AI’s convenience now has a physical footprint
For millions of users, an AI prompt feels weightless. A question is typed, a response appears, and the exchange seems to happen entirely in the cloud. But the cloud is not a metaphor anymore; it is a fast-expanding network of land, buildings, servers, electricity lines, cooling systems and water pipes.
That reality is becoming harder to ignore. A new Guardian analysis published on June 8, 2026, found that about two-thirds of planned data centres in the United States are set to be built in areas that have experienced drought conditions over the past year. Of 809 planned data centres, 517 are in drought-affected locations, according to the analysis.
The issue is not that a single AI prompt “drinks” water in a direct, visible way. The concern is that prompts trigger computation, computation produces heat, heat requires cooling, and cooling often relies on water—either directly inside data centres or indirectly through the power plants that supply them.
“The public debate still often treats AI as software, but AI is also physical infrastructure: data centres, electricity generation, cooling systems, transmission networks, chips, minerals, land and water.”
— Kaveh Madani, United Nations University Institute for Water, Environment and Health, quoted by Reuters.
How much water does AI really use?
The exact number is difficult to pin down because water usage varies by model, location, cooling design, energy source, time of day and data-centre efficiency. Some estimates suggest a 100-word AI prompt may be associated with roughly a 500 ml bottle of water, mainly due to cooling needs, though such figures should be treated as approximate rather than universal.
What is clearer is the scale of the infrastructure. Brookings notes that a typical data centre can use about 300,000 gallons of water per day, while very large facilities may use up to 5 million gallons daily—roughly comparable to the water needs of a town of up to 50,000 people.
This is why the debate has shifted from “Does AI use water?” to “Where is the water coming from, who else needs it, and who gets priority during shortage?”
The drought problem: data centres are moving into dry regions
The most politically sensitive part of the AI water story is geography. Data centres are often attracted to places with cheaper land, tax incentives, easier construction conditions and access to power. Some of those places are also water-stressed.
The Guardian’s June 2026 analysis reported that more than 60% of the contiguous United States was experiencing varying stages of drought, while the majority of upcoming data-centre projects were planned in drought-affected locations. Large data centres may require millions of gallons of water per day for cooling, and total U.S. data-centre water demand is projected to rise sharply—from about 17 billion gallons in 2023 to as much as 73 billion gallons annually by 2028.
The tension is already surfacing in rural and agricultural communities. Farmers, ranchers and local residents are increasingly asking why new AI infrastructure should receive large water allocations while existing users are being told to conserve.
“When we get into a situation where there’s a limited amount of water available, are we going to limit water to residents and businesses before datacenters?”
— Christopher Dalbom, water resources law expert at Tulane University, quoted by The Guardian.
The global numbers are now too large to dismiss
A United Nations University report, covered by AP and Reuters, warns that the environmental footprint of data centres is now approaching the scale of major national economies. AP reported that global data centres used 448 trillion watt-hours of electricity last year and that producing that energy consumed about 1.2 trillion gallons, or 4.5 trillion litres, of water. By 2030, data-centre electricity use could nearly double, with AI’s share of data-centre energy demand rising from around 20% today to 40%.
Reuters reported that data-centre water consumption is expected to reach 9.3 trillion litres by 2030, while annual power consumption could rise to 945 TWh—roughly comparable to Japan’s electricity usage.
These numbers matter because AI is moving from occasional use to embedded infrastructure. Search, office tools, coding platforms, customer support, education, healthcare workflows, banking operations and entertainment products are all being rebuilt around AI. That means the environmental cost is no longer limited to model training; it increasingly comes from day-to-day usage.
Big Tech says cooling is improving—but trade-offs remain
Technology companies argue that the industry is becoming more efficient. Closed-loop liquid cooling, air cooling, immersion cooling, better chip design and renewable power sourcing can all reduce water intensity. Brookings notes that closed-loop cooling systems can reduce freshwater use significantly, while other cooling methods can also limit water demand.
Microsoft has recently highlighted newer AI data-centre designs that use closed-loop systems and sharply reduce direct water consumption. But critics point out that many existing facilities still rely on older cooling designs, and a low-water cooling system may sometimes require more electricity. If that electricity comes from fossil-fuel power plants, the water burden may simply shift from the data centre to the power-generation system.
That is the uncomfortable truth: AI’s water cost is not just inside the data-centre fence. It also includes electricity generation, semiconductor manufacturing and hardware supply chains.
Regulators are beginning to respond
The European Union is now considering minimum energy-efficiency standards and sustainability labels for data centres. Reuters reported that the EU proposal would include public disclosure of metrics such as water consumption and clean-energy supply for large facilities.
This marks a significant shift. For years, the AI race was measured mainly in model capability, speed, cost and market share. Now, regulators are beginning to ask for a different scoreboard: how much power, water, land and carbon are being consumed to produce those digital gains?
The user’s role: small, but not meaningless
No ordinary user should feel guilty for every useful AI prompt. AI can improve productivity, education, accessibility, scientific research and public services. The problem is not individual curiosity; the problem is invisible scale without transparent accounting.
Still, usage habits matter when multiplied across billions of interactions. AP reported that the UN study found more concise AI queries can reduce energy demand, with one estimate suggesting that cutting request length by 30% may reduce energy use by 25%.
For users, the practical takeaway is simple: use AI thoughtfully. Ask clear questions. Avoid unnecessary repeated prompts. Use smaller or simpler tools when the task does not require a powerful model. But the bigger responsibility sits with companies and governments: disclose water use, avoid building in severely water-stressed regions, invest in low-water cooling, use cleaner energy, and involve local communities before permits are granted.
The bottom line
AI is not just software. It is a resource-consuming industrial system wrapped in a friendly interface. Every prompt may feel instant and invisible, but behind it is a chain of servers, chips, cooling systems, electricity and water.
The central question is not whether AI should grow. It almost certainly will. The real question is whether its growth will be planned responsibly—or whether the world will discover too late that the intelligence boom was built on resources communities could not spare.
