What ChatGPT, Data Centers, and Steak Can Teach Us About Water Scarcity

From beef production to videos on social media, we're misjudging the real drivers of water use. Here's why outrage often misses the mark and what we should pay attention to instead.

The All-American Canal, in Southern California bifurcating the desert and watering the fields that grow our vegetables and beef with Colorado River water.

Suddenly, everyone seems to be talking about how much water data centers use. The coverage, the tweets, the outrage, it all ramps up fast, and a lot of it hinges on numbers that sound big but aren't placed in any meaningful context. Take millions of gallons of water per day, the projected use for Project Blue in Tucson. Water use measured in gallons made the rounds on social media and in headlines, but rarely with comparisons that helped people understand scale. It’s become increasingly common to see water use reported in gallons, likely because it sounds more dramatic, but this only makes it harder for the public to compare across sectors like agriculture, power, and tech. And so, understandably, it feels like a crisis. But when you zoom out, much of the discourse misses the mark. Not because water isn’t important but because we’re focusing on the wrong parts of the story.

To be clear, I don’t think people are wrong to scrutinize large infrastructure projects. I think it’s healthy (and necessary). But what’s missing from this conversation is a sense of history. The systems we’re reacting to, whether in agriculture or tech,, are downstream from over a century of policy choices. The idea that we could turn deserts into farmland, backed by postwar irrigation subsidies and 19th-century water laws, gave rise to the agricultural footprint we have today. These assumptions were baked into our institutions, and they still shape what kinds of water use we consider normal, necessary, or inevitable. This isn’t the whole story. It’s not even the start. It’s the tip of the iceberg or more accurately, the edge of a massive alfalfa field.

What gave me pause wasn’t the concern about water use, that’s warranted. It was the realization that much of the conversation seemed to begin only when a tech company entered the scene. People who are newly awakened to water politics, especially around Project Blue, are right to be concerned. But the real story goes deeper than one data center. This moment revealed something missing from the broader dialogue: a consistent reckoning with the largest, most entrenched drivers of water use in the West. The outrage over Project Blue (and data centers more broadly) highlighted anxieties not just about ecology, but about corporate power, surveillance, and the pace of technological change. And those are valid. But if we want that outrage to translate into real change, we need to bring it into alignment with the full landscape of water politics and resource use.

A ghost of a river, carved into dust.This is the Colorado River at its end—near the delta in Mexico. Once a vibrant wetland ecosystem, it now flows only rarely, if at all, due to upstream diversions and over-allocation. What water remains is negotiated, rationed, and contested. The image is a reminder that the debates over AI and agriculture aren’t theoretical—they echo downriver, all the way to where the river was once supposed to end.

The Imperial Valley grows much of what ends up on your plate. This desert region, one of the most intensely irrigated in North America, relies entirely on water diverted from the Colorado River via the All-American Canal. From cattle feed to winter greens and vegetables, its fields feed millions across the U.S. and beyond. But this agricultural abundance is fragile. It sits downstream from legal battles, climate change, and collective decisions about what kinds of water use we prioritize. Seen from the air, this landscape reveals both the engineered precision of modern farming and the massive scale at which our food and water systems are entangled.

This is where risk psychology comes into play. Research on environmental and technological risk perception has shown that public outrage tends to be more acute when people encounter new technologies, particularly those they don’t feel in control of. This is often less about empirical risk and more about trust, perception, and familiarity. As Mukherjee and Jensen (2020) demonstrate in their comparative study of water reuse in the US and Australia, novel water technologies like direct potable reuse triggered strong public resistance despite their safety and efficacy, largely because of inadequate risk communication and institutional trust deficits. The psychological effect here isn’t unique to water, it applies broadly to how the public processes AI, data centers, or anything else they perceive as imposed upon them.

Humans are wired to worry more about novel threats than familiar ones. We respond strongly to emerging technologies like AI, especially when accompanied by large, abstract numbers. But we tend to ignore ongoing, normalized harms, like industrial agriculture or fossil fuel infrastructure, because they feel embedded in daily life. Psychologists refer to this as the "availability heuristic," where risks that are vivid or emotionally evocative feel more dangerous. And when those risks are perceived as imposed upon us - large, involuntary, or inescapable - they get coded as what scholars call "dread risks." These are the kinds of threats we instinctively respond to with greater urgency, regardless of the actual statistical risk involved. That’s why 1.6 million gallons of water per day from a data center, equivalent to about 1,910 acre-feet per year, sounds catastrophic, while the 5,984,000 acre-feet used annually for irrigating alfalfa and hay in the Colorado River Basin, or the approximately 1,234,000 acre-feet lost annually to thermoelectric cooling in the same region, register as background noise. These large-scale, ongoing water uses are normalized in our economy and policies, whereas a single, visible project like a data center triggers alarm due to its novelty and scale being reported in isolation.

We also gravitate toward action when we perceive immediacy and agency. A local data center proposal feels stoppable. Global beef consumption or federal energy policy does not. But if we don't align our attention with actual impact, we end up fighting the wrong battles or worse, missing the most important ones. This isn't just about pointing fingers. It's about learning how to read the landscape of risk and power more clearly.

Let’s be clear: data centers do use water. And it's fair to ask questions about their scale and purpose. But when we zoom out, the asymmetry of scrutiny becomes hard to ignore.

Take beef, for example. Producing one 8 oz steak takes roughly 6,800 liters of water by some estimates. That’s about 1,800 gallons. In comparison, a recent study led by Professor Shaolei Ren at the University of California, Riverside, estimated that a single ChatGPT prompt uses around 0.42 watt-hours of energy and about 0.2–0.5 milliliters of water, depending on the energy mix and cooling infrastructure of the data center. That means one steak conservatively uses as much water as 1.3 to 9 million prompts, depending on assumptions.

You may have seen headlines like the one from Brut America: “AI data centers in Texas used 463 million gallons of water, as residents are told to take shorter showers.” That sounds shocking because it’s meant to. But 463 million gallons is roughly 1,420 acre-feet of water. For comparison, that’s about what it takes to irrigate 355 acres of alfalfa for a year. That’s a mid-sized field, not a planetary crime. These headlines use gallons because the number looks bigger and more emotionally evocative. But without context, what kind of water was used, where it came from, how it compares to agriculture, electricity generation, or cooling it’s a misleading metric. These posts don’t help people understand the tradeoffs; they just provoke outrage clicks. And they reinforce a dangerous illusion: that taking shorter showers will somehow solve the structural issues driving western water scarcity.To be clear, there’s still significant uncertainty around these numbers for both AI and Beef Production. AI models like GPT-4o are hosted on varied infrastructure with different cooling systems, energy sources, and grid emissions. Many data centers are secretive about their configurations, and researchers must estimate environmental footprints using public APIs, regional water use factors, and performance benchmarks. So while we don’t know the exact number of prompts per steak, we know the magnitude and can compare. And that matters. Because our instincts are often misaligned with the real scale of things. We worry about the water footprint of the tools we don’t understand, while rarely questioning the impacts of what we consume every day.

But that water footprint doesn’t exist in a vacuum, it’s rooted in the vast agricultural system that supports cattle. In fact, according to Richter et al. (2020), irrigation of cattle-feed crops. including alfalfa, hay, and corn silage, is the largest single consumptive use of river water in the western United States. These crops alone account for 55% of all water consumption in the Colorado River Basin. That water use has downstream effects: it’s a leading driver of river depletion and the imperilment of native fish species due to low flows. The same study found that 88% of fish species threatened by low river flows in the western U.S. were impacted primarily by water withdrawals for irrigated cattle-feed crop. Alfalfa is favored not just for its resilience under variable conditions but because it aligns with structural incentives—cheap water, low labor, and the demands of beef and dairy production. It fixes nitrogen in the soil, which helps regenerate fertility without synthetic fertilizers, and it thrives in hot climates where many other crops can’t survive. In that sense, alfalfa isn’t just a water hog, it’s a rational response to the economic and environmental constraints that many small, mid-sized, and industrial farmers face. At the same time, the political economy around crops like alfalfa is deeply entangled with systems of capital, subsidies, and regulatory inertia. Farmers are bound by a 'use it or lose it' water doctrine that punishes conservation, and by commodity markets that reward water-intensive feed crops. The tech sector, meanwhile, is its own version of this logic: concentrated capital, high energy use, and infrastructure that most people interact with but don’t control. One is rural and rooted. The other is sleek and global. But both reflect the ways extractive systems shape the margins we live and work within. We’re not choosing between wholesome land stewards and greedy tech bros. We’re looking at two differently engineered pipelines of value extraction, one wrapped in pickup trucks and subsidies, the other in server racks and venture capital. If we want to get serious about water, we can’t ignore either. It’s incredibly water-intensive and accounts for a disproportionate share of agricultural water use in the West. Agriculture overall makes up about 72% of water use in Arizona. Municipal and industrial use combined? About 22%. Power generation makes up around 6%. Within that agricultural share, alfalfa alone is the largest single use.

In Arizona, agriculture accounts for 72% of the state’s water use—dwarfing municipal (22%) and industrial (6%) consumption. Source: Arizona Department of Water Resources (ADWR, 2020)

So when a data center using a few hundred acre-feet of water per year becomes the lightning rod for concern, while the tens of thousands of acre-feet dedicated to growing alfalfa, which ultimately ends up as beef, escape broader scrutiny, it points to a gap in our understanding of what’s really driving scarcity. This isn’t a callout, it’s an invitation to deepen the conversation.

Let’s break it down. The proposed facility in Tucson, reportedly to be owned by Amazon Web Services (AWS), was expected to use roughly 1,910 acre-feet of reclaimed water per year for its two sites, or about 6% of the city's reclaimed water supply. On its face, that number sounds enormous. But let’s put it in perspective.

Alfalfa typically uses about 4 acre-feet of water per acre per year in Arizona conditions. So, 1,910 acre-feet is the water needed to irrigate roughly 475 acres of alfalfa. That’s a medium-sized field, not even a blip on the radar compared to the hundreds of thousands of acres planted annually statewide in Arizona.

It’s also important to understand what reclaimed water is. Unlike potable water pulled from aquifers or the Colorado River, reclaimed water is treated wastewater - sewage, essentially - that’s been processed and disinfected for reuse. It’s not drinking water, but it’s clean enough for industrial, irrigation, and construction purposes. In many cases, this water would otherwise be lost to evaporation or discharged into the Santa Cruz River.

University of Arizona researcher raised a valid concern: that diverting this water for industrial use would reduce discharges that currently support riparian restoration and ecological flows in the Santa Cruz. That’s true to a point and it’s a tradeoff worth discussing. But it’s also the kind of dilemma we’re going to face more often: choosing between ecosystem restoration, industrial innovation, agricultural legacy, and population growth in a tightening water budget. Building the infrastructure to use reclaimed water more widely, like the pipeline proposed in Project Blue, was one proposed step toward better reuse. Whether or not it was the right one, it asked a question we still need to answer.

Cotton in the desert. Despite Arizona’s heat and deepening water crisis, cotton continues to thrive here thanks to a mix of history, infrastructure, and economics. The plant loves sun and heat, and the arid climate helps reduce pests. Generations of farmers have built their livelihoods around it, supported by long-standing irrigation rights, local gins, and federal subsidies. Much of the cotton grown here, especially premium varieties like Pima, is exported worldwide. But with the Colorado River in decline, fields like these raise a difficult question: In a hotter, drier future, what crops should still be grown with our most precious resource?

So this raises a broader question we haven’t collectively answered: what’s already using water, and what might deserve to replace it? Should data centers, controversial as they are, be considered a more justifiable use of water than cotton? The answer isn’t simple. But our screen time, streaming habits, and constant digital connectivity suggest which direction we’re already choosing. The question now is whether our water policy will catch up to our behavior and whether we’re willing to have that conversation in full view.

AWS is the backbone of the internet as we know it. It doesn’t just serve Amazon, it hosts government platforms, small businesses, universities, and countless apps, including some of the very ones people use to voice their outrage. From Netflix to Reddit, the services that fill our lives run on infrastructure like this. It’s understandable to be skeptical of corporate consolidation and the scale of tech’s footprint, those critiques are valid. But we should also be honest about how this infrastructure exists to serve our habits. The real story here isn’t just about a corporate land and water grab, it’s about how our content consumption has become one of the fastest growing sectors of the economy, and how the demand for streaming, cloud storage, and 24/7 connectivity is driving a quiet explosion in infrastructure. Criticizing Amazon is fair. But it doesn’t change the reality that our collective behaviors (our hunger for convenience, connection, and entertainment) are also shaping what infrastructure gets built and where. We can hold the wolves at the door for a while. But the wolves are us.

Project Blue is now dead. The Tucson City Council voted unanimously to pull the plug, citing community concerns about water use and transparency. And to be fair, those transparency concerns weren’t unfounded. Questions were raised about how the project was communicated to the public, and whether proper disclosure had occurred when the operator’s identity, Amazon AWS, was accidentally revealed, potentially violating a non-disclosure agreement . These procedural missteps understandably eroded public trust. Personally, I don’t have strong feelings either way about the project from a water-use angle. It would have used reclaimed water and built critical greywater infrastructure that could have benefited the region long-term. But the way it was handled politically left a lot to be desired. Ultimately, the optics, and the politics, were too much.

That’s not to say data centers are benign. They are part of an extractive infrastructure that includes not just water use but energy use, mineral demand, and e-waste. And public scrutiny is necessary. But much of the panic online didn’t match the actual scale or context. What we weren’t talking about was what’s really driving demand for more data centers. It’s not just AI. It’s us. Our phones. Our content consumption. Our constant refreshing of short-form video feeds. If you want to understand why more servers are being built, start with your own screen time report.

There’s also an irony in how AI is being singled out when it's not the full picture of total data load. A signification portion of energy demand within data centers still comes from video streaming, advertising, and search, the everyday plumbing of the consumer internet. But “Project Blue” made for a better villain than “Instagram Reels,” and it gave people a target.

And we haven’t even touched electricity. One of the most overlooked components of data center impact is their energy demand. Large-scale data centers can consume hundreds of megawatt-hours per day depending on their load, backup systems, and cooling requirements. At full build out Project Blue was expected to need about 700MW in electrical generation capacity which is as much power as a small town. If that power comes from fossil sources, as much of Arizona’s still comes from natural gas, it carries a secondary water burden through thermoelectric cooling. Natural gas plants, in particular, use significant volumes of water to condense steam. Most public scrutiny doesn’t fall on the Arizona Corporation Commission, which has long rubber-stamped utility expansion and delayed renewable portfolio standards. Instead, attention drifts to whatever headline trends next, often missing the underlying structures that got us here in the first place.

If we want our water politics to be rooted in impact, not just reaction, we need to broaden the scope of our concern. That means talking about agricultural reform. It means reckoning with energy policy. It means holding state-level regulators accountable. These are long, unglamorous fights. They involve wonky regulatory bodies, tribal consultation processes, and regional water boards. But they’re also where the future of the Colorado River is being decided. If you’re looking to plug in, start by following the 2026 Colorado River renegotiation process and how tribal nations and frontline communities are, or aren’t, being included.

My goal here isn’t to defend big tech. I understand the discomfort. And I understand the stakes. But for me, the arugment over data centers, is a reminder of how easily our public attention can be captured, and how rarely we take the opportunity to zoom out. The uproar gave me pause not because it was unjustified, but because it revealed a narrow frame through which we tend to view water, risk, and responsibility.

If you’re interested in going deeper, I’d recommend the first episode of the Outside TV series I made, Landscape: Running Dry (yes, I’m fully aware that watching this will consume water and emit carbon into the atmosphere for a shameless plug). I began at the parched delta of the Colorado River in Mexico, where the river no longer meets the sea, and journeyed north by bike and boat, tracing its path through the Imperial Valley, Navajo lands, and eventually up to cattle ranchers in Colorado. The series weaves together different perspectives, urban, rural, Indigenous, industrial, to explore how water use, policy, and history collide across the basin. It’s not a policy explainer. It’s a ground-level look at what it means to live inside these systems.

This six-part series follows my human-powered journey up the Colorado River—from the dried-out delta in Mexico to the ranchlands of Colorado. Along the way, I spent time in the Imperial Valley, with water managers, cattle ranchers, and communities whose lives are deeply entangled with the river’s fate. If this essay zooms out on our discourse, the series zooms in on the people living inside the system.

The uproar about data centers has never been just about water. And I’ve heard a lot of people say, with sincerity, that we can’t just point fingers at alfalfa or agriculture, we have to fight on all fronts. I don’t disagree in theory. But I also think that argument, when used to justify a kind of all-or-nothing activism, can become a deflection. It avoids the much harder, more sustained work of engaging in electoral politics or trying to influence the Colorado River negotiations now underway. There’s a limit to how much public attention we can mobilize. And if we’re going to get people to stay engaged in the long haul, we have to be strategic about where that energy flows.

My hope is that the same momentum and scrutiny that showed up around Data Centers can pivot toward these bigger structural conversations. Because this isn’t just about stopping things. It’s about building a more coherent, informed public understanding of the systems we all rely on. And maybe, if we’re lucky, nudging them toward something more sustainable.

Yes, we should ask hard questions about how digital infrastructure impacts ecosystems. But we also need to step back and ask why certain forms of impact spark immediate outrage, while others pass under the radar. It’s easy to see the threat when it’s unfamiliar and corporate. Harder when it’s part of daily life. Maybe the bigger question is: what would it look like to align our attention with our actual values and our actual impact?