Particle.news
Download on the App Store

AI Buildout Hits a Power Wall as New Study Tallies Its Carbon and Water Costs

A peer‑reviewed Cornell analysis puts hard numbers on U.S. AI server emissions plus water demand by 2030.

Overview

  • Researchers project 24–44 million metric tons of CO2 per year and 731–1,125 million cubic meters of water use from U.S. AI servers by 2030, with siting, grid decarbonization and efficiency offering cuts of roughly 73% and 86% respectively.
  • Two completed data centers in Santa Clara totaling nearly 100 megawatts are sitting idle for lack of electricity, with the city utility pursuing $450 million in upgrades targeted for completion in 2028.
  • Microsoft’s Satya Nadella says power access is now the chief constraint, as hyperscalers move past chip shortages and plan roughly $400 billion in 2025 infrastructure spending.
  • Utilities are turning to fast‑deploy options such as natural gas and backup generators, with some coal retirements delayed and a Georgia utility seeking approval for about 10 gigawatts of new gas‑fired generation.
  • Analysts warn the U.S. could lose ground to countries adding generation at far faster rates, highlighting permitting delays, long transmission timelines and community opposition that has stalled tens of billions of dollars in projects.