AI Water Use: ChatGPT Prompts Have a Hidden Cost

A new dashboard called watercost.live tracks the estimated water consumption of AI queries. Each time you use ChatGPT, roughly 5 milliliters of water reportedly evaporates to cool the servers that respond. What's interesting is this estimate is contested, but the dashboard aims to make this abstract number personal. It shows running totals and compares water usage to things like bottles or household daily use. A sibling project, aiwaterusage.com, also tracks water consumption. US data centers consumed about 449 million gallons of water per day in 2021. Hyperscale facilities can use up to 5 million gallons daily. Phoenix-area data centers currently use around 385 million gallons per year, a figure that could reach 3.7 billion with planned expansions. These trackers use modeled estimates, not live sensor readings. They multiply a fixed per-query water factor by assumed usage. While the 5ml figure is real, it's an incomplete picture, as direct cooling is only about one-fifth of the total footprint. The bottom line is these public dashboards act as tools to raise awareness about the environmental impact of AI.