GPU-Driven AI's Environmental Cost: Water, Pollution, and Public Health Risks

The article notes that GPUs are deployed globally in data centers at unprecedented rates, consuming enormous energy and water. Training Meta's Llama 3.1 model produced air pollution equivalent to 10,000 round-trip car trips from New York to Los Angeles. Researchers estimate AI-related public health costs could exceed $20 billion by 2028, with 1,300 premature deaths annually from air pollution. By 2025, AI's electricity consumption may surpass Bitcoin mining, accounting for nearly half of global data center electricity use. For communities near U.S. data centers, these facilities bring tangible pollution and noise. The NAACP has sued xAI over air pollution. For developers using AI services, these externalities are not yet priced. Researcher Shaolei Ren proposes 'community-integrated data centers' to reduce resource consumption and pollution through optimized operations. But in the short term, the GPU-driven AI ecosystem may face increasing public and regulatory pressure.