Looking for the truth about AI data centers and advice that’s neither blind acceleration nor reflexive panic? This article is my (i.e. Gabriel Cassady’s) evidence-based position on AI data centers, their actual ecological footprint, and what citizens should demand. Written from Springfield, Missouri, by someone who bet his career on AI in 2022 and has been answering this question every week since.
A Week in the Life of the AI Buildout
On Wednesday, May 20, 2026, Missouri Governor Mike Kehoe stood next to Google executives in High Hill, Missouri, and announced a $15 billion data center investment in Montgomery County — what state leaders called the largest single private investment in Missouri history. While the governor was celebrating, a group of residents called Preserve Montgomery County LLC was preparing for a June 1 court hearing in a lawsuit alleging that county commissioners violated the Missouri Sunshine Law by withholding water-use data and conducting closed-door meetings about the projects.
That was Wednesday. Three weeks earlier, Elon Musk’s xAI committed $20 billion to a new Mississippi data center — the largest private investment in that state’s history. In February, three major tech companies locked in multi-million-gallon-per-day water allocations in Virginia, Louisiana, and Indiana, with associated water infrastructure costs approaching a billion dollars in a single month.
This is not a story about one announcement. This is the new normal. And every region in America is going to have a version of the same fight.
I get asked some variation of “what about the environmental cost of AI?” every week — at coffee shops, in client meetings, at family gatherings. So I am going to lay out, once, the honest position I have arrived at after years of working with this technology and reading the actual research. The truth about AI data centers is more complicated than either the doom-mongers or the cheerleaders will tell you, and the only useful place to stand is in the middle, with your eyes open and your civic muscle engaged.
This is meant to be the foundational piece I link back to from everything else. If you want my actual evidence-based take on AI, climate, and water — bookmark this one.
The Truth About AI Data Centers Starts With Concrete and Steel
I made my bet on AI in 2022. I have not regretted it. I still believe AI may become one of the most useful technologies humans have ever built — drug discovery, materials science, grid optimization, accessibility, the works. I make my living helping small and mid-sized business owners figure out how to use it without setting their hair on fire.
None of that gives the buildout an ecological free pass.
Here is the first thing most people get wrong. When normal people hear “cloud,” they tend to imagine something abstract — software, light, ones and zeros. The reality is industrial. AI runs on physical buildings full of servers, substations, transmission lines, backup diesel generators, and cooling systems that draw electricity and water from real communities at real cost. A modern hyperscale AI campus can require 100 megawatts of power or more — sometimes drawing as much electricity as an aluminum smelter, on a footprint measured in hundreds or thousands of acres.
If a multi-billion-dollar developer wants to build one of these facilities near a real community, using real electricity and real water from already-stressed systems, residents are not paranoid for wanting unredacted answers. They are doing exactly what citizens are supposed to do.
The grown-up position is not ban all data centers. It is not trust the tech companies. It is this:
AI may well be worth building. The infrastructure behind it must be built responsibly, transparently, and with enforceable local safeguards.
That is not a hedge. That is the only honest place to stand.
The Climate Question Is Mostly an Electricity Story
When people worry about AI and climate, what they are usually worrying about — whether they say it this way or not — is power. Training and running modern AI models takes immense computation. Computation takes electricity. If that electricity comes from high-emission sources, the climate impact tracks.
Here is what the actual research says.
A 2024 Lawrence Berkeley National Laboratory report, published by the U.S. Department of Energy, found that American data centers consumed about 176 terawatt-hours of electricity in 2023 — roughly 4.4% of total U.S. electricity use. The same report projected that figure to land somewhere between 325 and 580 terawatt-hours by 2028, or between 6.7% and 12% of total U.S. electricity consumption.
That range is wide on purpose. The lower end assumes meaningful efficiency gains. The upper end assumes the current trajectory continues without major technical breakthroughs. Either way, the direction is the same. Demand is rising, and rising fast.
The International Energy Agency tells a similar story globally. Data centers consumed about 415 terawatt-hours worldwide in 2024 — roughly 1.5% of total global electricity — and the IEA’s base case projects that doubling to around 945 TWh by 2030, more than Japan’s entire annual electricity consumption today. The United States holds about 45% of global data center electricity consumption, and by 2030, American data centers are expected to draw more power than all the country’s energy-intensive manufacturing — aluminum, steel, cement, chemicals — combined.
Read that line again. Data processing will out-consume heavy industry.
In plain English: even though data centers are a modest share of global energy demand on average, the impact is brutally uneven on the ground. Nearly half of U.S. data center capacity is already clustered in five regional hubs. In Northern Virginia, data centers already consume more than a quarter of all metered electricity in the state. When one 100-megawatt facility lands in a specific region, it reshapes the entire local load profile overnight.
That is the climate story, honestly. Not apocalypse. Not nothing. A real, growing, geographically concentrated demand spike that the grid was never designed to absorb without serious planning.
The Truth About AI Data Centers and Water Is Worse — and Better — Than Either Side Says
This is where the public conversation tends to get fuzzy. Bear with me, because the nuance matters.
Some data centers use water-intensive evaporative cooling. Some use closed-loop direct-to-chip systems that practically eliminate evaporation. Some use reclaimed wastewater. Some still draw potable drinking water from municipal supplies during peak summer heat — which is exactly the wrong combination of choices, and which is still legal in most of America.
A March 2026 study from researchers at the University of California, Riverside, in collaboration with Caltech, finally quantified the part of this debate that most local utilities had been quietly worrying about for years: the summer spike. On the hottest days of the year, evaporative cooling at a single large data center can draw more than a million gallons of water per day — and the UCR team found that some facilities under construction have been allocated up to 8 million gallons per day, enough to supply multiple small towns at once. Daily demand on hot days can be six to ten times the average. At some planned facilities, more than thirty times the winter baseline.
Municipal water systems are not built for average demand. They are built for peak demand. So if a hyperscale data center lands in your county and uses evaporative cooling, your local utility has to engineer and finance peak capacity that mostly sits idle the rest of the year. The UCR researchers estimate the national cost of building this required water infrastructure through 2030 at between $10 billion and $58 billion, depending on how fast the buildout actually moves.
Here is the part of the truth about AI data centers and water that should land hardest. The Environmental Protection Agency itself, in its updated Water Reuse Action Plan 2.0 released April 16, 2026, now explicitly identifies data center cooling as one of the central pressure points on American water infrastructure. The EPA’s own guidance for data centers recommends air cooling, closed-loop liquid cooling, and reclaimed water — not potable drinking water — as the default.
So when a developer says “we have to use drinking water, that’s just the cost of progress,” they are not telling you the truth. They are telling you the cheapest option for them.
When Industry Cleans Up, It Is Because It Was Pushed
Microsoft is a useful proof point.
In December 2024, the company announced that all new data center designs initiated from August 2024 onward would consume zero water for evaporative cooling — using closed-loop, chip-level liquid cooling that recycles coolant continuously. Microsoft estimates the change saves more than 125 million liters (roughly 33 million gallons) of water per data center per year, with pilot facilities in Phoenix and Mt. Pleasant, Wisconsin starting up in 2026 and broader deployment slated for late 2027.
Amazon Web Services, in its public statement supporting the new EPA plan, now operates 22 facilities cooled with reclaimed water and plans to expand to “nearly 100 U.S. locations by 2030.” Google’s new Missouri build is being marketed on closed-loop air cooling that limits water use to domestic purposes like kitchens.
I am not citing these companies because they deserve an award. I am citing them because they make the inverse argument impossible. If Microsoft can engineer zero-water cooling, then evaporative-water cooling is a choice — usually the cheaper one. And the choices industry makes are downstream of the pressure communities apply.
That pressure can be applied. It is being applied right now. The question is whether we will apply it in our own backyards.
Local Governance Is the Lever That Actually Works
Here is the part most people miss. Most national debates about AI infrastructure are theater. The actual decisions — who pays for transmission upgrades, how summer water peaks get managed, whether legacy ratepayers cross-subsidize a hyperscale customer’s substation — get made at the local utility, county commission, and state public service commission level.
That is where citizens still have leverage. That is also where it is most often given away without a fight.
I will use my own city as one example among many. City Utilities of Springfield, Missouri, is a community-owned utility — not an investor-owned utility legally obligated to maximize shareholder returns. It has served Springfield since 1945, delivers electricity, gas, water, broadband, and transit to over 111,000 customers across 320 square miles, and is governed by an eleven-member Board of Public Utilities — local citizens appointed by Springfield City Council. Hundreds of American cities have a version of this structure — community-owned utilities, municipal utilities, public power districts. Wherever you live, find out who governs yours.
State-level policy can either support or undermine that local leverage. Missouri’s Data Center Sales Tax Exemption Program grants 15 years of sales tax relief for projects meeting fairly modest thresholds. Missouri’s Senate Bill 4, signed in April 2025, introduced large-load tariff requirements designed in theory to prevent costs incurred to serve data centers from being passed to residential ratepayers. Both the Missouri Coalition for the Environment and the Consumers Council of Missouri have flagged structural loopholes — including a definition of “large load” that lets developers split projects into smaller facilities to avoid the rules. Even Senator Josh Hawley, no friend of progressive environmental policy, publicly disputed state leadership claims that residents would not pick up part of the data-center tab.
Variations of these stories are happening in every state with data center activity. The specifics differ. The pattern does not.
What Communities Should Demand — In Writing
If a hyperscale data center is proposed in your county, the questions to ask are specific, local, and on the record:
- Where does the electricity come from? Existing generation, new generation, grid upgrades, long-term power purchase agreements? Is new fossil-fuel infrastructure being built specifically to serve this customer?
- Who pays for the new transmission? The developer, directly and contractually — or existing ratepayers, indirectly?
- How much water will the facility use on the hottest days of the year? Not the annual average. The peak summer day.
- What cooling technology? Evaporative, closed-loop direct-to-chip, hybrid? Potable water or reclaimed wastewater?
- What is the drought contingency plan? What happens during the next bad heat dome?
- What are the tax abatements — and what is the public actually getting back?
- What environmental and operational data will be reported publicly, in real time? Energy use, water draw, backup-generator emissions, expansion plans, incident disclosures.
These are not gotcha questions. They are the basic, contractual terms of any responsible large infrastructure project.
The current best-practice tool for getting them in writing is a Community Benefit Agreement — a legally binding contract between developer and community that spells out cooling mandates, water caps, infrastructure funding, exit fees if the project is abandoned, wage commitments for construction labor, environmental monitoring with teeth, and direct community-betterment payments that survive any tax abatement. The Brookings Institution has solid frameworks for what these contracts should look like. So does the NAACP’s Stop Dirty Data Centers campaign. The legal tools exist. The political will is the variable.
Voluntary corporate commitments are not enforceable contracts. Anyone who has run a small business knows the difference.
Both Extremes Get This Wrong
One extreme treats every data center as if it were automatically a climate crime. The other dismisses every legitimate concern as anti-progress hysteria from people who do not understand technology.
Neither is serious enough for the moment we are actually living in.
AI is going to be woven into the future, whether we love it or fear it. That raises the stakes for everyone — including, and especially, the people who host the infrastructure that makes it work. The fight is not only over whether the buildings get built. The fight is over how they get built, where they get built, who pays the externalities, and who actually benefits.
That is where citizens still have leverage. That is where local governance still matters. That is where the truth about AI data centers becomes a civic question, not a technical one.
The Bottom Line
I remain cautiously optimistic about AI itself, and deeply cautious about the political economy growing up around it. I do not think the right answer is blind acceleration. I also do not think the right answer is reflexive moratoriums everywhere on principle.
The right answer is pressure.
Pressure for cleaner power. Pressure for closed-loop cooling. Pressure for reclaimed water over potable water. Pressure for transparent local permitting. Pressure for legally binding Community Benefit Agreements. Pressure for honest tax math. Pressure for developers to prove, not merely promise, that the upside of their project is worth the burden it places on the places around it.
AI may help solve some of the hardest problems humans have. Climate change among them. But it will not do that automatically — and it will not do it at all if we build the infrastructure powering it in the dumbest, cheapest, least accountable way available.
The cloud is a lot of buildings. The buildings are coming. The question is who gets to set the terms.
That, in the end, is the truth about AI data centers worth defending.
Anxious about AI’s footprint in your community? Or running a small or mid-sized business trying to figure out the AI question honestly, without getting taken for a ride? Let’s talk. Reality-check before hype. Always.
For more on how I think about AI’s place in human life, business, and community, read my AI manifesto or learn more about my approach.
