Data centers are criticized as "bad" because they concentrate enormous electricity demand, water consumption, land use, and local air pollution into single sites, and because the grid powering them still runs largely on fossil fuels. The U.S. Energy Information Administration reports that U.S. data centers consumed roughly 4 percent of national electricity in 2023, a figure projected to reach 6 to 12 percent by 2028. Those are real burdens, and they are also partly manageable ones.
The efficiency paradox: why scale cuts cost but raises total demand
The core mechanism is not waste in the naive sense. Modern hyperscale facilities are among the most energy-efficient buildings ever constructed, with power usage effectiveness (PUE) ratios near 1.1 to 1.2, meaning almost all electricity reaches computing equipment rather than cooling or losses. A legacy enterprise server room often ran at PUE 2.0, wasting half its power on overhead. The problem is Jevons paradox at industrial scale: cheaper, more efficient computing invites vastly more computing. Total demand rises even as per-unit efficiency improves.
Inside the rack, most electricity becomes heat by design. A processor converts nearly all its input power into thermal output, so every megawatt of computation requires roughly a megawatt of cooling, minus what free-air cooling and liquid loops can offset. Cold-climate siting and evaporative cooling reduce that load but shift it to water. The efficiency gains are genuine; the aggregate curve is still climbing.
Quantified context: power, water, and grid pressure
A single large hyperscale campus can draw 100 to 300 megawatts continuously, comparable to a small city. The International Energy Agency estimates global data center electricity use at roughly 415 terawatt-hours in 2024, about 1.5 percent of world electricity, projected to more than double by 2030 under current trends. In the United States, utility forecasts have repeatedly revised load-growth projections upward since 2023, largely citing data centers and electrification.
Water matters too. Evaporative cooling can consume millions of gallons per day at large sites; operators report water usage effectiveness figures in the range of 0.2 to 1.5 liters per kilowatt-hour depending on climate and design. Land use is comparatively small per megawatt but concentrated, and backup diesel generators create localized air-quality permits that draw community scrutiny.
| Cooling approach | Typical PUE range | Water use | Grid and siting trade-off |
|---|---|---|---|
| Evaporative (open loop) | 1.1–1.25 | High | Cheapest in hot, dry regions; strains local water |
| Closed-loop with chillers | 1.2–1.4 | Very low | Higher electricity; works anywhere |
| Free-air (cool climates) | 1.1–1.2 | Low | Needs favorable climate; seasonal limits |
| Direct-to-chip liquid | 1.05–1.15 | Low to moderate | Retrofit cost; supports high-density AI racks |
| Immersion liquid | 1.02–1.1 | Very low | Immature ecosystem; maintenance complexity |
The time course: how a campus becomes a local issue
Public opposition rarely arrives at once. It builds through a predictable sequence as promises meet grid reality.
| Phase | Typical timeframe | What happens |
|---|---|---|
| Announcement and incentives | Month 0–6 | Tax abatements, land deals, job projections |
| Load-study and grid queue | Year 1–3 | Utility upgrades, transmission delays surface |
| Construction and backup permits | Year 2–4 | Diesel generator air permits trigger comment periods |
| Energization and ramp | Year 3–5 | Actual draw approaches contracted load |
| Rate-case adjustment | Year 3–7 | Other customers absorb or contest shared upgrade costs |
| Water and noise complaints | Ongoing | Local hearings, moratoria, litigation |
| Renegotiation or exit | Year 5+ | Contracts revised; stranded-asset risk debated |
The counterintuitive detail: much of the backlash lands years after the political win, when ratepayers see bills and residents hear generators. A professional caveat worth stating plainly: contracted capacity and actual consumption often diverge, and utilities plan for the higher number. That conservatism protects reliability but can overstate near-term load in public debate.
Misconceptions: what the criticism gets wrong
The first misconception is that data centers "waste" most of their power. At PUE near 1.1, overhead is small; the electricity is doing computation the economy demands. The second is that they employ few people and therefore add nothing. Direct employment is modest, often 50 to 200 permanent staff per large campus, but the tax base and fiber infrastructure can be substantial.
The third misconception conflates the facility with the workload. A data center hosting video calls, medical records, and grid-balancing software is not the same public good as one training generative models. Criticism aimed at "data centers" often targets specific workloads, and policy that ignores the distinction is blunt.
Bullet list of red flags worth watching in local proposals:
Projected load exceeds available transmission capacity with no funded upgrade plan. Water rights are senior to, or conflict with, residential or agricultural supply. Diesel backup permits allow unlimited testing hours without noise or emissions limits. Tax abatements extend beyond the depreciation life of the equipment. Utility rate design lets a single customer's upgrade costs flow to all ratepayers. Waste-heat reuse is promised but no pipe, customer, or contract exists.
Practical actions: how to evaluate a claim or a proposal
For residents and officials, the useful question is not "are data centers bad" but "which burdens are priced, disclosed, and bounded." Ask for the interconnection study and who pays for network upgrades. Request the water usage effectiveness figure and the source (municipal, groundwater, reclaimed). Demand the backup-generator permit terms and enforcement history.
For buyers and operators, concrete thresholds help. Target PUE below 1.2 and water usage effectiveness below 0.5 liters per kilowatt-hour where climate allows. Contract for hourly-matched clean energy rather than annual certificates if the goal is grid impact. Pursue waste-heat offtake agreements with district heating or greenhouses before construction, not after.
Cross-check thresholds explicitly. A "carbon-neutral" claim may rest on annual renewable energy certificates, while the local grid at 6 p.m. in July is still gas-fired. A "water-positive" claim may offset a different watershed than the one stressed. Reconcile units: megawatts of capacity, megawatt-hours of consumption, and liters per kilowatt-hour measure different things, and proposals often blur them.
Limitations and what remains uncertain
Forecasts for data center load are unusually wide because AI demand, chip efficiency, and model architectures are moving fast. The IEA and EIA projections differ by factors of two for 2028, and utility forecasts have been revised repeatedly. Water data are self-reported and inconsistently scoped. Local air-quality impacts from backup generators are real but poorly monitored in many jurisdictions. The honest position is that the mechanism is clear, the direction is upward, and the magnitude is genuinely contested.
This article is general educational information about energy and infrastructure, not engineering, legal, or investment advice. It does not promise outcomes or returns. For decisions about a specific project, permit, or contract, consult a qualified local engineer, attorney, or financial professional.