GPU Data Centers Drive Up Energy Costs, Face Local Backlash
Get the Tech newsletter
Daily tech — startups, AI labs, chips, the launches that shape the next decade. Free.
- Microsoft's Fairwater data center in Mount Pleasant, Wisconsin — built on the failed Foxconn LCD site — is the world's most powerful AI data center, packed with Nvidia B200 Blackwell GPUs that draw up to 1,200W each at max capacity and sit in racks of 72 stacked two-high.
- Data center expansion in the US is concentrated in low-income neighborhoods and communities of color, driving up local utility bills and pollution in the same places where tech companies are courting more affluent consumers.
- The NAACP has sued xAI (now doing business as SpaceXAI) over air pollution from on-site gas generators powering its massive data centers, after warning tech companies to "be on alert" as local groups mount campaigns.
- UC Riverside's Shaolei Ren, who grew up in a coal-mining region of northern China in the 1980s, now researches data centers' impact on local air quality and water scarcity and proposes a "community-integrated data center approach" to prevent harm to nearby residents.
- GPU environmental costs span the entire lifecycle — raw material mining, toxic chemicals in semiconductor factories, water and energy use at data centers, and e-waste at end of life — and aren't unique to AI.
- Gaming and consumer GPUs raise the same energy questions: Nvidia's RTX 5090 draws 575W, Sony has sold 92 million PS5s built around an AMD GPU drawing up to 220W, and Apple's A19 chip in roughly 1.5 billion active iPhones adds up to enormous aggregate power draw.
- VCs blaming AI's slow consumer uptake on environmental bad press get pushback from University of Staffordshire ethicist Catherine Flick, who asks: "Isn't it nice to have the environment as a scapegoat?"
Why it matters: The data center boom is concentrating environmental costs in US communities — often low-income ones — even as the chips spread from 1.5 billion iPhones to 92 million PS5s to Nvidia's 1,200W B200 data center chips. Local residents absorb the utility hikes and air pollution; the AI productivity gains accrue elsewhere, and the article makes clear there is no shared metric yet for deciding which GPU workloads earn their energy footprint.
