come to about five percent of the economic output of the sector—again, roughly $25 billion (with a plausible range of $10 billion to $33 billion depending on modeling assumptions) against more than $500 billion in output—suggesting the industry produces far more value than harm.  Zoom out further, and the picture improves further. The paper finds that if AI raises the level of GDP by a modest 0.5 percent, the associated environmental costs amount to just 2.4 percent of those gains. And if AI’s impact is larger—say, a one percent increase in GDP—the costs fall to roughly one percent of the upside. Back-of-the-envelope comparisons? Sure, but they point in the same direction. Of course, these are big-picture numbers. The pollution is not spread evenly across the country. Texas and Virginia alone account for about 30 percent of the total, and in a handful of places, the environmental damage actually exceeds the local economic benefit from data centers. Companies build where electricity is cheap, and cheap electricity is sometimes dirty electricity. The right response is to clean up the power AI runs on. The paper shows that environmental costs depend heavily on how electricity is generated, with damages varying significantly depending on whether power comes from coal, natural gas, or cleaner sources like nuclear and renewables. Rising data-center demand is also putting pressure on power systems in some regions, potentially accelerating investment in generation and grid infrastructure. Those pressures, if paired with cleaner energy sources—solar, wind, nuclear, geothermal—could actually reduce the environmental footprint of AI while expanding the capacity of the broader energy system. AI’s electricity use creates genuine costs. But trade-offs are part of life. And in this case, they’re just not the kind or magnitude that equates tocrisis—or even close to one. A rich and technologically advanced economy like America’s can handle harder problems than this one to maximize the upside and minimize the downside. The post AI’s Power Hunger Has a Price—Just Not a Ruinous One Given the Upsides appeared first on American Enterprise Institute - AEI .