It’s a question that’s starting to bubble up, and for good reason: how much water does artificial intelligence actually guzzle? Every time you ask ChatGPT to draft an email, or Claude to help you brainstorm, water is being used. And for many, that’s a cause for real concern. A recent survey from the University of Chicago found that a significant chunk of Americans are deeply worried about AI's environmental footprint, with data centers popping up in communities often facing local opposition.
At the heart of these worries lies AI's thirst. Pinpointing the exact amount, however, is trickier than you might think. You'll hear vastly different figures. On one hand, a Morgan Stanley report projects that AI data centers could consume a staggering 1,068 billion liters of water globally by 2028 – that's an eleven-fold increase from previous estimates. To put that in perspective, the average American uses around 243,174 liters a year. Yet, OpenAI's CEO, Sam Altman, has suggested that each ChatGPT query uses a mere "about 0.000085 gallons of water; roughly one fifteenth of a teaspoon."
So, how can both figures be out there? As science YouTuber Hank Green pointed out, the difference often comes down to when you start measuring. The smaller figure, like Altman's, typically counts only the water used during the immediate query. The larger, more encompassing figures, however, account for the entire lifecycle – the massive infrastructure, the manufacturing, and crucially, the cooling.
Why does AI need water in the first place? AI computations happen in data centers, vast buildings packed with powerful processors (GPUs). These chips work incredibly hard, generating immense heat. To prevent them from overheating and failing, these centers need robust cooling systems. Traditional air conditioning just doesn't cut it for the intense heat generated by AI hardware. Liquid cooling, whether through evaporative systems or more direct methods, is far more effective at dissipating this heat.
Evaporative cooling towers, while efficient in heat removal, can be water-intensive, literally evaporating large volumes. On the other hand, dry coolers use less water but require more energy to compensate. For cutting-edge AI, direct-to-chip cooling solutions are becoming essential, bringing coolant right to the processor. Some systems even immerse entire servers in special dielectric fluids. As the demand for more powerful AI processors, like those from Nvidia, continues to surge, so does the need for these advanced, and often water-reliant, cooling technologies.
Industry leaders are actively working on these challenges, with significant investment and innovation focused on developing more sustainable cooling methods. The race is on to balance the immense computational power of AI with its environmental impact, and water usage is a critical piece of that puzzle. While the exact numbers can be debated based on what’s included in the calculation, one thing is clear: the digital world has a very real, and growing, physical footprint.
