Large computing facilities are drawing organised opposition over electricity, water and land use. The argument has moved from local planning meetings to state government, where energy pricing, permits and tax incentives are actually decided.
Key takeaways
- Data centres have become one of the most contested categories of industrial development in the United States, largely because of the amount of electricity and land they require.
- The growth of artificial intelligence workloads has increased demand for computing capacity, which in turn increases the pressure on regional electricity grids.
- Opposition is no longer confined to individual neighbourhoods, and is increasingly directed at state-level decisions on tax incentives, utility regulation and permitting.
- The central financial dispute is who pays for new generation and transmission: the companies operating the facilities, or the wider base of electricity customers.
- Very little about any individual project is publicly verifiable in detail, because siting agreements, power contracts and incentive packages are often confidential.
What is actually happening around data-centre projects
Across many parts of the United States, proposals to build large computing facilities are meeting organised resistance. The objections cluster around a small set of themes: the electricity a site will draw once operating, the water some cooling systems consume, the noise generated by cooling equipment, the amount of land taken up by low-rise buildings and substations, and the tax concessions sometimes offered to attract the investment.
What has changed is the venue. Individual projects are approved or refused by county boards and municipal planning bodies, and for years that is where disputes stayed. Increasingly, campaigners argue that the decisions which matter most are made higher up: how a state regulates its utilities, how the cost of new transmission lines is allocated among customers, whether large industrial users pay a distinct tariff, and what conditions attach to economic development incentives. Those are state functions, which is why the argument has arrived at state capitols. The result is a set of policy questions about energy and taxation that sit uncomfortably alongside the same governments’ interest in attracting technology investment.
Why the argument has surfaced now
Two pressures have converged. The first is demand. After a long period in which electricity consumption in the United States grew slowly, utilities in several regions have begun forecasting significant load growth, and large computing facilities are a visible part of that forecast. Whether they are the dominant cause is disputed; electrification of vehicles and heating, and the return of some manufacturing, are also cited.
The second is price. Electricity bills have risen in many areas for reasons that include fuel costs, grid maintenance, and the cost of hardening networks against extreme weather. When bills rise at the same time as a large new industrial customer connects nearby, the connection between the two is easy to draw and hard to disprove without detailed regulatory data. That combination — visible construction, rising bills, and limited public information about the contracts involved — is what turns a technical planning matter into a political one. Election cycles then amplify it, because energy costs are a subject on which voters have direct experience.
The background a newcomer needs
A data centre is a building that houses computing and storage equipment, along with the electrical and cooling systems needed to keep it running continuously. Most of the internet’s everyday services — email, streaming, business software, payment processing — run inside such buildings, and they are not new. What is new is the scale of the largest recent projects and their power density.
Training and running large artificial intelligence models uses specialised processors that draw considerably more power per rack than conventional servers, and generate correspondingly more heat. That heat has to be removed, either by air cooling, which uses more electricity, or by water-based systems, which can use less electricity but consume or evaporate water. The choice of cooling method varies by site and climate.
Connecting a very large customer to the grid is not simply a matter of running a cable. It can require new substations, upgraded transmission lines, and sometimes new generation. Those assets take years to build and are paid for through regulated rates, which is the mechanism at the centre of the current dispute.
Who is affected, and how
Residential and small-business electricity customers are affected if the costs of grid expansion are spread across all users rather than charged to the facility that triggered them. Whether this is happening in any particular case depends on the tariff structure approved by that state’s regulator, and it is not uniform.
Communities near a proposed site are affected by construction traffic, by the appearance of large windowless buildings and switchyards, and in some cases by the low-frequency noise of cooling plant. Local governments are affected in both directions: the facilities can generate substantial property tax revenue but relatively few permanent jobs compared with other industrial developments of similar size.
Utilities face the difficult position of being obliged to serve new load while avoiding over-building capacity that customers would still pay for if a project were cancelled or downsized. Technology companies face rising costs and lengthening timelines for the capacity their services require. Workers in construction and the electrical trades benefit from the building phase, which is generally short relative to the life of the site.
Where informed people disagree
There is genuine disagreement among people who follow this closely, and it is not simply industry against opponents.
One dispute concerns forecasting. Utilities receive interconnection requests from developers who may be applying at several sites simultaneously for a project they will build only once. Some analysts therefore argue that headline demand figures overstate what will actually be constructed. Others argue that under-building is the greater risk, because grid infrastructure takes far longer to deliver than a building does.
A second dispute concerns efficiency. Improvements in chip design and cooling reduce energy use per unit of computing, and one school of thought holds that this will moderate total demand. The counter-argument is that cheaper computing tends to be met with more use of it.
A third concerns economic value. Supporters point to tax revenue, construction employment and the infrastructure that underpins modern services. Critics question whether the incentives offered are proportionate to the permanent employment created. Both positions rest partly on figures that are commercially confidential.
The practical implications
For policy, the effect is a shift towards rules rather than case-by-case negotiation. The instruments available to states include special tariff classes for very large users, requirements that a new customer fund the network upgrades it necessitates, conditions attached to tax incentives, disclosure requirements for water and energy use, and rules on whether a facility must bring its own generation or storage.
For local government, the practical implication is that approval is becoming slower and more conditional, with agreements covering noise limits, water sourcing, screening and decommissioning.
For companies planning capacity, the implication is that power availability, rather than land or construction capacity, is now often the binding constraint on where and when a facility can be built. That is already shifting interest towards regions with spare generation, and towards arrangements that pair sites with dedicated supply.
What to watch next
Several signals will indicate the direction of travel. The first is regulatory: whether state utility commissions approve distinct rate classes for very large customers, and what conditions they impose. These proceedings are public, though technical.
The second is legislative. Bills addressing disclosure of data-centre energy and water use, the terms of tax abatements, and cost allocation for grid upgrades are the clearest indicator of whether the backlash is producing durable rules or a temporary slowdown.
The third is corporate behaviour: whether announced projects proceed on their stated timetables, whether developers withdraw applications in contested areas, and whether more sites are paired with their own generation or storage.
The fourth is the demand data itself. Actual metered consumption, published by regional grid operators over time, will eventually show whether current forecasts were accurate. Until then, most confident claims in either direction should be treated as projections rather than facts.
Frequently asked questions
Why do data centres use so much electricity?
They run large numbers of processors continuously, and every watt consumed by that equipment becomes heat that must be removed by cooling systems, which consume further power. Facilities designed for artificial intelligence workloads use processors with higher power draw per unit than conventional servers, so a building of a given size can require substantially more electricity than an older facility of similar footprint.
Do data centres raise household electricity bills?
It depends on how a state’s regulator allocates the cost of new generation and transmission. If a large customer funds the upgrades it requires, the effect on other bills is limited. If those costs enter the general rate base, they are spread across all customers. Bills also rise for unrelated reasons, including fuel prices and grid maintenance, so attributing an increase to one cause is difficult.
Why do some data centres use water?
Water-based cooling can remove heat using less electricity than air cooling, particularly in hot climates. Some systems evaporate water, meaning it is consumed rather than returned; others circulate it in a closed loop. The trade-off is essentially between electricity and water, and the choice varies by site, climate and local conditions. Disclosure of actual water use is not consistent across jurisdictions.
Do these facilities create many permanent jobs?
Construction employs a substantial workforce, but that phase is temporary. Once operating, a large facility is typically run by a comparatively small permanent staff of technicians, engineers and security personnel, because the equipment is designed for remote management and automation. Precise figures vary by operator and site and are not consistently published, which is part of why the economic debate remains unresolved.
Can local residents stop a data-centre project?
Local planning and zoning bodies usually decide whether a specific site may be developed, and organised opposition has led to refusals and withdrawals in various places. However, decisions on electricity tariffs, grid cost allocation and state tax incentives are made by state regulators and legislatures, so local objection alone does not determine the wider economics of where facilities are built.
Is the demand for computing capacity expected to keep growing?
Forecasts vary and should be treated with caution. Interconnection requests submitted to utilities may overstate what will be built, because developers sometimes apply at multiple sites for a single project. Efficiency improvements may moderate consumption per unit of computing, while wider adoption may increase total use. Metered consumption data published over time is the most reliable eventual indicator.
Sources and further reading
- Regional transmission organisations and grid operators, for published load forecasts and interconnection queue data.
- State public utility commissions, whose rate cases and tariff proceedings are matters of public record.
- The US Energy Information Administration, for national electricity consumption and generation statistics.
- Trade and technical press covering data-centre construction, cooling technology and power procurement.
Surfaced from the reddit:technology signal “state-level data-centre backlash”. AI-assisted draft, editorially reviewed.
