Most new U.S. AI data centers are being built in drought zones

Recent analysis reveals that approximately two-thirds of the 809 planned AI data centers in the U.S. are being built in areas affected by drought over the past year. This trend raises concerns about w... Read original →

The concentration of U.S. AI data centers in drought-stricken regions reveals a fundamental mismatch between computational expansion and ecological carrying capacity. Technologically, sub-3nm nodes using EUV lithography consume over 2 tons of water per wafer, while liquid-cooled systems like GB200 address only 4% of the water footprint—the real strain lies in power generation and semiconductor fabrication. Regulatory efforts targeting cooling water are superficial; they ignore the hidden water intensity of fabs (e.g., TSMC in Taiwan, China) and thermal power plants, escalating ESG compliance costs and supply chain volatility. In response, Microsoft and Google may fast-track green power and desalination partnerships, while NVIDIA leverages platforms like Vera Rubin to boost algorithmic efficiency. Within 18 months, water access will become the second critical siting constraint after power availability, forcing co-design across chip architecture, cooling tech, and regional policy—or risking deployment delays from local water restrictions.