Environment

AI Data Centers Are Driving Fast-Tracked Gas Plants With Limited Public Scrutiny

At least 57 proposed or developing private plants totaling roughly 73,000 megawatts are testing whether the AI boom will outpace environmental review and community oversight.

By Serena Tao · June 16, 2026
Email Reporter
AI Data Centers Are Driving Fast-Tracked Gas Plants With Limited Public Scrutiny
CGN News / Cook Global News Network / Environment / All Rights Reserved

CHICAGO | The artificial-intelligence boom is driving plans for at least 57 off-grid power plants intended primarily to serve individual data centers, a buildout totaling roughly 73,000 megawatts that is moving through accelerated approval processes with limited public scrutiny in several states.

The plants, most of them fueled by natural gas, are proposed or under construction as technology companies seek electricity faster than utilities can build transmission and generation. Reuters found that more than a dozen projects had been approved in less than a year and that some state procedures allow decisions within weeks.

Supporters say private generation allows data centers to expand without pushing the full cost onto ordinary utility customers. Critics argue that rushed permits, confidentiality agreements and fragmented oversight can leave communities with air pollution, emergency risks and little time to influence projects large enough to power cities.

Why data centers want their own power

AI systems require enormous computing capacity. Data centers run servers, cooling systems, storage and network equipment continuously. Their demand can arrive faster than a regional grid can add substations, transmission lines and new generation.

A company that builds or contracts for an onsite plant can begin operations without waiting years for utility upgrades. It can also design power around a specific campus and use the public grid mainly for backup or balancing.

That speed has economic value. Technology companies are competing for customers and model capacity. States are competing for investment and tax revenue. The pressure can make ordinary planning and public review appear too slow.

Off-grid does not mean no public impact

A plant serving one data center still uses land, water, pipelines and air. It may connect to roads and emergency services. Emissions cross property lines, and accidents can affect nearby communities.

The label off-grid can obscure those impacts. A project may be electrically separate while depending on public infrastructure or receiving tax incentives. It may also connect to the grid later.

Regulators should evaluate the plant as industrial infrastructure, not merely as an accessory to a private building.

The scale is substantial

Reuters identified about 73,000 megawatts of proposed or developing off-grid capacity. That total is larger than the peak demand of many states and illustrates how quickly AI is reshaping electricity planning.

Not every announced plant will be built, and capacity figures may describe maximum output rather than average operation. Some facilities may run only during high demand or until grid connections are available.

Even a fraction of the total would affect gas demand, pipelines, emissions and equipment supply. Public agencies need a consolidated view rather than evaluating each project in isolation.

Fast-track laws compress public review

Reuters reported that Ohio procedures can allow some decisions within 45 days and without a traditional hearing. Other states have created special pathways or economic-development processes intended to accelerate projects.

Speed can reduce uncertainty, but it can also prevent residents from understanding technical documents, hiring experts or organizing comments. A permit issued before a community knows the full site plan is difficult to challenge later.

Minimum notice periods should reflect project size and risk. A multibillion-dollar campus and power plant should not receive less scrutiny simply because officials want to win an investment competition.

The Bowling Green example

Reuters examined a Meta data-center project near Bowling Green, Ohio, and an associated gas plant developed by Apollo. The data center covered hundreds of acres, and the power project received approval in less than three months, according to the report.

The plant was described as capable of producing enough electricity for about 100,000 homes. Construction activity reportedly began before a draft air permit was issued, highlighting the way separate approvals can move on different schedules.

The companies and regulators should clarify which work was legally allowed before the air permit and which activities depended on final approval. Early construction can create pressure to approve a project because money has already been spent.

Air pollution

Natural-gas turbines emit nitrogen oxides and other pollutants. Nitrogen oxides contribute to ground-level ozone, which can worsen respiratory conditions. Fine particulate pollution can also affect heart and lung health.

The level of impact depends on technology, operating hours, fuel, pollution controls and local conditions. Health claims should be based on modeling and qualified medical or regulatory evidence rather than assumptions about any single plant.

Permits should specify emission limits, monitoring and consequences for violations. Communities should have access to data after operations begin.

Greenhouse-gas consequences

Gas generation emits carbon dioxide, and methane can leak during production and transport. A large buildout could complicate corporate and state climate commitments.

Technology companies often purchase renewable-energy credits or contract for wind and solar. Those measures can support clean generation, but they do not eliminate local emissions from a dedicated gas plant operating around the clock.

Companies should report both direct plant emissions and the assumptions behind any net-zero claim. Future conversion to hydrogen or carbon capture should not be counted as a current reduction unless the technology and supply are actually in place.

Water and noise

Data centers and power plants can use water for cooling, depending on design. In areas facing drought or competing municipal demand, withdrawals require careful assessment.

Turbines, compressors and cooling equipment also create noise. Continuous low-frequency sound can affect nearby residents even when it meets a simple property-line limit.

Permits should address cumulative water use and nighttime noise, not only average annual figures.

Emergency planning

Gas pipelines, turbines, transformers and battery systems create fire and explosion risks that require specialized response. Local departments may need training, equipment and site access.

Companies should fund the additional capability rather than assuming existing taxpayers will absorb it. Emergency plans should identify evacuation zones, communications and coordination with hospitals.

Confidential security information can be protected while still giving residents practical guidance about hazards.

Non-disclosure agreements limit debate

Local officials sometimes sign confidentiality agreements during economic-development negotiations. Companies argue that secrecy protects site selection and business plans. Excessive secrecy can prevent elected bodies from evaluating incentives and impacts publicly.

Once a site is chosen and permits are filed, basic information should be available: owner, capacity, fuel, emissions, water use, incentives and expected employment.

Confidentiality should not cover environmental facts or public obligations. Communities cannot consent meaningfully to a project they are not allowed to understand.

The argument about ratepayers

Private generation can prevent a data center from requiring utilities to build plants paid for by all customers. That is a legitimate benefit if costs truly remain with the company.

The analysis must include pipelines, transmission backup, grid services and incentives. If a facility receives favorable rates or public infrastructure, some cost may still be socialized.

Regulators should require data centers to pay for upgrades caused by their demand and protect other customers if the project closes before costs are recovered.

Jobs and tax revenue

Data centers involve large construction spending but relatively few permanent employees compared with factories of similar size. Power plants add operators and maintenance work, but automation limits staffing.

Officials should compare incentives with verified permanent jobs, wages and tax revenue. Temporary construction work is valuable but should be labeled separately.

Economic benefits should also account for housing, road, water and emergency-service costs.

Competition among states

Federal and state leaders view AI infrastructure as strategically important. They fear that slow approvals will send investment elsewhere. That competition can weaken standards if companies threaten to move.

Strong rules do not have to mean indefinite delay. States can establish clear timelines, complete application requirements and public hearings scheduled early. Predictability benefits companies and residents.

A race to the bottom can leave communities with pollution while tax benefits expire. A race to competent planning would be a more durable advantage.

Alternatives to dedicated gas

Efficiency, demand management, grid upgrades, storage, renewables and advanced nuclear power may supply parts of data-center demand. Each option has cost, timing and permitting challenges.

Companies can design computing workloads to shift some activity to times or places with available clean power. Not every AI task requires identical real-time operation.

Dedicated gas may be the fastest option today, but permits should consider whether a project can transition to lower-emission power and set enforceable milestones.

Cumulative impacts can exceed a permit's frame

A regulator may review one plant against a legal threshold while several projects are planned in the same region. Together they can increase ozone-forming pollution, pipeline demand, water withdrawals and truck traffic beyond what any single application shows.

Cumulative analysis should include existing industrial sources and the health of nearby populations. Communities that already face heavy pollution should not receive additional burden simply because each new project remains within an individual limit.

Regional planning can also identify shared infrastructure or cleaner alternatives. Several data centers might support one efficient grid project rather than building separate gas plants.

Construction and operations need labor standards

Fast schedules can pressure contractors to work long hours around heavy equipment, electrical systems and gas infrastructure. Economic-development agreements should require safety compliance, qualified trades and transparent injury reporting.

Permanent operators need training for turbines, emissions controls and emergency response. The specialized workforce should be part of the planning timeline rather than assembled after construction.

Public support is stronger when projects create skilled local jobs and apprenticeships, not only temporary labor brought from outside.

What regulators should require

Agencies should publish a national or state inventory of data-center generation, cumulative emissions and fuel demand. Project reviews should examine neighboring facilities rather than treating each as isolated.

Permits should require continuous monitoring where appropriate, public reporting, emergency plans and financial assurance for closure. Incentive agreements should include clawbacks if investment or jobs do not materialize.

Residents need enough time and technical assistance to participate. Environmental justice analysis should identify communities already carrying high pollution burdens.

A test of AI's physical footprint

AI is often described as software, but its growth depends on land, electricity, water, minerals and industrial equipment. The off-grid gas boom makes that physical footprint visible.

Private power can accelerate innovation and protect ratepayers if it is designed honestly. It can also move costs outside utility proceedings and hide them in air, climate and public infrastructure.

The choice is not between AI and oversight. It is whether the infrastructure supporting AI will be built under rules that measure its full cost before communities are committed to living with it. The speed of AI competition should not become a reason to approve energy infrastructure before the public can see its emissions, subsidies and emergency obligations or compare them with alternatives that may take longer but impose lower costs over the plant's full operating life and eventual closure, cleanup or conversion to cleaner generation under enforceable public environmental operating permit conditions.

Additional Reporting By: Reuters; U.S. Environmental Protection Agency; state environmental and utility regulators; Meta; Apollo; Data Center Coalition; local-government permitting records.

What This Means

Dedicated generation can help data centers grow without placing every cost on utility customers, but an off-grid plant still affects air, climate, water and emergency services. Fast approval should not mean incomplete review.

Communities should watch permit data, cumulative emissions, public incentives and who pays for pipelines, backup service and eventual closure. Regulators need a consolidated view of the 73,000-megawatt buildout rather than isolated project files.

Advertisement
Advertisement
Sponsored placement