Water Consumption of African Data Centers from AI

Carnegie Mellon University Africa researchers published a study estimating water usage efficiency (WUE) for data centers across 41 African countries and presented the work at the 2025 ACM COMPASS conference.

• Main action: The research team modeled WUE and estimated AI model water consumption both onsite (direct cooling) and offsite (indirect via electricity generation) across 41 African countries and five climate regions; key numeric findings include Llama-3-70B: 0.7 liters for a 10-page report and GPT-4: up to 60 liters for the same task, with 8 of 11 sampled African countries using less water than the global average and Morocco and South Africa using less than the U.S. average.
• Background & details: The paper identifies five climate regions (rainforest, savanna, desert, steppe, Mediterranean), notes high water transmission leakage as a major contributor to water costs, analyzes onsite vs offsite water sources (e.g., hydroelectric or thermo-electric offsite impacts), lists co-authors Pengfei Li and Shaolei Ren (UC Riverside), and was presented at ACM COMPASS 2025; media inquiries contact: Hannah Diorio-Toth (hdiorio@andrew.cmu.edu).