The fight over AI data centers is still current, but the water story needs a sharper frame. National totals make the panic look overstated, while local water stress makes some projects genuinely hard to defend.
AI data centers have become one of the hardest infrastructure projects to approve in America, and water is now at the center of the fight. The concern is not imaginary. These facilities can draw heavily from local systems, and when they are planned in drought-prone regions, the burden lands on households, farms, and utilities long before it shows up in any national spreadsheet.
That distinction matters because the debate is moving fast. As The Guardian reported this week, roughly two-thirds of 809 planned AI data centers in the United States are expected to be built in areas experiencing severe drought. The same report cited projections that U.S. data center water use could rise from about 17 billion gallons in 2023 to as much as 73 billion gallons a year by 2028. Those numbers explain why local opposition has become so intense. They do not, by themselves, prove that data centers are a national water crisis.
The national comparison cuts in the other direction. Agriculture remains the dominant water user in the United States, and thermoelectric power generation has historically accounted for a large share of withdrawals. Data centers are much smaller on that scale. That does not make their footprint irrelevant, but it does make broad claims about AI draining the country too blunt to be useful. A facility drawing from an abundant municipal supply in a wet region is not the same project as one proposed over a stressed aquifer in Arizona, Georgia, or Texas.
This is where the public argument often gets sloppy. Opponents are right to ask who bears the cost of serving massive server farms. They are on weaker ground when they use national water numbers as if every gallon comes from the same place. Water is not a national commodity in the way capital spending is. It is local, seasonal, political, and tied to infrastructure that may already be under strain.
The polling shows why the industry cannot dismiss the backlash as a niche environmental campaign. Gallup found in May 2026 that 71 percent of Americans oppose new data centers in their area, a level of resistance stronger than public opposition to nearby nuclear plants. Among opponents, resource use and environmental impact were leading concerns. That is a serious permitting problem for Microsoft, Amazon, Meta, Google, Oracle, CoreWeave, and every developer trying to turn AI demand into physical capacity.
Companies have started to respond, but unevenly. Microsoft has promoted closed-loop cooling designs for newer AI campuses, saying some can operate with sharply reduced ongoing water consumption after an initial fill. Amazon has said it is making progress toward replenishing more water than it consumes by 2030. Google has made similar stewardship pledges. These commitments matter, but they are still too often framed at the company level, when communities want facility-level answers.
That disclosure gap is the practical problem. A county board does not need a glossy sustainability target for 2030. It needs to know how much water a specific site will withdraw, what source it will use, how much is potable, how demand changes during heat waves, and what happens during drought restrictions. Without that information, opponents can fill the vacuum with worst-case assumptions, and companies are left arguing about averages that do not answer the local question.
The same pattern applies to indirect water use. A data center may use less water on site if it relies on air cooling or closed-loop systems, but the electricity it consumes may still involve water-intensive power generation somewhere else. That does not mean every project should be blocked. It means the accounting has to be honest enough to cover both direct cooling and the power supply behind it.
The current backlash also reflects a broader trust problem. Many communities see enormous buildings, limited permanent job creation, tax incentives, higher power demand, and unclear environmental reporting. Even where a project would be manageable from a water standpoint, developers are walking into hearings with a credibility deficit. Once that happens, technical corrections rarely win the room.
The better path is not to pretend the water concerns are fake. It is to separate the real local constraints from the exaggerated national rhetoric. Data centers are not the largest water user in America, and the aggregate figures do not support a blanket claim that AI infrastructure is draining the country. But in drought-stressed watersheds, a large facility can still be a serious local burden. Both things can be true.
The AI buildout is not going to stop, because the demand for compute is too large and the capital already committed is too deep. What can change is the standard for approval. The next phase of the market will reward companies that disclose water use at the watershed level, site facilities where local systems can support them, and treat community concerns as operating constraints rather than public relations problems. The companies that keep arguing from national percentages may find that the permits they need are decided by much smaller numbers.
Also read: Apollo's hunt for a second headquarters signals a permanent shift in where private equity capital calls home • Taiwan is using a $200 million data center to keep its last South American ally • Europe 2031 gives the continent until summer to avoid AI irrelevance
