Summary: This post explains where the United States should build data centers if the AI boom continues, and why those locations matter for carbon emissions and water use. The analysis rests on a recent Nature Communications study and the real-world commitments from major tech firms. I follow the paper’s reasoning, restate the facts, weigh trade-offs, and offer concrete policy and siting suggestions so industry and regulators can make better choices now rather than pay later.
Why this question matters right now
Tech companies are pledging staggering sums to build computing capacity on U.S. soil. Meta told the White House it plans $600 billion in U.S. infrastructure by 2028; OpenAI has pledged $1.4 trillion. Those numbers are not abstract—each dollar converts to servers, power hookups, cooling systems, and water needs. The Nature Communications study models how that buildout could affect emissions and freshwater stress through 2030. If the AI boom keeps growing, these choices will shape emissions and water stress for years. The question is simple: where should we put these facilities so the environmental damage is the smallest?
What the study did and why it is credible
Researchers used projections for AI chip demand, mapped state-level electricity mixes, and layered in water-scarcity metrics to build scenarios for data center impacts through 2030. The paper models multiple outcomes depending on how fast AI demand grows, how quickly computing gets more efficient, and how the electricity grid evolves. Nature Communications is a peer-reviewed outlet; the lead author, Fengqi You of Cornell, is an energy-systems expert. That gives the study weight—though no model is a crystal ball. The authors are careful to present ranges, not single forecasts.
Location drives both water and carbon footprints
The study’s headline point is straightforward and worth repeating: water and carbon footprints depend heavily on where data centers are situated. Where you site a server farm matters because (1) states with greener grids produce fewer emissions per kilowatt-hour, and (2) states with abundant freshwater handle evaporative or liquid cooling with less stress. Put simply: location is one of the most effective levers we have for lowering damage.
Which states come out best — and why
The analysis identifies four states that strike the best balance between energy and water availability: Texas, Montana, Nebraska, and South Dakota. Each offers strengths that reduce combined environmental harm: Texas has an existing data-center industry and growing renewable capacity; Montana and South Dakota have low population density and grid mixes that can be relatively clean; Nebraska has huge wind potential. These states can host a lot of compute while keeping incremental emissions and water withdrawals lower than more stressed regions.
Why industry still clusters in Virginia and California
Historical patterns persist. Northern Virginia hosts over 650 data centers, largely because it is near Washington, DC, dense fiber routes, a skilled workforce, and favorable tax treatment. California, with more than 320 centers, benefits from proximity to Silicon Valley and local talent. Those locational advantages explain clustering even when environmental metrics are worse than in other areas. Industry says latency, connectivity, and workforce matter—and they do.
Specific environmental problems in cluster areas
Virginia isn’t water-stressed, but the state’s clean-energy goals could be derailed if an expanding data-center sector consumes large amounts of electricity from fossil-heavy sources. California faces a different constraint: chronic water shortage. Governor Newsom vetoed a bill requiring data centers to disclose water use, arguing the state can support development—but environmental advocates see an escalating risk. Arizona, with more than 160 centers, is already flagged in the paper for severe water scarcity. In short: clustering in familiar places concentrates risk.
Why continued concentration causes trouble
Fengqi You uses a useful image: when everyone shops at the same grocery store at once, even the best store gets overwhelmed. The same happens when data centers pile into the same grid and watershed. Exceeding natural resource capacity is not a distant possibility—it’s a near-term hazard if buildout continues unchecked in stressed regions. No, market convenience does not erase physical limits.
States on the rise and the current on-the-ground picture
Texas is already the second most popular state for data centers; construction there quadrupled between 2023 and 2024. Montana, Nebraska, and South Dakota have far fewer installations now, but their counts are increasing. A Los Angeles firm plans a hyperscale campus in South Dakota; Nebraska already hosts 39 centers including Meta and Google projects. Despite that progress, these states still lag the established hubs.
How big could the carbon hit be?
If the grid does not transition to lower-carbon sources while demand for AI compute keeps rising faster than efficiency gains, the study finds U.S. data-center expansion could add up to roughly 44 million tons of CO2-equivalent per year. To give that a frame: that’s more than the annual emissions of countries like Hungary, Portugal, or New Zealand in 2022. That is a lot of new emissions tied to one sector’s buildout.
Key uncertainties that change the outlook
Several unknowns can push outcomes in either direction: improvements in model efficiency, better cooling systems, broader adoption of on-site renewables, changes to the grid mix, or even an AI investment slowdown. Noman Bashir at MIT warns that data centers could install self-contained generation (natural-gas plants, solar+battery) or benefit from new power sources such as advanced nuclear. These moves could shift emissions far more than location alone. So the model’s numerical projections are sensitive to policy and technology choices.
Political choices matter—big time
Power-policy direction is decisive. Nebraska, for example, has excellent wind potential but utilities invested heavily in natural gas this year. The current federal posture has been favorable to fossil fuels, which can lock in higher emissions if data centers rely on those supplies. Conversely, a strong policy push for renewables and transmission buildout would lower the emissions cost of nearly any site. No single factor will decide outcomes—policy and market moves combined will.
What responsible siting and policy would look like
If the goal is to reduce harm while letting the economy gain the benefits of AI investment, consider these measures:
- Site servers in states with cleaner grid mixes or realistic, funded plans to decarbonize their electricity supply.
- Prioritize locations with low water stress or require dry-cooling and other low-water technologies where water is scarce.
- Require transparent, standardized disclosure of energy and water use—like nutrition labels for data centers—so stakeholders can compare projects honestly.
- Tie tax incentives and approval to commitments: for example, tempo-limited permits that require transitions to renewables or to off-grid low-carbon resources over time.
- Invest in grid transmission and storage where buildout will cause local congestion; that reduces reliance on local fossil generation for new loads.
How companies can reduce their footprint today
Tech firms can lower impacts by combining better siting with technology choices: more efficient AI models, improved cooling (liquid or immersion where appropriate), on-site renewables plus batteries, and demand-management strategies that shift loads to cleaner hours. Again: where you place capacity matters, and how you design the facility matters. The two levers together multiply benefits.
What the study’s critics say—and where they agree
Outside experts caution against overreliance on exact numbers from a fast-moving sector. Bashir points out that on-site generation and future power technologies could change emissions outcomes materially. Yet there is consensus on the paper’s central message: net-zero claims are fragile if companies keep building massive facilities without transparent accounting and without ensuring clean power for them. Google and Microsoft themselves have acknowledged in sustainability filings that AI makes reaching net-zero harder. That collective admission is social proof that the problem is real.
Practical trade-offs and political reality
Industry chooses locations for clear business reasons—latency, fiber access, workforce, and incentives. Those same reasons explain why Virginia and California remain hubs. The question is not whether business should pursue those advantages, but how to align private decisions with public goods. Policymakers can make that alignment easier by offering incentives for cleaner siting and denying generous breaks for projects that lock in fossil-based power or stress water supplies.
What a sensible short-term plan looks like
Here is a practical sequencing to reduce risk while allowing the sector to scale responsibly:
- Immediate transparency: mandatory, standardized disclosures of projected and actual energy and water use for large projects.
- Conditional permitting: link tax breaks and fast-track approvals to credible decarbonization and water-management plans.
- Investment in transmission: fund lines to move renewable power from wind- and solar-rich states to high-demand regions.
- Support for cooling R&D: public-private partnerships to make low-water, high-efficiency cooling cheaper and faster to deploy.
What I’d tell a governor or CEO in straight terms
If you are a state leader: do you want the short-term tax revenue and jobs from a data-center boom that may leave you with higher electricity demand and harder clean-energy targets? If you are a CEO: do you want to lock your facilities into fossil-heavy power because it was easiest to site, or pay a little more now to avoid regulatory and reputational costs later? These are trade-offs. They are political choices, not natural laws.
How to read the numbers without panic
No model predicts every future twist. Improvements in efficiency or breakthroughs in power technology could cut the projected harm. Likewise, if demand for AI compute plateaus, some planned capacity may never come online. But the study’s real value is not the single top-line number; it is the comparative analysis showing which locations reduce combined water-and-carbon risk. That comparison should inform policy and investment decisions today.
A short checklist for better siting decisions
Before greenlighting a new hyperscale campus, ask:
- Does the local grid have a credible path to low-carbon power for the expected lifetime of the facility?
- Is freshwater abundant, or can the facility be designed to operate with minimal water?
- Are incentives tied to verifiable decarbonization and water plans?
- Can local transmission or storage be expanded so the facility doesn’t force use of local gas peaker plants?
Final synthesis
The Nature Communications study gives a clear, practical takeaway: siting choices matter. Texas, Montana, Nebraska, and South Dakota appear to offer the most favorable balance of energy and water factors today. Virginia and California remain dominant for business reasons, but concentrated growth there risks higher emissions and water stress. No single policy will fix this. No, net-zero promises, left unchecked, are unlikely to hold if buildout continues without transparency and conditional incentives. Yet the problem is tractable: better siting, cleaner grids, improved cooling, and honest disclosures will bend outcomes toward lower impact.
Questions for readers and policymakers
If you were deciding where to permit a $1 billion data-center campus, what trade-offs would you accept between jobs, tax revenue, and environmental limits? If you are in industry: what would make you commit publicly to a phased transition to clean power for new projects? …What evidence would convince you that a given state can host large-scale compute with limited harm? These are open questions. They require negotiation among business, regulators, and communities—so let’s talk about the trade-offs honestly.
#DataCenters #AIInfrastructure #EnergyPolicy #WaterSecurity #Sustainability #NetZero #CornellResearch #NatureCommunications
Featured Image courtesy of Unsplash and Đào Hiếu (2snX_tffBxM)
