Everything About AI
NextFin News - The American West is entering one of its driest stretches in recorded history at the same moment artificial intelligence is turning data centers into some of the thirstiest industrial tenants on the grid, and the collision is forcing a reckoning that reaches far beyond environmental advocacy: water reuse is being repriced as a core infrastructure investment. Lake Mead fell to a record low of 1,040.4 feet above sea level in August, while more than two-thirds of the data centers built since 2022 sit in watersheds already classified as water-stressed. The question investors face is not whether the technology exists - it does - but whether the regulatory and commercial machinery can move fast enough to turn recycled water into a bankable asset class before the next drought cycle bites.
The Situation: Two Crises Colliding in the Same Watersheds
The facts, laid side by side, define the tension. The Colorado River system - which supplies municipal water to roughly 40 million people and irrigates about 5.5 million acres of farmland across seven states - is at a historic low. The combined storage in Lakes Mead and Powell has fallen to levels researchers expect to remain at record lows every day until spring runoff arrives. Ninety-five percent of the Colorado River basin is in drought, and the region has been gripped by a megadrought for more than a quarter of a century.
At the same time, the AI buildout is accelerating water demand in precisely those stressed basins. An analysis of the sector found that more than two-thirds of new data centers built since 2022 are located in water-stressed regions, with approximately 160 new AI-focused facilities constructed in the United States over the past three years - a 70 percent increase from the prior three-year period. The concentration is striking: five states - Virginia with 67 facilities, Arizona with 26, Texas with 26, Illinois with 23, and California with 17 - account for 72 percent of all new data centers sited in locations of acute water stress. At least 59 more facilities are planned in US water-stressed regions by 2028.
The water intensity is real and measurable. A medium-sized data center consumes roughly 110 million gallons of water annually, comparable to the yearly needs of a town of 50,000 residents. Large hyperscale facilities in hot climates can draw 1 million to 5 million gallons per day. Google's 2026 Environmental Report disclosed consumption of 10.9 billion gallons in 2025, up 34 percent year over year and more than double its 2021 level, even as the company improved its water-use efficiency to 0.30 liters per kilowatt-hour - a 39 percent improvement since 2021. The Lawrence Berkeley National Laboratory's December 2024 assessment put direct US data center water consumption at 17.4 billion gallons, with an additional 211 billion gallons consumed indirectly through electricity generation.
The response is taking shape in three channels: corporate commitments, state-level regulation, and federal tax policy. Amazon plans to expand recycled-water cooling from 24 data centers to more than 120 locations across the United States by 2030, a move the company says will preserve more than 530 million gallons of drinking water annually. In May 2026, Senators Ben Ray Luján and Katie Britt introduced the bipartisan Advancing Water Reuse Act, which would create a 30 percent federal investment tax credit for qualifying water recycling and reuse projects. And the EPA released WRAP 2.0 in April 2026, positioning water reuse as a federal coordination priority for data center deployment.
The investment case is being underwritten by market-research forecasts that treat water reuse as a decade-long growth vector. The global industrial water reuse systems market is projected to grow from USD 23.9 billion in 2026 to USD 67.2 billion by 2036, a compound annual growth rate of 10.9 percent, with membrane filtration leading by technology and cooling-water reuse dominating by application. The broader water recycle and reuse market is expected to reach $27.1 billion by 2030, growing 10.5 percent annually from 2024.
But market size is not the same as investability. The question is whether this is a cyclical weather trade or a structural regime shift - and that distinction determines everything about how the theme should be priced.
The Mechanism: Why Water Is Different From Power
The first-order story is obvious: data centers need water for cooling, droughts reduce water, therefore reuse becomes valuable. That chain is true but shallow. The mechanism that makes water reuse structurally different from, say, a spike in natural-gas prices is that water cannot be piped in from a distant basin the way electricity can be transmitted across a grid. Water rights are adjudicated watershed by watershed, and in the American West they are seniority-based: the newest, largest industrial user is the first to be cut when the reservoir falls.
This creates a physical lock-in that power does not impose. A data center's structure has an expected lifetime of at least 20 years, so a facility permitted today in a stressed watershed is committing to a water source that may not exist in its current form by 2040. The buildout is already concentrated in the wrong places, and relocation is not a clean escape - the capital is already sunk, the interconnection queue is already paid, and the latency requirements that drive site selection near fiber hubs and power substations do not move easily.
The second-order effect is where the investment thesis actually lives. Water stress does not merely raise operating costs; it raises the cost of capital for projects that cannot demonstrate a resilient water supply. Lenders and rating agencies are beginning to treat water risk as a credit factor, and a facility dependent on potable municipal supply in a stressed basin will face higher financing costs than one with a contracted reuse source. That is the transmission channel: drought risk becomes water-risk repricing, which becomes a differentiated cost of capital, which becomes a competitive advantage for operators with reuse infrastructure. The beneficiaries are not only the technology vendors; they are the utilities and developers that can secure long-term reuse offtake agreements.
There is also a Scope 2 trap that most sustainability disclosures obscure. A data center that switches to air cooling can report near-zero on-site water use, but air cooling consumes more electricity, and thermoelectric power generation evaporates water upstream. The Lawrence Berkeley National Laboratory estimates indirect consumption at roughly 12 times direct consumption. So a "zero water" claim based only on on-site measurement can shift the burden rather than eliminate it. This is why reuse - which addresses the actual watershed draw regardless of cooling technology - is more durable than cooling-system substitution alone.
The friction is already showing up in the capital pipeline. A tracker of development opposition recorded at least 75 data center projects worth approximately $130 billion blocked or delayed in the first quarter of 2026 alone - roughly matching the scale of all of 2025 in just three months. Communities are organizing around electricity costs, water consumption, and noise, and the number of active opposition groups more than doubled across 49 states between year-end 2025 and March 2026.
Cyclical or Structural: The Call
This is the judgment the market has not priced cleanly. Drought itself is cyclical: the Colorado River has wet years and dry years, and a series of strong snowpacks in Colorado, Utah, and Wyoming would refill the reservoirs and ease the immediate pressure. If this were only a weather story, water reuse would be a mean-reverting trade - buy the panic, sell the rain.
But three forces make this structural, and they will not self-correct in a wet year.
First, demand is on a secular path. The International Energy Agency projects global data center electricity consumption will more than double between 2022 and 2026, reaching 1,000 terawatt-hours annually - roughly equivalent to Japan's total consumption. Because the majority of that electricity becomes waste heat that must be rejected, cooling water demand scales with compute demand. AI inference load grows with deployment, not just with construction, so the water intensity compounds as models move from training to production.
Second, the regulatory architecture is being built now, and regulation is sticky. Illinois has advanced legislation requiring data centers to disclose water and energy usage to the Illinois Power Agency; companion proposals would require data centers to pay for water infrastructure costs. West Virginia has proposed limiting data center usage where it causes adverse supply impacts. Minnesota enacted data center water reporting in 2025. Virginia is pushing for statewide disclosure of energy, water, and emissions. These are not one-season policies; once disclosure regimes exist, they create the data infrastructure for performance standards, and performance standards create the market for reuse technology.
Third, the physical capital is being sunk in the wrong watersheds. Even if every new facility from 2027 onward were sited in water-rich regions, the two-thirds already built or in development since 2022 would remain in place for two decades. That installed base is the addressable market for retrofit reuse systems, and it is not going anywhere.
The cyclical leg and the structural leg should be separated, not blended. The cyclical leg is the weather-driven urgency that compresses permitting timelines and spikes political attention - that will ebb in wet years. The structural leg is the combination of secular compute demand, regulatory lock-in, and physical site lock-in - that persists. The investment implication follows: drought headlines create the political window, but the thesis does not depend on the next drought.
The Counter-Thesis: Technology and Relocation Could Defuse the Crisis
The strongest case against this thesis comes from the technology side, and it deserves its full weight. New cooling technologies now make close-to-zero direct water consumption possible. Microsoft launched a next-generation datacenter design in August 2024 that consumes zero water for cooling, with pilot sites in Phoenix, Arizona, and Mt. Pleasant, Wisconsin, expected to begin coming online in late 2027. The closed-loop system saves more than 125 million liters of water per facility each year, and the company has reduced its Water Usage Effectiveness from 0.49 liters per kilowatt-hour in 2021 to 0.30 - a 39 percent drop. If these designs become the default, the direct-water problem that motivates reuse largely disappears.
The relocation argument runs alongside it: developers can simply site new facilities in water-rich regions, and the market has already begun to price water risk into site selection. And on the macro scale, the counter-thesis has a factual anchor - data centers still account for a very small fraction of total US water consumption, and agriculture dominates in states like Arizona, where it accounts for roughly 86 percent of total usage. From that vantage point, the data center water story is a rounding error dressed up as a crisis, and the reuse market is being oversold.
The answer to the technology counter-thesis is not that it is wrong, but that it is incomplete in three ways. First, zero-water designs carry higher upfront capital costs and are not yet scalable across the existing installed base - the hundreds of facilities already operating with evaporative cooling will not be rebuilt. Second, zero-water cooling trades water for electricity, which raises the Scope 2 water burden upstream unless the power is solar or wind; in a grid still heavily dependent on thermoelectric generation, the watershed benefit is partial. Third, hot climates - precisely the water-stressed regions where the buildout is concentrated - are where free-air and dry cooling perform worst, because the ambient air cannot absorb heat efficiently. The places that need water savings most are the places where water-free cooling is hardest to deploy.
The relocation counter-thesis fails on the lock-in point: the capital is already committed, the fiber and power interconnections are already built, and the latency-sensitive workloads that justify those sites do not move. And the "small fraction of total consumption" argument, while true at the national level, is irrelevant at the watershed level - which is where water is actually allocated. A data center drawing 3 million gallons a day in a basin at 1,040 feet is not competing with national averages; it is competing with the farmer downstream and the suburb that just got asked to cut lawn watering.
The counter-thesis would be proven right, and this structural call proven wrong, by a specific observable: if zero-water or air-cooled designs reach more than 40 percent of new hyperscale capacity by 2028, or if new hyperscale builds in the five high-stress states fall below half of currently planned capacity, the reuse thesis loses its structural foundation and reverts to a cyclical weather trade.
Who Benefits, Who Is Exposed
The beneficiaries fall into three buckets. Technology vendors with reuse and treatment capability sit closest to the theme: Xylem, which completed its approximately $7.5 billion acquisition of Evoqua Water Technologies to create what it calls the world's largest pure-play water technology company, reports that its technology enables 18.1 billion cubic meters of water reuse globally. Veolia, trading on the Paris exchange, has made resource preservation a pillar of its GreenUp strategic program and is already executing reuse projects with hyperscalers - including an Amazon partnership in Mississippi expected to reuse more than 83 million gallons of potable water per year once operational. Other named players in the reuse value chain include SUEZ, Fluence, Alfa Laval, Pentair, and Pall.
The second bucket is utilities and developers that can contract long-term reuse offtake. Essential Utilities' $26 million investment in Pennsylvania to build an 18-million-gallon-per-day treatment plant supplying a data center and its onsite power facility is the template: a regulated utility earns a return on reuse infrastructure while the data center secures a water source insulated from drought curtailment. Municipal wastewater utilities in stressed basins gain a new revenue stream by selling treated effluent instead of discharging it.
The exposed are the operators dependent on potable municipal supply in stressed basins without reuse alternatives, and the municipalities that have promised water to data centers on the assumption that historical flows will persist. The friction is rising, and the voice of the water utilities themselves captures the shift.
John Hernon, strategic development manager at Thames Water in the UK, put it directly in an interview:
Water is often one of the last things that gets thought about, so when you're thinking about data centers, it's not just about the speed you're going to get, it's not just about making sure there's a lot of power available - we need to make sure that water is factored in at the earliest possible thinking ... at the forefront, rather than an afterthought.
Thames Water's service area holds about 80 percent of the UK's data centers, with another 100 or so proposed - and the utility has said it is willing to restrict, reduce, or object to data center connections when supply cannot accommodate them.
Outlook: What to Watch and What Would Break the Thesis
The mechanism cashes out into a clear asymmetry. Water reuse is not a single technology bet; it is infrastructure that becomes more valuable as the alternative - potable supply in a stressed watershed - becomes less reliable. That is why the theme is structurally underwritten rather than weather-dependent.
Split by time horizon: in the short term, the trade is driven by drought headlines and permitting friction - sentiment and liquidity, and it will be volatile. In the medium term, the earnings visibility comes from the retrofit backlog: the two-thirds of facilities already in water-stressed regions cannot wait for the next wet year. In the long term, the thesis stands or falls on whether reuse becomes codified in state performance standards - and the legislative activity in Illinois, West Virginia, Minnesota, and Virginia suggests codification is underway.
Three scenarios frame the path. The base case: reuse adoption grows at the modeled 10-11 percent annual rate, driven by a combination of state disclosure rules, the federal 30 percent investment tax credit if enacted, and corporate water-positive commitments. The upside case: a severe multi-year drought in the Colorado River basin triggers emergency curtailments that force accelerated retrofit spending, pushing growth above the model. The downside case: zero-water cooling reaches the 40 percent threshold in new hyperscale capacity ahead of schedule, or a string of wet years defuses political urgency and stalls the disclosure regimes - in which case the theme compresses to a niche retrofit market.
What to watch, specifically: state-level data center water disclosure bills moving from proposal to enactment; the share of new hyperscale capacity announced with zero-water or closed-loop cooling; and the quarterly water-use disclosures from the hyperscalers, where the gap between consumption growth and replenishment is the tell. The single falsifying signal: if new hyperscale builds in the five high-stress states fall below half of planned capacity by 2028, or zero-water designs exceed 40 percent of new capacity, the structural call is wrong.
This is not a drought trade. It is a repricing of water as a constrained input for the AI economy, and the companies that can turn wastewater into a contracted, drought-proof supply are the ones that will capture the value - wet year or dry.
Explore more exclusive insights at nextfin.ai.
