Water Runs Through Everything We Build
March 22 was World Water Day, and I want to use it to talk about something that matters to me as someone who works at the intersection of making things and understanding systems: where water actually goes, and what that has to do with the tools many of us have started relying on.
This isn’t an alarm post. I’m not here to tell you that AI is draining the planet or that you should feel guilty for asking ChatGPT a question. I wrote a fully sourced, 26-page white paper on this topic last year (“Thirsty Intelligence: What AI Costs and What It Could Save,” which you can read at bit.ly/CostOfAI), and the conclusion I came to then is the same one I hold now: AI is a tool. It reflects the values and decisions of the people who build and deploy it. The environmental impact is real, and it deserves scrutiny, and scrutiny is different from panic.
So let’s look at what’s actually happening.
The water most people never see
The average American family uses about 300 gallons of water a day indoors. Toilets, showers, faucets, and laundry account for most of it. Leaking pipes alone waste around 12% of that total, which adds up to roughly 10,000 gallons per household per year. That’s the water you can see and feel.
Then there’s the water you can’t. Every product has a water cost embedded in its production. A pair of jeans takes about 2,000 gallons. A smartphone, over 3,000. A single hamburger with beef runs about 660 gallons when you count the water to grow the feed, support the animal, and process the meat. Agriculture alone claims about 72% of all global freshwater withdrawals.
Data centers are part of this picture, too. They use water for cooling (either directly through evaporative towers or indirectly through the power plants that supply their electricity).
AI workloads, because they’re computationally intensive, tend to run hotter, which means more cooling.
The numbers vary enormously depending on the facility, the cooling system, and the energy source. Google published a technical report in 2025 showing that a median Gemini text prompt uses about 0.26 milliliters of water, roughly five drops.
Independent researchers have estimated that a more complex query on a less efficient system could use significantly more. The gap between those numbers tells you something important: infrastructure decisions matter as much as the technology itself.
A distinction worth understanding: withdrawal vs. consumption
One of the things I try to do in my work is give people the vocabulary to evaluate claims for themselves. When it comes to water, the most useful distinction is between withdrawal and consumption.
And this distinction matters for evaluating any industry, not just tech.
Withdrawal means borrowing water from a source. A power plant might pull 10 million gallons from a river, run it through its cooling system, and return 9.5 million gallons. That water isn’t gone, but pulling too much at once can still stress the local ecosystem.
Consumption means the water doesn’t come back. It evaporated, it was incorporated into a product, or it was contaminated beyond safe return.
When you hear a number about how much water a data center “uses,” ask which kind of use they mean. A facility running evaporative cooling towers is consuming a certain amount of water (it’s lost to the atmosphere).
A facility with a closed-loop system is mostly withdrawing it (the same water circulates). Although it’s worth noting that closed-loop systems aren’t entirely zero-impact either, since they require chemical treatments to prevent bacterial growth and mineral buildup, and those chemicals have their own environmental considerations.
What’s happening right now in legislation
Here’s where it gets actionable, and where I think the World Water Day conversation gets most useful.
The Water Resources Development Act of 2026 is actively moving through both the House and Senate. WRDA is the primary bill that decides which water infrastructure projects get federal funding. It’s been enacted on a bipartisan basis in each of the last five Congresses. This year’s version will shape decisions about lead pipe replacement, PFAS remediation, flood management, and green infrastructure like permeable pavement that recharges local aquifers. If you care about clean water in your community, this is a bill worth contacting your representatives about.
On the data center side, the picture is even more active. At the federal level, the Data Center Transparency Act (H.R. 6984), introduced by Rep. Rob Menendez of New Jersey, would require the EPA to submit quarterly reports to Congress on data center water and energy use, pollution, and impacts on local water supplies.
At the state level, the momentum is remarkable. At least 12 states have introduced moratorium bills that would pause new data center construction while communities and regulators study the impacts. Over 60 local moratorium actions have been introduced or adopted across dozens of towns and counties. And here in Illinois, the POWER Act would require data centers to pay for their own energy, source it from renewables, and manage water use responsibly. As of this week, the bill is still in committee with active negotiations.
This isn’t fringe activity. This is the political system starting to set guardrails, the way it eventually does with any large-scale industrial use of shared resources.
What you can do
I built a free template kit with two customizable letters you can send to your legislators: one about WRDA 2026, one about data center transparency. You can download it. Personalize the brackets, add a sentence about a local water issue, and send it. Handwritten or individually composed letters carry more weight than form letters.
Â
Beyond advocacy, there are practical choices worth making. Fix your leaks. A toilet flapper that’s silently running can waste 200 gallons a day. Match the AI tool to the task. If you’re drafting a quick email, you don’t need the largest, most resource-intensive model available. Write from scratch (still a perfectly valid option). And if you feel you must use one for brevity or complexity’s sake, smaller models exist for a reason, and they use less compute, less energy, and less water. Extend the life of your devices. E-waste is a massive contributor to resource extraction and pollution, and keeping your phone or laptop an extra year or two matters.
And if you want the full context on AI’s environmental footprint, including how it compares to agriculture, transportation, and crypto mining, read the white paper. I wrote it to give people the facts, the comparisons, and the nuance in one place: bit.ly/CostOfAI.
The posture that matters
I come back to something I wrote in that paper: AI is not a moral actor, but it reflects and amplifies the values of whoever holds the reins. The same technology that consumes water can be used to conserve it. AI-driven irrigation has cut water use by roughly 25% in early case studies while maintaining crop yields. AI is helping detect deforestation from satellite data in near-real-time. The question was never whether the technology is good or bad. The question is who’s making the decisions about how it gets deployed, and whether the rest of us have enough transparency to hold those decisions accountable.
World Water Day is a good day to remember that water connects everything. The coffee you drank this morning. The clothes you’re wearing. The data center helping you book your flight for summer vacation. It’s all drawing from the same finite supply. The goal isn’t to stop using things. It’s to understand what we’re using, to make sure the people using the most are being honest about it, and to ensure they’re actively working to reduce their consumption.
Choose wisely. Act kindly. And if your representatives haven’t heard from you yet about water, today’s a good day.