As of September 12, 2026, the U.S. is facing an unusual resource conflict: the infrastructure being built to power the artificial-intelligence boom is increasingly drawing on the same water and electricity systems that farmers depend on to produce food. The conflict is not as simple as data centers literally taking irrigation water from wheat fields. In most places, the two demands are separated by watersheds, utilities and regulatory systems. But they increasingly collide through a shared resource: infrastructure capacity. A data center can require enormous amounts of electricity, while the generation of that electricity may itself consume water. A drought can simultaneously reduce farm water availability, hydroelectric output and the reliability of local power systems. And higher electricity demand can raise the cost of pumping irrigation water.

That makes 2026 an important turning point. AI infrastructure is no longer merely an issue for technology companies and electric utilities. It is becoming a rural water-planning issue, an agricultural policy issue and, in some states, a question about whether economic-development incentives should give priority to computing infrastructure or established agricultural users.

AI's electricity problem is becoming a land-and-water problem

RAND researchers Konstantin F. Pilz, Yusuf Mahmood and Lennart Heim warned in their 2025 study, AI's Power Requirements Under Exponential Growth, that AI data-center electricity requirements could grow extraordinarily quickly. Their central high-growth projection was for global AI data-center power demand to reach 68 gigawatts (GW) by 2027 and as much as 327 GW by 2030. The authors also estimated that a single frontier AI training operation could require up to 1 GW at one location by 2028 and potentially 8 GW by 2030 if the underlying scaling trends continued.

The significance for rural America is not simply the size of those numbers. It is their geography. In a later 2026 study, RAND researchers Ismael Arciniegas Rueda, Austin Smidt, Robin Wang, Hye Min Park, David Gill and Henri van Soest examined 22 potential AI data-center sites and concluded that existing substations, transmission connections and previously developed industrial sites could be decisive advantages because major new grid infrastructure is difficult to build quickly. Their assessment makes the problem increasingly local: the question is not just whether America has enough electricity in aggregate, but whether it can deliver enough power to the particular place where a giant computing campus is proposed.

As the RAND team put it, the challenge is no longer only how much additional power the United States might add by 2030: It is also where that power can realistically be delivered. That distinction is critical for farming regions. Electricity is not a single national pool that can be freely moved without constraints. Transmission bottlenecks, substations, generation capacity and local distribution infrastructure determine who can actually receive additional power.

The original RAND analysis is available in its AI power-demand study, while its 2026 site assessment shows why existing infrastructure has become one of the most valuable assets in the AI race.

Farmers experience the same constraint from the opposite direction

Farmers rarely describe their problem in gigawatts. They experience it through pumps, wells, canals, reservoirs, soil moisture and crop yields.

The USDA's 2026 winter-wheat numbers illustrate why this matters. The National Agricultural Statistics Service initially estimated about 33 million acres of winter wheat planted for the 2026 harvest. By March, USDA estimated 32.41 million acres, 2.2% below the previous year. The crop subsequently suffered from difficult conditions in important producing regions, particularly the southern and central Plains. USDA's June estimate put 2026/27 winter-wheat production at roughly 1.03 billion bushels, compared with 1.402 billion bushels in 2025/26. The August estimate remained around 990 million bushels.

That is a striking decline in a crop whose production is closely tied to water availability and weather conditions. The 2026 crop is also part of a broader pattern: the U.S. Drought Monitor reported that 59.05% of the Lower 48 was in drought as of September 1, while severe-to-exceptional drought affected a long list of states. The drought picture is not uniform, but the agricultural implication is straightforward: water reliability has become an increasingly valuable input.

USDA's own crop-monitoring system makes the distinction between drought and crop production important. Agricultural drought is not simply a measure of rainfall. It incorporates precipitation, evaporative demand, soil moisture and surface- and groundwater conditions. A farm can therefore be under increasing agricultural stress even when a short period of rain makes the landscape look temporarily greener.

The latest USDA reporting also illustrates a timing problem for policymakers. Winter-wheat harvest is essentially complete by September, but USDA's final 2026 Small Grains Summary is not scheduled until September 30. As of September 12, policymakers therefore have a fairly good picture of the year's production damage but not yet the final statistical accounting. The USDA National Agricultural Statistics Service reporting calendar shows that distinction clearly.

The hidden connection is electricity used to move water

The most important link between AI and agriculture may not be the water consumed inside a data center. It may be electricity.

Irrigated agriculture is an energy consumer. Moving groundwater to the surface, pressurizing irrigation systems and transporting water through large networks all require power. When a region's electricity demand rises sharply, utilities must expand generation, transmission and distribution capacity. Those costs can eventually affect other customers, including farms.

That creates an unusual feedback loop. A data center can compete with agriculture for electricity even if it never uses a drop of the farm's irrigation water. If the data center causes a utility to build a new substation, transmission line or generating plant, the resulting costs may influence electricity rates. Higher pumping costs then make irrigation more expensive precisely when drought is already making water scarcer.

The water side of electricity is equally important. Cooling power plants can consume substantial quantities of water, depending on the generation technology. This means the water footprint of an AI campus does not necessarily stop at the fence surrounding the data center.

A 2026 analysis by Ceres made this point directly. Its study of Virginia, Texas, California, Illinois, Georgia, Ohio and Arizona concluded that power generation can account for the largest portion of the water associated with data-center operations. The finding changes the policy question from How much water does this data center use? to How much water does the electricity system need to supply this data center?

Arizona shows what happens when water and power collide

Arizona is the clearest example of the problem because it combines rapid technology investment, extreme heat, agricultural water demand and long-term drought.

The state's Department of Water Resources reported that Arizona experienced its sixth-hottest and 37th-driest April-to-June period on record in 2026. More importantly, its long-term drought assessment found extreme or exceptional drought somewhere in every county during the April-June period.

At the same time, Arizona is becoming a major destination for advanced semiconductor and computing infrastructure. The combination is forcing policymakers to think about water and electricity together rather than separately.

Arizona Attorney General Kris Mayes made the political stakes unusually explicit on August 31, 2026, calling for a statewide pause on new data-center approvals, particularly hyperscale AI projects. The only sane thing to do is to pause the approval of new data centers, Mayes said, adding that data centers are power- and water-intensive, and Arizona is challenged on both fronts right now.

That statement is important because Arizona is not simply arguing that data centers use too much water. It is arguing that the electricity required by those facilities can create another layer of infrastructure demand. New power plants may require water. New transmission infrastructure must be constructed. Cooling requirements become more difficult during periods of extreme heat, when both electricity demand and water stress can be elevated.

Arizona also demonstrates why comparing data-center water use directly with farm water use can be misleading. Agriculture remains a much larger water user in the state. The emerging policy conflict is instead about the marginal value of scarce water: whether a new industrial user should receive guaranteed supplies in a basin where existing agricultural users are being asked to reduce consumption, fallow acreage or invest in efficiency.

Georgia reveals a different version of the same conflict

Georgia demonstrates that this is not solely a Western drought story.

The state has become a major data-center market because of its electricity infrastructure, connectivity and business environment. But water planners are increasingly examining whether the development pattern fits regional water availability.

Georgia's Lower Flint-Ochlockonee Water Planning Council discussed data centers in 2026 in a region where water withdrawals had already been subject to long-running restrictions. The council considered both cooling-water consumption and the energy requirements associated with data centers. In another regional discussion, planners noted that development could take agricultural land out of production and that data-center electricity demand could affect the water requirements associated with power generation.

The agricultural connection is especially significant in southwest Georgia, where irrigation is central to crops such as peanuts, cotton and other agricultural products. USDA's 2026 planting data also show that Georgia remains a meaningful winter-wheat producer, although its acreage is tiny compared with the major Plains states.

Georgia has simultaneously taken a relatively aggressive approach to the electricity side. In December 2025, the Georgia Public Service Commission approved arrangements supporting 9,985 MW of new generation, with roughly 80% expected to serve data centers. The commission has emphasized that existing residential customers should not be left carrying those costs.

That approach effectively turns the AI boom into a test of utility cost allocation. If data centers pay their full incremental electricity costs, the conflict becomes easier to manage. If infrastructure costs are broadly socialized, farmers and households can end up indirectly subsidizing an industry that was attracted partly through economic-development incentives.

Ohio shows that water-rich states are not immune

Ohio offers a third model. It has far more freshwater availability than Arizona, but that does not mean water is an unlimited input.

The Ohio Consumers' Counsel reports that the state has more than 200 data centers and that companies could invest as much as $40 billion more in the sector by 2030. Large facilities can consume millions of gallons of freshwater per day for cooling, while hyperscale campuses can require electricity on the scale of a small city.

Ohio lawmakers responded in 2026 with proposals to improve transparency. House Bill 784 would require data centers to report monthly and annual water consumption and prevent local governments from placing that information under nondisclosure agreements. Senate Bill 378 separately proposes rules governing water withdrawal and consumptive use by data centers.

The state's debate is revealing because Ohio does not face Arizona-style chronic Colorado River scarcity. Instead, it is confronting a different question: what happens when an industry with enormous localized water and electricity requirements grows faster than the regulatory system designed to measure it?

The Ohio EPA also reversed course in 2026 on a proposed statewide general wastewater permit for data centers, deciding that facilities discharging wastewater should instead receive individual NPDES permits. That means water policy is moving toward site-specific scrutiny rather than assuming every data center has the same environmental footprint.

WaterSMART offers an important clue about the policy solution

The federal response to water scarcity has traditionally focused heavily on making existing water systems more efficient. The Bureau of Reclamation's WaterSMART program is an important example.

In June 2026, Reclamation announced $38.9 million for 31 Water and Energy Efficiency projects in 10 Western states. The program provides 50/50 cost-share funding to irrigation and water districts, tribes, states and other eligible entities for projects that conserve water, improve delivery efficiency and strengthen water-supply reliability.

These projects can include canal improvements, automation, measurement systems, piping and other modernization work. Reclamation also selected 36 small-scale projects worth $3.7 million in 13 Western states in 2026.

The program's structure is important. It does not simply pay farmers to use less water. It invests in the physical infrastructure that makes water accounting and delivery more efficient. The Bureau of Reclamation's Water and Energy Efficiency Grants explicitly link water efficiency with energy and water-supply reliability.

But there is a crucial policy distinction that becomes more important in the AI era: water saved is not automatically water available for a new data center.

If modernization reduces seepage from a canal, improves measurement or allows farmers to deliver the same crop output with less diversion, the resulting conservation benefit may be committed to drought reserves, groundwater recovery, environmental flows, municipal reliability or future agricultural use. Treating every efficiency gain as a new pool of industrial water would undermine the purpose of the investment.

The next policy fight may be over “who gets the saved water”

This is where AI changes the economics of water conservation.

For decades, governments could justify irrigation modernization primarily on the grounds of agricultural productivity and drought resilience. The arrival of enormous industrial water users creates another possible beneficiary. A city or state that spends millions of dollars modernizing an irrigation district may suddenly be asked whether the resulting water savings can support new industrial development.

That creates a difficult political question. If farmers accept restrictions, invest in more efficient irrigation or fallow acreage during drought, should the resulting water savings support food production, replenish aquifers, restore rivers, protect future agricultural capacity or be allocated to new industrial demand?

Arizona is already demonstrating how politically sensitive such questions can become. In August, a groundwater agreement involving Riverview Dairy in Cochise County included commitments to fallow or transition 2,000 acres of irrigated farmland over 12 years. That arrangement illustrates the direction of travel: water conservation is increasingly being treated as a real economic asset with competing claimants.

The real competition is for reliability, not just gallons

The emerging AI-agriculture conflict should therefore not be reduced to a simple comparison between gallons used by farms and gallons used by data centers.

AI companies need extraordinarily reliable electricity. Farmers need extraordinarily reliable water. Both depend on infrastructure that was generally designed for smaller and more predictable loads. Both are exposed to climate variability. And both increasingly depend on decisions made by regulators who must allocate the cost of upgrading systems.

The difference is that data centers can often move geographically. A hyperscale operator can consider Arizona, Georgia, Ohio, Texas, Indiana or another location based on power availability, tax treatment, water conditions and transmission access. A farm cannot move its soil, irrigation district or local aquifer nearly as easily.

That asymmetry may become one of the most important facts in rural resource policy. If AI developers can bid across states for the most favorable combination of electricity, land, water and incentives, while farmers remain tied to their watersheds, state governments will have to decide whether economic-development policy should compete with agricultural continuity or explicitly protect it.

The strongest policy response may therefore be neither a blanket moratorium on data centers nor unrestricted development. It may be a requirement that new hyperscale facilities demonstrate where their electricity will come from, how much water their cooling systems will consume, how much water is associated with their power supply, who will pay for grid upgrades, and whether the proposed project increases pressure on drought-stressed agricultural systems.

That approach would also give irrigation modernization a new role. Instead of treating water-efficiency grants as a way to find more water for whatever new demand appears next, policymakers could use them to make existing agricultural systems more resilient while separately requiring AI infrastructure to provide the incremental water and power capacity it needs.

In 2026, the AI boom is therefore changing the meaning of infrastructure policy in rural America. The scarce resource is no longer simply water, and it is no longer simply electricity. It is the ability to reliably deliver both at the same place, at the same time, under increasingly variable climate conditions. The states that manage that intersection well may discover that AI investment and agriculture can coexist. Those that treat power, water and land as separate policy questions may find that the competition becomes much harder to resolve once the largest new electricity loads in decades are already connected to the grid.