The hidden cost of AI's power hunger: from data centers to your neighborhood

Graphics processing units—from massive AI chips in data centers to the GPUs in your iPhone and gaming console—have quietly become central to nearly every computation we perform. Yet few people understand the true environmental toll these chips exact, especially now that artificial intelligence has dramatically amplified their power demands and put them squarely in the crosshairs of environmental scrutiny.

Nvidia's rise to become the world's most valuable company underscores how thoroughly GPUs now permeate technology. But behind the scenes, a much larger conversation is brewing about whether the promise of AI innovation justifies the real costs being imposed on communities where data centers operate.

The scale of AI's energy appetite#

A single Nvidia B200 Blackwell chip designed for AI workloads can draw 1,200 watts at full capacity. Microsoft's Fairwater data center in Wisconsin—built on a failed manufacturing site—stacks thousands of these chips into massive supercomputers. When you multiply that across the hundreds of thousands of GPUs now being deployed globally, the energy requirement becomes staggering.

Research from Lawrence Berkeley National Laboratory found that GPU-powered AI servers consumed roughly 40 terawatt-hours annually in the US as of 2023, up from just 2 terawatt-hours in 2017. Projections suggest this could balloon to between 165 and 326 terawatt-hours by 2028—roughly equivalent to what more than 8.7 million American homes consume in a year.

To put that in perspective, training a single large model like Meta's Llama 3.1 can generate as much air pollution as 10,000 round-trip car journeys between New York and Los Angeles. A 2024 study from UC Riverside and Caltech researchers estimated this could translate into more than $20 billion in public health costs by 2028 and around 1,300 premature deaths annually from air pollution by 2030.

Why the scrutiny feels uneven#

Venture capitalists and tech executives have begun complaining that AI is being unfairly singled out for environmental criticism, especially compared to industries like fast fashion. The argument goes: if consumers willingly buy cheap clothes despite environmental concerns, why shouldn't AI get similar latitude?

But consumers aren't seeing the promised payoff. Unlike cheap clothing—where customers perceive immediate value—most people haven't experienced the transformative AI benefits that tech leaders keep promoting. Meanwhile, the environmental costs are becoming impossible to ignore, particularly because data centers are no longer being built exclusively in distant countries. They're now sprouting across the United States, raising local energy prices and creating air quality problems in communities that never consented to host them.