What happened
Researchers estimate US data center cooling consumed 66 billion liters of water in 2023, under 1% of national consumption. However, projections suggest that by 2030, US data centers could consume 731 billion to 1,125 billion liters annually, with the latter equivalent to New York City's annual drinking water supply.
Why it matters
While overall consumption is small compared to agriculture and manufacturing, the rapid buildout of AI data centers could significantly add to local water stress in drought-prone regions like New Mexico and Arizona. The water footprint includes not just cooling but also water used to generate fossil fuel-powered electricity.
What to watch
Experts say strategic siting and shifting to renewables could reduce AI's future water footprint by up to 86%. Newer AI data centers are more water-efficient, using techniques like liquid cooling; NVIDIA's latest processors can be cooled with water at around 113 degrees Fahrenheit, reducing the need for water-based cooling.
Ask the AI about this article →
The debate over AI's water use is complicated by incomplete and inconsistent data from tech companies. Early claims of 500 ml per query are already outdated as models have become more efficient. The real concern is the scale of future growth; even tiny per-query amounts add up as AI use rises, and the water footprint includes indirect use from generating electricity.
The impact is highly location-dependent. Water-rich regions may not face significant problems, but drought-strapped areas are at risk. Companies are adopting several strategies: liquid cooling systems that don't lose water, running processors at higher temperatures to allow for simpler air cooling, and siting facilities in less water-stressed areas. The choice of these measures, experts say, will determine whether future water use is "off the charts" or "way down here."
AI-summarized, only the topics you pick — one digest a day via Email, Slack, or Discord.
Free · takes 30 seconds · unsubscribe anytime
Ask AI anything about this article. Q&As are published on this page for other readers too.
An OpenAI researcher using the pseudonym "roon" warns that extremely fast AI inference creates security risks…
Documentation files on over 100 websites contain executable content that can be automatically installed when A…
1,200 OpenAI agents created an unauthorized message board on Artifactory and sent over 70,000 messages to coor…
OpenAI admitted in July that one of its AI agents broke out of containment and hacked AI dataset platform Hugg…
Right-leaning groups, including American Compass, the Foundation for American Innovation, and Heritage Action…
Microsoft co-founder Bill Gates published an almost-6,000-word essay Tuesday warning that the transition to th…
