The Hidden Environmental Cost of the Cloud

A data centre is a machine for turning electricity into heat, and then getting rid of the heat.

Its power demand is measured in megawatts, the same unit used for power stations. A single large campus can draw as much as a small city. The industry's appetite in Virginia has grown fast enough that the state's main utility has repeatedly raised its forecast for future demand, and has delayed the retirement of fossil fuel plants to meet it.

Its water demand is less often reported. Many data centres use evaporative cooling, which is efficient in energy terms and consumes water — water that evaporates and does not return to the river or aquifer. In a dry summer, that competes directly with household supply.

There is also the land. Server halls are low-density buildings surrounded by parking and security fencing, and they generate almost no employment per acre. Counties that approve them gain tax revenue and lose developable land.

And there is noise. Cooling equipment and backup generators run continuously, and the low-frequency hum travels further than the sound of traffic. Residents in the towns next to these campuses have complained of sleep disturbance for years, and in some cases have taken their objections to county boards.

The carbon picture is complicated by the grid. A data centre's emissions depend on what supplies its power. As demand grows faster than renewable construction, utilities fill the gap with what they have.

The complaints from people who live near these buildings are consistent and specific.

It is the hum at two in the morning. It is the diesel generators tested monthly, which can be heard and smelled across a neighbourhood. It is the substation built at the end of a road, and the transmission lines widened to carry power that goes somewhere else.

It is also the money. Data centres pay property taxes, and in a county with a limited commercial base that revenue funds schools and services. Some residents who dislike the noise accept the trade. Others ask why they should bear the burden of a facility whose service they use but whose power lines cross their back garden.

Farmers describe a different cost. Transmission corridors take land and divide fields. Solar projects built to supply the grid compete with farming for the same acreage. And the water question, in a region that has spent decades arguing over a shared river, is unsettled.

The contributor's own experience of the noise and proximity of these facilities is testimony. The energy and water figures used here come from published forecasts and regulatory filings, and are cited separately.

The cloud is a metaphor that hides a building, and that matters for how people reason about it.

When a service is described as invisible, its costs become invisible too. Research on how people judge consumption finds that impacts which are distant, hard to picture and not attached to a product are systematically underestimated. Nobody sees a data centre when they watch a video, so nobody weighs it.

There is also a distribution question about attention. Noise complaints are made by a small number of residents, who must be persistent and organised to be heard at a county meeting at seven in the evening. The benefits are diffuse and mostly invisible. In local politics, intensity beats breadth.

Trust in the industry follows the pattern found in other risk disputes: it depends less on the accuracy of a company's figures than on whether the figures can be checked and whether the company admits uncertainty. Confidential rate agreements and voluntary water reporting tend to reduce trust rather than increase it.

Finally, there is a fairness instinct that people apply consistently even where they cannot name it. Asked whether a facility should pay for the infrastructure it needs, almost everyone says yes. Asked whether they want the tax revenue, almost everyone says yes to that too. The conflict is not hypocrisy; it is two legitimate claims held at once, which is exactly what local decisions have to resolve.

Three groups carry different parts of the cost. Residents near the buildings carry noise, traffic and visual impact. Electricity customers carry the grid costs, through rates. Communities near power plants carry the emissions.

Only the first group receives anything in return, in the form of local tax revenue. The second and third pay without compensation, and the second group has no ability to choose a different supplier.

The distributional question is sharpest in water. In a basin where allocations have been fought over for decades, an industrial user with capital and lawyers is in a stronger position than a household with a meter.

Procedural justice is thin. Land use decisions are public, but the technical evidence behind them — noise modelling, water estimates, load forecasts — is often supplied by the applicant and rarely contested. Rate cases involving large customers can be partly confidential.

There is also a question of scale. A single county approving one data centre is making a local decision. Fifty counties approving fifty is a regional transformation, and no institution is responsible for the sum.

Recognition matters here too: the workforce that builds and runs these facilities is largely invisible in the debate, though its safety and working conditions are part of the story.

No single body decides how much data centre capacity a region can carry. Counties decide land use, the utility decides supply, the grid operator decides transmission, the state regulator decides who pays, and the state environmental agency decides air and water permits. Each is competent within its own boundary.

The consequences are visible. Virginia's utility has revised its demand forecast upward repeatedly and delayed plant retirements. Counties have adopted noise ordinances after facilities were built. Water reporting remains patchy. Rate cases have become the venue where cost allocation is fought, because it is the only venue that sees the whole system.

The industry has also done things it did not have to. Some operators have signed long-term renewable contracts, some have switched to closed-loop cooling that uses far less water, and some have funded local energy efficiency. Those are choices, not obligations.

What is missing is coordination with a published limit: an assessment of how much capacity a county or a grid region can host given its water, its land and its carbon budget. That would require counties, the utility and the state to plan together, and none of them is currently required to.