From Texas to Arizona, residents have taken to the streets with signs that read “Protect our water” and “Water for people, not AI.” The protests reflect growing concern over the rapid expansion of artificial‑intelligence data centers and the water they may consume.
Current water use is modest
Data‑center cooling does require water, but the amount is tiny compared with the nation’s overall consumption. Researchers estimate that cooling systems for all U.S. data centers used about 66 billion liters of water in 2023—less than one percent of total national water use. By contrast, agriculture and heavy manufacturing account for the vast majority of withdrawals.
Efficiency gains are already under way
Engineers stress that newer AI‑focused facilities are far more water‑efficient than earlier designs. Cornell University energy‑systems expert Fengqi You notes that a combination of better cooling methods and water‑replenishment projects could eventually make on‑site cooling “net‑zero.”
Shaolei Ren, an electrical engineer at the University of California, Riverside, points out that early estimates—such as the claim that a single chatbot query consumes 500 milliliters of water—are outdated. AI models have become more efficient, and Google now reports that its Gemini chatbot uses only five drops of water per median‑length query.
Future projections and mitigation
Even with these efficiencies, the sheer scale of AI growth could raise water demand. You’s research group projects that by 2030 U.S. data centers might use between 731 billion and 1,125 billion liters annually—roughly the amount of water New York City consumes each year. However, those figures include water used to generate electricity, much of which comes from coal‑ and gas‑fired plants that require steam‑condensing water.
Switching to solar and wind power, which need little to no water, would dramatically lower the overall footprint. You emphasizes that locating new facilities in water‑rich, renewable‑energy‑friendly regions—such as parts of Montana, Nebraska, Texas, and South Dakota—can cut future water use by up to 86 percent.
Cooling technology evolves
Traditional data‑center cooling relied on evaporative towers that waste water. Today, many AI‑heavy facilities employ liquid‑cooling loops that circulate water (or specialized coolants) directly over processors. Because the system is closed, no water is lost; the heated fluid is simply cooled and recirculated.
When ambient temperatures soar, some centers add misting to boost heat removal, but even that is limited. In Texas, mechanical engineer Vaibhav Bahadur notes that most AI data centers use misting only during the hottest summer days. He estimates that Texas data centers currently consume less than 1 percent of the state’s water, a share that could rise to 3‑9 percent by 2040 if unchecked. He expects efficiency improvements to keep the actual figure lower.
Companies are also testing immersion cooling, where servers sit in a tank of non‑conductive liquid, and ocean‑based installations that use seawater for heat exchange. NVIDIA’s sustainability head, Josh Parker, explains that their latest processors run hot enough to be cooled by air alone in most conditions, reducing the need for water‑based methods.
Industry commitments
Major cloud providers have set ambitious water‑positive goals. Amazon Web Services aims to return more water to communities than its operations use by 2030 and reports being more than halfway there as of 2024. Google pledges to replenish 120 percent of the freshwater its data centers consume by the same year. OpenAI also outlines a water‑stewardship plan on its public website.
While local activists remain vigilant, the data suggest that AI’s water impact is manageable and improving. Continued investment in renewable energy, smarter siting, and advanced cooling will help ensure that the nation’s water resources stay protected even as AI capabilities expand.
Original reporting: KTVZ (Central Oregon) — read the source article.
