The Data Center Boom and Who Pays the Energy Bill

Somewhere in rural Virginia, a farmer watches construction crews pour the foundation for a data center the size of several football fields. The county offered a twenty-year tax abatement to attract the project. The power company is building a new substation to feed it. The farmer's electricity bill went up last year to help cover grid upgrades. The data center, when it opens, will employ about fifty people.

This scene is repeating across the United States and around the world. The AI industry's appetite for compute translates directly into an appetite for electricity, water, and land — resources drawn from communities that will see little of the wealth generated inside those windowless buildings.

The Building Boom

The numbers are staggering. Microsoft announced over $80 billion in data center spending for fiscal year 2025 alone. Meta committed $60 billion. Amazon and Google have each pledged tens of billions more. In total, the major technology companies are pouring more than $200 billion into data center construction in the 2024-2026 period (IEA, 2024).

This isn't speculative investment. It's a direct response to the compute demands of generative AI. Training a single frontier model requires thousands of GPUs running for months. Inference — actually serving responses to users — requires even more aggregate compute, because it runs continuously at scale. Every ChatGPT query, every AI-generated image, every automated customer service interaction draws power from a data center somewhere.

The scale is accelerating. Data centers that once measured their capacity in megawatts now measure it in gigawatts. Microsoft's planned campus in Mount Pleasant, Wisconsin would consume more electricity than some small countries.

The Energy Appetite

A single ChatGPT query consumes roughly ten times the electricity of a standard Google search (IEA, 2024). That ratio matters because of the volume: hundreds of millions of AI queries per day, growing rapidly.

The International Energy Agency projects that global data center electricity consumption could reach 1,000 terawatt-hours by 2026. To put that in perspective, it is roughly equal to Japan's total electricity consumption — the world's fifth-largest economy, 125 million people, every factory and home and train (IEA, 2024).

This projection may already be conservative. It was made before the latest wave of AI investment announcements. Goldman Sachs estimates that data center power demand in the US alone will grow 160% by 2030 (Goldman Sachs, 2024). Grids that were designed for gradual demand growth are suddenly facing exponential load increases.

The electricity has to come from somewhere. In the United States, that "somewhere" is increasingly natural gas. Despite corporate pledges to operate on renewable energy, the surge in demand is outstripping renewable supply. New gas-fired power plants are being built specifically to feed data centers. The US Energy Information Administration projects that power sector carbon emissions, which had been declining, will flatten or increase through the late 2020s as data center demand absorbs new renewable capacity that would otherwise have displaced fossil fuels (EIA, 2025).

The AI industry's climate impact is not hypothetical. It is measurable, growing, and running counter to decarbonization goals.

Water: The Hidden Cost

Electricity is not the only resource data centers consume. Cooling thousands of processors generates enormous heat, and the most common cooling method uses water — vast quantities of it.

A single large data center can consume millions of gallons of water per day. Google's own environmental reports revealed that its data centers consumed 5.6 billion gallons of water in 2023, a 17% increase over the previous year, driven largely by AI workloads (Google, 2024). Microsoft reported a 34% increase in water consumption over the same period.

In arid regions, this creates direct conflict. Data centers in Arizona, Texas, and parts of the American West are drawing from the same stressed aquifers and municipal systems that serve agriculture and residential users. In The Dalles, Oregon — a town of 15,000 that hosts major Google data centers — concerns about water consumption have become a persistent source of community tension. The water that cools servers is water that does not irrigate crops, fill reservoirs, or flow through taps.

As Kate Crawford documents in Atlas of AI, this is a pattern that repeats across the material infrastructure of computing: the physical costs are borne locally while the economic returns flow elsewhere. The "cloud" is not ethereal. It is concrete, steel, water, and wattage, situated in specific places and drawing on specific resources (Crawford, 2021).

Where the Data Centers Land

Data center developers seek three things: cheap electricity, available land, and cooperative local governments. Rural communities often have all three. They also tend to have limited economic alternatives and less political power to negotiate terms.

The incentive packages are substantial. Tax abatements lasting ten to twenty years. Reduced utility rates. Expedited permitting. Infrastructure built at public expense. Virginia's Loudoun County — "Data Center Alley" — has become the densest concentration of data centers on Earth, and the county has granted billions in tax relief to attract them.

The promised benefits are jobs and tax revenue. The reality is more complicated. Data centers are among the most capital-intensive and least labor-intensive facilities ever built. A billion-dollar data center might employ 30 to 50 permanent workers. Compare that to a manufacturing plant of similar cost, which might employ 500 to 1,000. The jobs that do exist tend to be specialized — requiring skills that local workers may not have.

What communities do get: noise from industrial cooling systems. Diesel generators for backup power. Truck traffic during construction. Elevated water and electricity demand that strains local systems. Visual blight from massive, featureless structures. And the knowledge that the facility's primary economic output — compute cycles — generates wealth captured almost entirely by shareholders in San Francisco, Seattle, and New York.

The Nuclear Gambit

Faced with the challenge of powering AI at scale, technology companies have turned to an unexpected source: nuclear energy.

Microsoft signed a twenty-year power purchase agreement with Constellation Energy to restart a reactor at Three Mile Island — yes, the site of America's worst nuclear accident. Google announced agreements to purchase power from Kairos Power's small modular reactors (SMRs), a technology that does not yet exist at commercial scale. Amazon signed deals with Talen Energy for nuclear-adjacent power in Pennsylvania.

The logic is straightforward, and in fairness, it addresses a real problem. Nuclear provides carbon-free baseload power — exactly what data centers need. Unlike solar and wind, it runs around the clock regardless of weather. And unlike new renewable installations, a restarted reactor or a dedicated nuclear facility can provide power directly to a single customer without competing for grid capacity. The technology companies pursuing nuclear are, at minimum, taking the energy problem seriously rather than papering over it with offsets.

But the implications are worth examining. When a technology company contracts the entire output of a power plant, that electricity is no longer available to the broader grid. If Microsoft buys Three Mile Island's output, Pennsylvania ratepayers lose access to that carbon-free generation. The grid becomes more dependent on fossil fuels to serve everyone else. The environmental benefit accrues to Microsoft's sustainability report; the environmental cost is distributed across the public.

This is redistribution in its most literal form: clean energy flows to those who can pay the most, while the communities around them absorb the emissions from whatever fills the gap.

Environmental Justice

The distributional pattern is not random. Research consistently shows that data centers — like other heavy industrial infrastructure — disproportionately affect lower-income communities and communities of color.

A 2024 analysis by the Energy Justice Lab found that census tracts hosting data centers in Virginia had median household incomes 15% below the state average and higher proportions of non-white residents (Raimi et al., 2024). This mirrors the pattern Virginia Eubanks identified in Automating Inequality: the systems that automate and optimize modern life impose their heaviest costs on those with the least power to resist (Eubanks, 2018).

The communities receiving data centers rarely have meaningful input into the decision. Deals are often negotiated between developers and state or county officials before residents learn what is planned. Public comment periods, where they exist, occur after key decisions have been made. The asymmetry is structural: technology companies have legal teams and lobbyists; affected communities have public meetings.

The Redistribution Ledger

The data center boom presents a clear case of externalized costs and concentrated returns.

Who pays:

• Local communities bear the environmental burden — noise, water depletion, visual impact, and grid strain.
• Utility ratepayers fund grid upgrades through higher electricity bills, regardless of whether they use AI services.
• Taxpayers finance the incentive packages that attract data centers.
• The atmosphere absorbs the carbon emissions from fossil-fueled generation that fills the gaps left by corporate clean energy purchases.
• Future generations inherit the climate consequences of an energy-intensive technology deployed at planetary scale.

Who benefits:

• Technology companies capture the value generated by AI services.
• Their shareholders receive the returns.
• Consumers of AI products receive convenience and capability — though they may not realize its environmental cost.
• A small number of specialized workers earn high salaries.

The imbalance is stark. The AI industry is executing one of the largest transfers of environmental costs from corporations to communities in recent history. The externalities are socialized; the returns are privatized.

What Could Be Different

This arrangement is a product of choices, not inevitability. Different policies could distribute both the costs and benefits more equitably.

Renewable energy requirements. Rather than allowing data centers to claim renewable energy through offsets and certificates while actually running on fossil-generated grid power, jurisdictions could require direct renewable generation as a condition of permitting. Some European countries already do this.

Community benefit agreements. Data center developers could be required to negotiate binding agreements with affected communities — not just tax incentives for the county, but direct benefits for residents: reduced utility bills, community funds, workforce development, noise mitigation, and water conservation measures.

Energy transparency reporting. Technology companies should be required to disclose the actual energy source, carbon intensity, and water consumption of their AI services on a per-facility and per-product basis. Current sustainability reporting is aggregated to the point of meaninglessness.

Grid impact fees. When a data center imposes costs on the local grid, those costs should be borne by the developer, not distributed across all ratepayers. The price of AI should reflect its true infrastructure costs.

The Nordic model. Sweden, Norway, Finland, and Iceland have attracted data centers by offering genuinely renewable energy — hydro, geothermal, wind — rather than the fossil-heavy grids of Virginia and Texas. Cold climates also reduce cooling requirements, cutting water consumption dramatically. If AI must consume vast amounts of electricity, there are better and worse places to do it. Policy can steer these decisions. The Nordic experience demonstrates that good infrastructure siting is achievable when governments set clear standards rather than racing to the bottom on incentives.

The Question of Scale

There is a deeper question beneath the policy specifics: should we be building this much infrastructure this fast?

The current data center boom is premised on the assumption that AI demand will continue to grow exponentially and that the economic returns will justify the environmental costs. Both assumptions may prove correct. They may not. The history of technology investment includes plenty of overbuilding — fiber optic networks in the late 1990s, for instance — where supply dramatically outran demand.

If the boom slows, communities will be left with massive industrial facilities that consume resources but generate less revenue than projected. The tax abatements will already have been granted. The grid upgrades will already have been paid for. The water will already have been drawn.

Whether the bet pays off or not, the same communities will have borne the costs. That asymmetry — between those who take the risks and those who bear them — is the redistribution problem at its most concrete.

The cloud is not in the sky. It is on the ground, consuming water and electricity, generating noise and heat, built on land that used to be something else. The question is not whether we want AI. It is whether the people who live next to its physical infrastructure should subsidize its construction while the profits flow elsewhere.

Related

• Compute Is the New Oil
• Hardware Bottleneck
• Redistribution Stack
• Environmental Cost of AI
• Magnificent Seven
• AI Won't Solve Climate Change
• Edge AI
• Who Controls Supply Chain
• Frontier Lab Oligopoly

Sources

• Crawford, Kate. Atlas of AI. Yale University Press, 2021. Publisher
• EIA (US Energy Information Administration). "Annual Energy Outlook 2025." 2025. EIA
• Eubanks, Virginia. Automating Inequality. St. Martin's Press, 2018. Publisher
• Goldman Sachs. "AI, Data Centers, and the Coming US Power Demand Surge." 2024. Goldman Sachs
• Google. "2024 Environmental Report." 2024. Google Sustainability
• IEA (International Energy Agency). "Electricity 2024: Analysis and Forecast to 2026." 2024. IEA
• Raimi, Daniel et al. "The Environmental Justice Implications of Data Center Siting." Resources for the Future, 2024. RFF