Despite efficiency gains in other areas, the exponentially growing demand for AI has led to an increase in the water footprint.
Driven in part by the growth in AI, Google’s scope-1 onsite water consumption (cooling) in 2022 increased by 20% compared to 2021, and Microsoft saw a 34% increase over the same period. The majority of water consumption for server cooling is derived from potable sources.
Additionally, Scope-2 offsite water consumption (electricity generation) is experiencing growth because of AI’s enormous use of computing power. Scope-3 water consumption within the AI supply chain is also mounting. For instance, the fabrication of a single microchip necessitates approximately 2,200 gallons of ultra-pure water. The staggering demand for AI processing necessitates the utilization of thousands of these chips.
The escalating water usage associated with AI is a matter of significant concern. We need to address water scarcity, a pressing global concern exacerbated by a rapidly expanding population, dwindling water resources, and ageing water infrastructures. This concern extends beyond the mere magnitude of AI models’ water usage, underscoring the imperative for developers to proactively address the shared global challenge of water scarcity.
Shaolei Ren elucidated this challenge in: ‘How much water does AI consume?’. See here.
I have earlier mentioned this issue in ‘The ecological footprint of Artificial Intelligence’.
Published in a LinkedIn post, December 2024.