Water, Not Chips, Is Blocking AI Data Centers

Published: September 10, 2026 | Category: Technology | By Mahesh

A Guardian analysis published in mid-2026 found that 517 of 809 planned U.S. data centers, roughly two-thirds, are sited in locations that experienced drought within the past year.^[1] That single geographic fact explains why water, not gas turbine backlogs or chip packaging constraints, has become the fastest-growing source of actual project cancellations and delays in the AI infrastructure buildout. Data Center Watch, an industry tracking firm, documented at least 75 projects worth a combined $130 billion disrupted by local opposition in the first quarter of 2026 alone.^[2] Depth Grid's coverage this week has walked through the grid capacity bottleneck, the gas turbine backlog, and the chip packaging shortage, three purely technical and manufacturing constraints. Water is a genuinely different kind of bottleneck: it is not primarily a supply chain problem at all, it is a local political and hydrological one, and it is already killing projects that have already cleared every other hurdle.

The Guardian Analysis That Made the Geography Impossible to Ignore

The scale of the mismatch between where AI infrastructure is being built and where water is actually available is documented with unusual precision in recent reporting. Beyond the Guardian's own two-thirds drought-exposure finding, the same report calculated that large data centers can require up to 5 million gallons of water a day for cooling, roughly equivalent to the daily water use of 50,000 people, and projected total U.S. data center water demand could climb to 73 billion gallons annually by 2028, up from about 17 billion gallons in 2023, more than a fourfold increase in five years.^[1] Bloomberg News reporting cited in a separate 2026 analysis adds a further confirming data point: about two-thirds of U.S. data centers built since 2022 sit specifically in areas already classified as high water-stress regions, a consistent finding across multiple independent data sources rather than an artifact of any single analysis's specific methodology.^[3]

Lawrence Berkeley National Laboratory, a U.S. Department of Energy national laboratory rather than an industry advocacy group or media outlet, provides the most authoritative federal research on the actual scale of current consumption. The lab's own report estimates 2023 direct water consumption by U.S. data centers, which host roughly 40 percent of the world's total data center capacity, at approximately 17.5 billion gallons, and separately projects that figure could double or even quadruple by 2028.^[3] That federal projection lines up closely with the Guardian's own independently sourced 73 billion gallon estimate for the same year, giving genuine cross-validation to a number that might otherwise be dismissed as advocacy-driven exaggeration from either side of the debate. The Berkeley Lab report also notes that on a purely national level, data center water use remains relatively modest, an important qualifier this piece returns to directly in a later section, but stresses that in the specific regions where data centers are concentrated, particularly ones already facing water shortages, the strain on local water systems can be significant.^[3]

Why Water Is a Political Bottleneck, Not Just an Engineering One

The defining characteristic that separates the water bottleneck from the power and chip constraints covered earlier this week is that water scarcity translates almost immediately into visible, organized local political opposition in a way that a distant gas turbine order backlog or an abstract chip packaging shortage simply does not. A March 2026 Gallup poll found seven out of ten Americans oppose data centers being built near them, with water use cited as a top specific concern driving that opposition.^[4] That opposition is not merely a polling abstraction, it has already translated directly into commercial outcomes: the 75 projects worth $130 billion disrupted by local opposition in the first quarter of 2026 alone, documented by Data Center Watch, represent capital that had already cleared financing, chip allocation, and power procurement, only to stall over local water and community concerns after those other, more heavily covered constraints had already been resolved.^[2]

The protest movement itself has become geographically specific and organized enough to carry recognizable slogans and named locations. Reporting from Knowable Magazine documents demonstrations with signs reading "Protect our water," "Water for people not AI," and "Don't mess with our water" appearing from Texas to New Mexico to Arizona, with specific, named local opposition emerging in communities including Tremonton, Utah.^[5] This pattern echoes what Depth Grid's earlier reporting found in the electricity context, where wholesale prices near specific data centers rose sharply enough to generate local backlash, but water carries an additional emotional and political charge that electricity price increases alone typically do not, since water is more directly and immediately understood by ordinary residents as a survival resource rather than a utility bill line item. A separate 2026 industry analysis notes this tension has already led to project delays and outright blocks worth a further $64 billion, with critics specifically pointing to what they characterize as opaque corporate water usage disclosure as compounding the community distrust driving these blocks.^[6]

Putting the Scale in Honest Context

A rigorous accounting of this issue requires acknowledging a genuine tension in the data that neither side of the current public debate fully resolves on its own. On one hand, the specific, large numbers cited throughout this piece, billions of gallons, multi-fold projected growth, are genuinely significant in the specific water-stressed regions where AI infrastructure has concentrated. On the other hand, independent analysis is careful to note that data center water consumption, at the purely national level, remains modest relative to other major water-consuming industries. One 2026 analysis states directly that agriculture dominates water consumption in many affected states, accounting for 86 percent of total usage in a state like Arizona, dwarfing the industrial category that includes data centers, estimated at around 8 percent of total use in the same state.^[7] The Knowable Magazine analysis cited above similarly notes that data center cooling systems consumed an estimated 66 billion liters of water in 2023, less than 1 percent of total U.S. water consumption nationally.^[5]

Both of these framings are simultaneously true, and the apparent contradiction resolves once the geographic concentration point made earlier in this piece is taken seriously: a resource that represents under 1 percent of national consumption can still represent a genuinely decisive, locally significant addition in a specific watershed that is already operating near its sustainable limit, particularly when that addition arrives concentrated within a handful of new, large, geographically clustered facilities rather than spread evenly across the existing water-using population. Texas offers a concrete regional illustration of this dynamic: a 2026 analysis found Texas data centers consumed over 50 billion gallons of water in 2024 alone, enough to supply a city the size of Austin for months, with 47 data centers concentrated specifically in the Central Texas region, an area already managing meaningful water stress independent of the data center buildout.^[8] Robert Mace, executive director of a water research organization quoted in that same reporting, captured the underlying framing problem directly: people don't think of data centers as industrial water users, but they are, a distinction that matters because local water policy and permitting processes were generally not designed around a rapidly growing industrial user of this specific scale and geographic concentration.^[8]

What Hyperscalers Are Actually Doing About It

The major technology companies building this infrastructure are responding with genuine, measurable operational changes rather than treating the criticism as purely a public relations problem, and the specific technical and policy responses are worth understanding in some detail. Amazon has developed a system that reduced the company's North American water use by 946 million liters in 2024 alone, described as enough drinking water for 1.3 million people for a year, while improving its overall water efficiency by 17 percent in the same period.^[2] Microsoft is separately experimenting with technology allowing AI computer chips to function reliably at higher operating temperatures, an approach the company's own vice president for data center infrastructure engineering, Steve Solomon, described directly: the warmer the temperature, the more efficient the energy use, an approach that reduces the cooling burden, and therefore the water burden, at its physical source rather than simply managing the water more efficiently after the fact.^[2]

Both Google and Amazon have adopted an explicit policy of prioritizing water-saving techniques specifically in identified drought-prone regions such as Phoenix and Cape Town, South Africa, rather than applying a single, uniform water strategy across every facility regardless of local conditions. Google's own Townsend, quoted directly in coverage of this policy, described the company's stance in unambiguous terms: in areas of high stress or scarcity, no evaporative cooling towers, it's a hard no.^[2] Amazon's own Schlitz, in the same reporting, tied the company's use of recycled water directly to business risk management rather than presenting it purely as an environmental commitment: using recycled water also limits risk for our business as well, since the company does not want to use potable water for cooling in locations where future water supply may not be reliably available, and the company has specifically helped fund water recycling infrastructure in Loudoun County, Virginia, one of the most heavily concentrated data center regions in the country.^[2]

A July 2026 report from the Information Technology and Innovation Foundation, a technology policy think tank, makes a more optimistic technical argument worth weighing directly: new cooling technologies now make it possible to consume almost zero water directly for data center cooling, at a modestly higher cost than traditional evaporative systems, and some hyperscalers have already begun adopting these zero-water designs.^[9] The same report distinguishes carefully between direct water consumption at the data center itself and indirect water consumption tied to the electricity generation that powers the facility, noting that gas-fired power generation consumes considerably more water than solar generation, meaning a data center's true water footprint depends significantly on the specific mix of power sources feeding it, not solely on the cooling technology installed on-site.^[9] Dry cooling, which the report describes as functioning much like a very large car radiator using air-cooled heat exchangers instead of evaporative towers, is identified as one of the more mature zero-water-consumption alternatives already commercially available, though the report notes it typically carries a higher capital cost and can reduce cooling efficiency in the hottest climates compared with evaporative systems.^[9]

Why This Is a State-by-State Problem With No Federal Fix

Unlike the electricity grid interconnection process, which involves federal regulatory bodies including the Federal Energy Regulatory Commission alongside state utility commissions, water policy in the United States is regulated almost entirely at the state level, a structural fact the Information Technology and Innovation Foundation's own report states directly: water consumption is regulated by the states, not the federal government, meaning any effective policy response must be driven by individual state regulation focused on the specific watershed in question rather than a single national standard.^[9] The same report notes there is no genuine national water shortage in the aggregate, but significant, real shortages exist in specific arid regions such as Arizona and in drought-stricken areas such as California, and each individual watershed carries its own distinct water capacity and its own existing base of competing users, from agriculture to municipal drinking water systems to industrial users, that a new data center project must be weighed against on a case-by-case, watershed-specific basis.^[9]

This state-by-state, watershed-by-watershed regulatory structure is precisely why the water bottleneck cannot be resolved the way federal CHIPS Act funding has attempted to address chip manufacturing capacity, or the way federal loan guarantee programs have supported specific nuclear power projects. A data center developer facing water-related permitting resistance in one Arizona watershed gains no benefit from a favorable federal water policy, because no such uniform federal water policy exists to invoke, and the same developer's next project in a different state, or even a different watershed within the same state, may face an entirely different set of local water rules, water rights holders, and community stakeholders to negotiate with from scratch. This fragmentation is a large part of why the water bottleneck, unlike the power grid and chip packaging constraints covered earlier this week, is likely to remain a persistent, unpredictable, project-by-project risk factor for the AI infrastructure buildout for the foreseeable future, rather than a constraint any single policy intervention could meaningfully resolve all at once.

What This Means for Anyone Siting or Evaluating a Project

Treat local water availability and community sentiment as a primary site-selection criterion, not a secondary permitting detail to resolve after a location is chosen. Given that 75 projects worth $130 billion were disrupted by local opposition in a single quarter after presumably already clearing power and financing hurdles, a developer or investor evaluating a specific data center project should weight local water stress data and documented community sentiment as heavily as power availability and chip supply timelines when assessing genuine project risk, rather than treating water as a formality to be handled once the more heavily covered constraints are resolved.

Favor water-abundant, lower-controversy regions when latency and other siting constraints genuinely allow it. Current research specifically names Montana, Nebraska, parts of Texas, and South Dakota as regions where strategic siting can avoid adding stress in already drought-prone states such as Arizona, New Mexico, and California, and a developer with genuine geographic flexibility should weight this water-availability factor explicitly against the latency and grid-access considerations that have historically dominated site-selection decisions.

Disclose water usage transparently and proactively, rather than defensively after opposition has already organized. Given that critics specifically cite opaque corporate water usage data as compounding local community distrust, and given that Amazon's and Google's own public, quantified water efficiency commitments appear to have helped manage this exact tension in some markets, proactive, specific, and verifiable water usage disclosure before a project faces organized local opposition is a meaningfully different, and likely more effective, community relations strategy than responding defensively once opposition has already formed.

Common Questions

This analysis is editorial commentary based on publicly available sources cited above. It is not financial, investment, engineering, or policy advice. Water consumption figures, regional projections, and regulatory details cited reflect data and reporting available as of publication and vary by source methodology; verify current figures with the cited research institutions, government agencies, and companies before making decisions based on this information.

Sources

1. ConstructConnect, "AI Data Center Boom Is Running Into America's Water Problem," citing Guardian analysis, June 12, 2026. Link
2. E&E News by POLITICO, "AI's No-Win Choice: Using Huge Amounts of Water or Energy," citing Data Center Watch and Gallup polling, July 9, 2026. Link
3. IEEE Spectrum, "AI Water Usage," citing Lawrence Berkeley National Laboratory and Bloomberg News reporting. Link
4. E&E News by POLITICO, "AI's No-Win Choice: Using Huge Amounts of Water or Energy," citing March 2026 Gallup poll, July 9, 2026. Link
5. Knowable Magazine, "How Much of a Problem Is AI's Water Use?" 2026. Link
6. AKCP, "The Data Center Water Footprint: 2026 AI Impact & Statistics," August 17, 2026. Link
7. AKCP, "The Data Center Water Footprint: 2026 AI Impact & Statistics," citing Arizona agricultural and industrial water use comparison, August 17, 2026. Link
8. Yahoo News, "AI's Thirst Trap: Data Centers Guzzle Water While Droughts Drain Communities," citing Robert Mace, executive director, water research organization. Link
9. Information Technology and Innovation Foundation (ITIF), "The Data Center Water Problem Is Soluble," July 6, 2026. Link

Read More on Depth Grid

• The Compute-and-Power Bottleneck: Why AI's Next Decade Will Be Decided by Electrons, Not Algorithms
• Order a Gas Turbine Today, Get It in 2031: The Backlog Quietly Gating Every AI Data Center
• The Real AI Chip Shortage Isn't Chips, It's a Packaging Process Called CoWoS

Article by Mahesh | Depth Grid