The Hidden Pollution Behind the AI Boom: How Data Centers Are Expanding Fossil Fuel Use in Texas

Artificial intelligence is often presented as an almost weightless technology.

People type a question, generate an image, summarize a document, or ask software to analyze enormous amounts of information. The result appears instantly on a screen, with little visible indication of the physical infrastructure working behind it.

But AI does not operate in an invisible cloud.

It runs inside enormous data centers filled with servers, networking equipment, cooling systems, electrical substations, batteries, gas turbines, and backup generators. As demand for AI computing grows, these facilities require increasingly large and uninterrupted supplies of electricity.

In Texas, some developers are responding by building power generation directly beside their data centers.

A recent Floodlight investigation found that at least 38 Texas data centers have received minor air permits for onsite power equipment since 2024, collectively authorizing more than 2,100 diesel backup generators. Those generators are permitted to emit nearly 2,500 tons of nitrogen oxides each year, although actual emissions will depend on how frequently the equipment operates.

The most prominent example is the Stargate AI campus near Abilene.

The site has already installed 10 natural-gas turbines and 62 diesel generators. Its current permits authorize more than 1.6 million tons of annual greenhouse-gas emissions and more than 1,000 tons of combined conventional air pollutants from that equipment. Developers are seeking approval for another 41 turbines and 18 generators, an expansion that could turn the campus into one of Texas’ largest fossil-fuel power facilities.

These are permitted maximums, not measurements proving that the facilities are already releasing those amounts.

That distinction matters.

But the permits reveal how much pollution regulators have legally authorized and how quickly the AI infrastructure boom is becoming an energy and public-health issue—not merely a technological one.

Why AI Data Centers Need So Much Electricity

All data centers use electricity, but advanced AI systems place especially heavy demands on computing infrastructure.

Training a large AI model can require thousands of specialized processors operating continuously for weeks or months. Once the model is deployed, every user request—known as an inference—requires additional computation.

A modern AI campus may contain hundreds of thousands of processors.

Electricity is needed to power:

• Graphics processing units and AI accelerators
• Central processing units
• Data-storage systems
• Networking equipment
• Cooling pumps and chillers
• Lighting and security
• Power-conversion systems
• Battery systems
• Backup generators
• Water-treatment and mechanical equipment

The challenge is not only the total amount of energy consumed.

AI companies also demand extremely high reliability.

A brief power interruption can stop workloads, damage equipment, interrupt customer services, and create expensive recovery procedures. Data centers therefore require several layers of electrical redundancy.

A typical large facility may use:

1. Electricity from the regional grid
2. Multiple utility connections
3. Onsite battery systems for immediate backup
4. Diesel generators for longer interruptions
5. Increasingly, onsite natural-gas generation for backup or primary power

The AI boom is expanding so quickly that utility companies cannot always build substations, transmission lines, and new power plants fast enough to meet requested connection dates.

Developers have responded with a strategy sometimes described as bring your own power.

Instead of waiting years for the grid, they build private energy systems beside the servers.

Why Texas Has Become a Data Center Magnet

Texas offers several advantages for large technology projects.

These include:

• Large areas of relatively affordable land
• A competitive electricity market
• Extensive natural-gas infrastructure
• Major wind and solar resources
• Business-friendly tax policies
• Rapid permitting
• Growing fiber-optic networks
• Proximity to expanding population centers
• Local governments eager for investment

The state also operates most of its electrical system through the Electric Reliability Council of Texas, or ERCOT, which is largely separate from the major eastern and western interstate grids.

That independence can make development flexible, but it also means Texas must manage most supply-and-demand challenges internally.

Data centers can bring construction jobs, tax revenue, infrastructure investment, and permanent technical employment.

They can also create very large new electrical loads in communities that were never designed to support them.

When developers fear that grid capacity will not arrive quickly enough, natural-gas turbines and diesel generators become attractive because they can be installed faster than large transmission projects or utility-scale power plants.

What Floodlight Found Across Texas

Floodlight analyzed Texas air-permitting records for onsite data center power systems.

The investigation found that since 2024:

• At least 38 data centers received minor permits for onsite electricity generation.
• More than 2,100 diesel backup generators were authorized.
• The generators were collectively permitted to emit nearly 2,500 tons of nitrogen oxides annually.
• More than half of the identified facilities submitted projected nitrogen-oxide emissions just below levels that could trigger more extensive review and public participation.

The generators are generally described as emergency or backup equipment.

However, they do not operate only during major blackouts.

Diesel generators must be tested and maintained to ensure that they will work when needed. Depending on the permit and operating conditions, they may also be used during grid emergencies, demand-response events, maintenance periods, or other disruptions.

A single generator operating occasionally may present a manageable local source of pollution.

A campus containing dozens or hundreds creates a different issue.

Even limited operating hours can become significant when multiplied across thousands of engines.

Artificial Intelligence and Machine Learning

Explore the transformative power of Artificial Intelligence and Machine Learning in this article. Discover how these technologies are revolutionizing industries, from healthcare to transportation, and learn about their potential impact on our lives. Understand the importance of ethical considerations and responsible development as we navigate the AI-driven future.

Revlox magazineMaruf Badhon

Nitrogen Oxides Are Not the Same as Nitrous Oxide

Some social-media posts and secondary reports incorrectly refer to the generator pollution as “nitrous oxide.”

The relevant pollutant category is nitrogen oxides, usually abbreviated as NOx.

NOx primarily refers to gases such as:

• Nitric oxide
• Nitrogen dioxide

These are produced when fuel burns at high temperatures.

Nitrous oxide, or N₂O, is a different gas commonly associated with agriculture, industrial activity, anesthetic use, and climate change.

The distinction is important because NOx has specific local and regional health effects.

It contributes to the formation of:

• Ground-level ozone
• Smog
• Fine particulate pollution
• Acid deposition

Exposure can aggravate asthma, reduce lung function, irritate airways, and increase risks for people with respiratory or cardiovascular illness.

Children, older adults, outdoor workers, and people already living in polluted communities can be especially vulnerable.

Why Diesel Generators Are a Public-Health Concern

Diesel engines can emit several harmful pollutants, including:

• Nitrogen oxides
• Fine particulate matter
• Carbon monoxide
• Volatile organic compounds
• Sulfur compounds
• Hazardous air pollutants
• Carbon dioxide

Fine particulate matter is particularly concerning because particles can penetrate deep into the lungs and some can enter the bloodstream.

Diesel exhaust has also been classified as carcinogenic to humans by the International Agency for Research on Cancer.

Modern generator engines may include pollution-control technology, and their emissions can vary significantly based on:

• Engine age
• Fuel quality
• Load
• Maintenance
• Operating time
• Pollution-control equipment
• Temperature
• Testing schedules

The issue is therefore not that every backup generator continuously produces the maximum amount listed in a permit.

The concern is the cumulative effect of installing thousands of engines near homes, schools, workplaces, and already polluted areas without always conducting a comprehensive regional assessment.

The Stargate Campus in Abilene

Stargate is OpenAI’s large-scale infrastructure platform involving partners including Oracle and SoftBank.

OpenAI announced in September 2025 that planned Stargate infrastructure had reached nearly seven gigawatts of capacity and more than $400 billion in expected investment over the following three years. The Abilene campus is the platform’s flagship site.

The Abilene development is being built by Crusoe and related project entities.

The campus is designed to host enormous numbers of advanced computing chips for AI training and operation.

Its physical scale makes its power strategy especially important.

According to permit records examined by Floodlight, the site obtained authorization in 2024 for:

• 10 gas-fired turbines
• 62 diesel backup generators

Those minor authorizations permit the combined equipment to emit:

• More than 1.6 million tons of greenhouse gases annually
• More than 1,000 tons of combined harmful air pollutants annually

Crusoe has stated that the gas turbines are intended ultimately to provide backup power rather than serve continuously as the campus’s primary electricity source. The company also says the turbines use selective catalytic reduction systems capable of sharply reducing nitrogen-oxide emissions compared with conventional uncontrolled equipment.

Both facts can be true:

The turbines may use advanced emissions controls and be intended primarily for backup service.

Their permits can still authorize very large amounts of pollution if their operation expands.

Permitted Emissions Are Not Actual Emissions

Air permits usually establish maximum allowable emissions based on assumptions about equipment, fuel use, operating hours, and pollution controls.

A facility’s actual annual emissions may be substantially below that ceiling.

For example, a backup turbine permitted to operate continuously may in practice run for only limited periods.

This means it would be misleading to claim that Stargate is already releasing exactly 1.6 million tons of greenhouse gases every year merely because its permits allow that amount.

The more accurate statement is:

The currently installed turbines and generators are legally authorized to produce emissions up to those stated levels under their permits.

Permitted totals still matter for several reasons.

They reveal:

• The maximum operation regulators have authorized
• The potential emissions if equipment is heavily used
• Whether the facility could operate beyond emergency conditions
• How the project was classified during review
• The scale of environmental risk accepted by regulators

Communities evaluating a new facility should be able to understand both actual operating plans and the full authority granted by its permits.

Why Minor Air Permits Are Controversial

Texas provides several forms of air authorization.

Minor mechanisms such as permits by rule and standard permits are designed to streamline approval for common sources with predictable emissions.

They can be appropriate for relatively small activities such as:

• A limited number of generators
• Gas stations
• Dry-cleaning operations
• Certain manufacturing equipment
• Small industrial sources

These pathways can avoid the cost and delay of evaluating every routine project through a lengthy custom process.

The controversy arises when a very large facility uses multiple streamlined authorizations for equipment that, taken together, resembles a major power plant.

Floodlight reported that the initial Stargate permits did not require the same environmental studies, public notice, or public-comment process associated with a major source review.

Former regulators interviewed for the investigation argued that a campus containing 10 turbines and 62 generators is fundamentally different from the small projects for which these simplified processes were designed.

TCEQ responded that it issues permits only when they comply with applicable state and federal permitting rules, including legally required public-participation provisions.

The disagreement therefore concerns not only whether the permits were technically available.

It concerns whether the regulatory system is appropriately evaluating the combined scale of modern data center infrastructure.

The “Small First, Big Later” Concern

The most serious criticism involves the sequence in which some data center projects receive authorization.

A developer may begin with equipment covered under minor permits and later apply for a much larger expansion requiring major-source review.

At Stargate, developers obtained the initial authorizations and subsequently applied to add:

• 41 additional natural-gas turbines
• 18 additional diesel generators

If approved, the total would reach:

• 51 gas turbines
• 80 diesel generators

Floodlight reported that the expansion could create one of the largest fossil-fuel power facilities in Texas, with enough generation capacity to supply more than one million homes.

The additional equipment must undergo a more extensive permitting process, including emissions-control requirements and public participation.

Critics argue that once an enormous data center has already been constructed and begun operating, later environmental review becomes less meaningful.

At that stage:

• Large investments have already been made.
• Local governments may depend on anticipated tax revenue.
• Thousands of construction workers may be involved.
• The data center may already require additional power.
• Political pressure to approve expansion becomes stronger.
• Alternatives may be more expensive to introduce.

Former EPA enforcement official Bruce Buckheit told Floodlight that air regulators are generally expected to assess physically and operationally connected equipment as one project rather than dividing a planned major source into smaller pieces.

Whether Stargate’s permitting sequence violates federal aggregation rules would ultimately require formal regulatory or judicial review.

It should not be treated as legally established merely because experts raised the concern.

Why Developers Prefer Onsite Gas Turbines

Natural-gas turbines offer several advantages for data center developers.

They can:

• Be installed faster than major utility infrastructure
• Provide large amounts of power
• Start quickly during an outage
• Operate independently of grid failures
• Support stable frequency and voltage
• Reduce reliance on diesel for long-duration backup
• Use existing Texas gas pipelines
• Supply power while grid connections are delayed

Modern turbines with effective controls can emit less local air pollution than older diesel engines or uncontrolled gas reciprocating generators for the same amount of electricity.

Crusoe says its Abilene turbines use selective catalytic reduction and produce significantly lower nitrogen-oxide emissions than traditional gas or diesel reciprocating engines.

However, “cleaner than diesel” does not mean carbon-free.

Burning natural gas releases carbon dioxide.

Methane can also escape during extraction, processing, and transportation. Methane is a powerful greenhouse gas, particularly over shorter timescales.

A large gas system built for AI infrastructure can remain operational for decades, potentially creating long-term dependence on fossil fuel even as electricity grids attempt to decarbonize.

Is Natural Gas Better Than Diesel?

For extended generation, modern gas turbines can offer advantages over diesel generators.

They may produce:

• Less particulate pollution
• Less sulfur pollution
• Lower carbon dioxide emissions per unit of electricity
• Lower nitrogen-oxide emissions when effective controls are used
• Fewer fuel-delivery challenges

But the comparison depends heavily on the technologies involved.

A poorly controlled gas engine can create substantial NOx emissions and methane leakage.

A modern diesel engine operated only briefly may produce less total annual pollution than a gas turbine operating continuously.

The relevant question is not simply:

“Is gas cleaner than diesel?”

It is:

• How many units are installed?
• How often will they operate?
• What controls will they use?
• What is their permitted operating limit?
• What is the upstream methane footprint?
• What cleaner alternatives were considered?
• Will the equipment delay retirement of fossil infrastructure?
• How close is it to populated areas?

Why the AI Boom Could Lock In Fossil Fuel Infrastructure

Data center developers are planning for rapid growth now.

Power equipment built in response may remain in service for decades.

A turbine installed to solve a temporary grid delay can become a permanent asset because:

• It has already been paid for.
• It increases operational flexibility.
• It can participate in electricity markets.
• It can provide backup during extreme weather.
• Retiring functional equipment creates financial losses.
• Future AI demand may justify more frequent use.

This creates the risk of carbon lock-in.

A supposedly transitional fossil-fuel solution can shape infrastructure decisions long after the original problem has passed.

Utilities may also build new pipelines, transmission systems, and gas plants to support data center demand.

Once constructed, those systems can influence energy policy and operating decisions for decades.

Data Centers Can Also Increase Grid Emissions Indirectly

A data center does not need its own gas turbines to increase fossil-fuel use.

When a large facility connects to the grid, utilities must generate enough additional electricity to serve it.

If new wind, solar, nuclear, storage, or other clean resources are not available at the same time and location, existing gas or coal plants may operate more frequently.

A company may purchase renewable-energy certificates or sign a power agreement with a wind farm, but its hourly consumption may still be met by fossil generation when renewable output is low.

This is the difference between:

• Annual renewable-energy accounting
• Actual hourly power-system operation

A company can match its annual electricity use with renewable purchases while still increasing fossil generation during many hours.

More rigorous clean-energy strategies attempt to match data center consumption with carbon-free electricity every hour and within the relevant grid region.

The Difference Between Carbon Dioxide and Local Air Pollution

Climate pollution and local air pollution overlap, but they are not identical.

Greenhouse Gases

These include:

• Carbon dioxide
• Methane
• Nitrous oxide

They accumulate in the atmosphere and contribute to global warming.

Their effects are global, regardless of exactly where they are emitted.

Conventional Air Pollutants

These include:

• Nitrogen oxides
• Particulate matter
• Carbon monoxide
• Sulfur dioxide
• Volatile organic compounds
• Hazardous air pollutants

Their effects are often most severe near the source or within the surrounding air basin.

A company could theoretically offset carbon emissions elsewhere while nearby residents continued breathing NOx and particulate pollution.

Climate accounting therefore cannot replace local air-quality protection.

Responsible AI infrastructure must address both.

Why NOx Is Especially Important in Texas

Texas contains several metropolitan areas that have struggled to meet federal ozone standards.

Ground-level ozone forms when nitrogen oxides react with volatile organic compounds in sunlight.

Hot, sunny conditions accelerate the process.

Texas’ climate therefore creates ideal conditions for ozone formation, particularly during summer.

Adding large numbers of combustion engines can worsen regional ozone problems even when no single facility appears dominant.

Ozone exposure can:

• Trigger asthma attacks
• Cause coughing and chest discomfort
• Reduce lung function
• Increase hospital visits
• Affect children’s lung development
• Make outdoor work and exercise more dangerous

Because ozone forms through regional atmospheric chemistry, regulators need to consider cumulative emissions rather than evaluate each generator in isolation.

Environmental Justice and Community Consent

Industrial facilities are not distributed evenly.

Polluting infrastructure is often concentrated near communities with:

• Lower incomes
• Higher proportions of racial or ethnic minorities
• Limited political influence
• Existing industrial pollution
• Lower property values
• Fewer health-care resources

When data center power equipment is approved without public notice, residents may learn about it only after construction begins.

That prevents meaningful participation in questions such as:

• Where turbines should be located
• How frequently they may operate
• Whether stronger emissions controls are required
• What monitoring should occur
• How noise will be managed
• Whether schools or homes are nearby
• What emergency procedures are planned
• Whether cumulative pollution has been evaluated

A technically legal permit does not automatically create community trust.

Trust requires transparency before construction, not only explanations after equipment appears.

Data Centers Create Noise as Well as Air Pollution

Large generator and turbine installations can create substantial noise.

Sources include:

• Turbine operation
• Generator testing
• Cooling fans
• Air-handling systems
• Transformers
• Construction
• Truck traffic

Low-frequency noise can travel long distances and may be particularly disruptive at night.

Potential impacts include:

• Sleep disturbance
• Stress
• Difficulty concentrating
• Reduced quality of life
• Property-use conflicts

Noise is not interchangeable with air pollution, but communities experience infrastructure as a combined burden.

A project that meets one regulatory limit may still create significant cumulative effects through noise, light, traffic, water demand, and emissions.

What About Water Consumption?

Data centers generate enormous amounts of heat.

Many facilities use evaporative cooling systems that consume water continuously as heat is released into the atmosphere.

This creates concern in drought-prone regions, including parts of Texas.

Crusoe says the Abilene Stargate campus uses a closed-loop, non-evaporative liquid-cooling system. Water is recirculated and heat is rejected through air-cooled chillers rather than traditional evaporative cooling towers. The company says this design greatly reduces operational water consumption for heat rejection.

That is a significant environmental feature.

However, water analysis should still include:

• Initial system filling
• Domestic water use
• Construction
• chip manufacturing
• Electricity generation
• Upstream fuel production
• Landscaping
• Emergency requirements

A facility can reduce direct cooling-water use while retaining a substantial indirect water footprint through the electricity and hardware it consumes.

AI Has Genuine Social Benefits

The environmental debate should not be reduced to the claim that AI provides no public value.

AI systems may contribute to:

• Medical imaging
• Drug discovery
• Climate modeling
• Scientific research
• Accessibility tools
• Language translation
• Education
• Industrial efficiency
• Fraud detection
• Disaster forecasting
• Software development
• Productivity

The question is not whether society should reject every AI application.

It is whether the physical expansion required to support AI should proceed without full accounting for climate, health, water, and community effects.

A technology that may improve health care should not quietly worsen respiratory health in the communities powering it.

AI’s potential benefits do not eliminate the obligation to build it responsibly.

They strengthen that obligation.

The Industry’s Efficiency Paradox

AI hardware is becoming more efficient.

New chips can perform more calculations per unit of electricity than previous generations.

Cooling and data center designs are also improving.

Yet total energy use may continue rising because demand grows faster than efficiency improves.

This is known as a rebound effect.

When computing becomes cheaper and more efficient:

• Companies train larger models.
• More users adopt AI tools.
• More products add AI features.
• Models process longer inputs.
• Video and image generation expand.
• Automated agents perform more tasks.
• Companies build additional data centers.

Efficiency reduces the energy needed for each calculation.

It does not guarantee a reduction in total calculations.

Can AI Data Centers Run on Renewable Energy?

Yes, but doing so reliably at enormous scale requires more than purchasing solar panels or annual renewable certificates.

Wind and solar output varies with weather and time.

Data centers operate continuously.

A low-carbon system may require a combination of:

• Wind
• Solar
• Battery storage
• Long-duration energy storage
• Geothermal power
• Nuclear power
• Hydroelectricity
• Strong transmission connections
• Demand flexibility
• Clean firm generation
• Regional power sharing

Texas has exceptional wind and solar resources.

Its major challenge is aligning those resources with the location and hour-by-hour demand of AI campuses.

A data center could improve the situation by adjusting flexible workloads based on grid conditions.

Not every AI task must run at the exact moment it is submitted.

Some model training, data processing, and maintenance workloads can potentially shift toward hours with abundant renewable electricity.

Batteries Can Replace Some Generator Functions

Battery systems already provide short-duration backup at data centers.

They respond almost instantly when grid power fails, allowing servers to continue operating while generators start.

Larger batteries could reduce generator use by:

• Covering brief outages
• Managing voltage disturbances
• Supporting grid frequency
• Shifting electricity consumption
• Reducing peak demand
• Avoiding generator testing during certain events
• Providing power until workloads shut down safely

However, batteries have limitations.

A severe outage may last hours or days.

Providing enough battery capacity for a massive data center during a long grid emergency would be expensive and require large amounts of material.

The best near-term strategy may combine batteries with cleaner long-duration backup rather than expecting one technology to solve every reliability problem.

Alternatives to Diesel Backup

Possible replacements or supplements include:

Battery Energy Storage

Best for immediate response and shorter outages.

Fuel Cells

Can provide onsite electricity with lower local pollution, depending on the fuel source. Hydrogen or biogas may reduce carbon emissions if produced sustainably.

Renewable Microgrids

Solar, wind, and batteries can provide part of the campus’s energy independently.

Geothermal Power

Offers continuous low-carbon electricity in suitable locations.

Nuclear Energy

Can provide firm carbon-free power but involves cost, regulatory, safety, construction-time, and waste considerations.

Grid-Interactive Workloads

Some computing can be reduced or shifted during supply shortages.

Multiple Grid Connections

Strong transmission and geographic diversity can reduce reliance on onsite combustion.

Cleaner Emergency Fuels

Renewable diesel or lower-carbon fuels may reduce some lifecycle emissions, although they can still produce local air pollution.

There is no universal solution.

A responsible design will usually combine several technologies.

What Stronger Permitting Could Look Like

Regulators do not need to prohibit data centers to improve oversight.

They could require:

• Review of the entire campus as one project
• Cumulative analysis of all turbines and generators
• Public notice before construction
• Health-risk assessments
• Air-dispersion modeling
• Consideration of nearby communities
• Clear limits on non-emergency generator use
• Continuous emissions monitoring
• Public operating-hour records
• Best available pollution controls
• Disclosure of expansion plans
• Evaluation of cleaner alternatives
• Reporting of actual greenhouse-gas emissions
• Noise studies
• Water-use transparency

Permitting rules designed for one or two emergency engines may not be adequate for campuses containing hundreds.

Regulation must reflect the real scale of modern computing infrastructure.

Companies Should Disclose More Than Annual Carbon Totals

Corporate environmental reports often provide broad annual numbers.

Communities need more specific information.

Data center operators should disclose:

• Electricity use by location
• Hourly carbon intensity
• Actual generator operating hours
• Turbine fuel consumption
• Methane assumptions
• NOx emissions
• Particulate emissions
• Water use
• Backup testing schedules
• Renewable-energy contracts
• Expected future expansion
• Construction emissions
• Equipment manufacturing impacts

Transparent reporting would allow researchers, regulators, investors, and residents to distinguish facilities making genuine progress from those relying mainly on marketing claims.

Should AI Companies Be Responsible for Supplier Emissions?

AI companies may lease computing capacity from infrastructure developers rather than owning every building, generator, and turbine directly.

That contractual separation should not erase environmental responsibility.

If an AI company’s demand causes construction of new fossil-fuel power, those emissions are part of the system enabling its products.

Responsibility may be shared among:

• AI developers
• Cloud providers
• Data center owners
• Utilities
• Turbine operators
• Hardware manufacturers
• Investors
• State and local governments

Effective accountability should follow the economic and operational relationships rather than only legal ownership.

The Need to Distinguish OpenAI From the Site Developer

The Abilene campus is widely described as OpenAI’s Stargate facility because it supports the Stargate infrastructure programme.

However, construction, permitting, and operation involve several companies and project entities.

Crusoe is the developer most directly associated with building the campus and has publicly explained its turbine and cooling strategies. Oracle and other partners are involved in computing and infrastructure deployment, while OpenAI is a principal customer and platform participant.

Environmental reporting should describe this structure accurately.

It would be too simple to imply that OpenAI alone applied for every permit or operates every generator directly.

It would also be too simple to argue that an AI company bears no responsibility because a contractor owns the physical equipment.

Is the AI Boom Necessarily Bad for the Climate?

No.

AI could help reduce emissions through:

• Improved grid forecasting
• Better battery design
• Efficient logistics
• Materials discovery
• Building-energy optimization
• Methane-leak detection
• Climate modeling
• Industrial-process control
• Renewable generation forecasting

But environmental benefits are not automatic.

An AI system used to optimize renewable power could be environmentally valuable.

The same infrastructure could also be used for advertising, surveillance, fossil-fuel exploration, disposable content, or computationally wasteful applications.

The climate value of AI depends on:

1. What the system is used for
2. How much energy it consumes
3. Where that electricity comes from
4. What infrastructure its growth causes
5. Whether its benefits exceed its environmental costs

What Consumers Can Do

Individual users have limited influence compared with companies and regulators, but they are not powerless.

People can:

• Support transparency requirements
• Ask technology companies for energy and water disclosures
• Avoid unnecessary repeated generation
• Use smaller models when sufficient
• Prefer services with credible clean-energy commitments
• Support local communities seeking public participation
• Distinguish useful AI applications from novelty consumption
• Vote for policies supporting clean electricity and modern grids

Consumers should not be made to feel individually responsible for solving an industrial infrastructure problem.

The largest decisions involve corporate investment, power-system planning, and regulation.

Still, public expectations can influence company behavior.

What Responsible AI Infrastructure Would Look Like

A genuinely responsible AI campus would not simply claim carbon neutrality through distant offsets.

It would aim to:

• Use carbon-free electricity matched hourly
• Avoid building unnecessary fossil-fuel generation
• Use high-efficiency hardware
• Reuse waste heat where practical
• Reduce direct and indirect water consumption
• Install advanced emissions controls
• Minimize diesel use
• Publish actual operating data
• Consult residents before construction
• Protect nearby communities
• Shift flexible workloads during grid stress
• Contribute financially to transmission and clean generation
• Plan for eventual equipment decommissioning
• Measure the full lifecycle footprint of chips and buildings

Responsible growth does not mean zero impact.

Every physical system has an environmental footprint.

It means acknowledging that footprint, reducing it transparently, and ensuring communities are not asked to carry hidden costs for private technological expansion.

Final Thoughts

The AI revolution is not happening only in software.

It is being built through steel, concrete, transmission lines, cooling systems, gas pipelines, turbines, and diesel engines.

In Texas, at least 38 data centers have received minor permits for onsite power sources since 2024. More than 2,100 backup diesel generators have been authorized, with permits allowing nearly 2,500 tons of nitrogen-oxide emissions each year.

At the Stargate campus near Abilene, 10 gas turbines and 62 diesel generators have already been installed. Their existing permits authorize more than 1.6 million tons of greenhouse-gas emissions and more than 1,000 tons of combined conventional air pollutants annually, although actual emissions may be much lower if the equipment operates only occasionally.

Crusoe says the turbines are intended as backup systems, use advanced nitrogen-oxide controls, and are cleaner than conventional diesel-based alternatives. The company has also adopted closed-loop, non-evaporative cooling to reduce direct water consumption.

Those measures deserve recognition.

They do not eliminate the larger questions.

Why were such substantial power systems initially approved through minor permitting pathways?

Were the original and planned expansions evaluated as one connected project?

Will backup turbines remain rarely used, or become long-term sources of power?

How will actual emissions be disclosed?

Will communities receive meaningful opportunities to influence projects before construction creates irreversible momentum?

Most importantly, will AI infrastructure accelerate the clean-energy transition—or create a new generation of fossil-fuel plants that remain in service for decades?

Artificial intelligence may contribute to medicine, science, accessibility, and productivity.

Its value should not depend on hiding its physical costs.

The industry’s future will not be judged only by how intelligent its models become.

It will also be judged by whether the systems powering those models protect the climate, respect surrounding communities, and avoid turning digital progress into another source of preventable pollution.

Frequently Asked Questions

Why do AI data centers use so much electricity?

AI models require large numbers of specialized processors for training and operation. Electricity also powers storage, networking, cooling, security, and backup systems.

How many Texas data centers received permits for onsite power?

A Floodlight investigation identified at least 38 Texas data centers that received minor permits for onsite power equipment beginning in 2024.

How many diesel generators were approved?

The identified permits authorized more than 2,100 diesel backup generators across Texas.

How much NOx could those generators emit?

Together, they are permitted to emit nearly 2,500 tons of nitrogen oxides annually. Actual emissions will depend on operating time and equipment performance.

Is NOx the same as nitrous oxide?

No. NOx refers mainly to nitric oxide and nitrogen dioxide. Nitrous oxide is a different greenhouse gas with the formula N₂O.

Why is NOx harmful?

NOx contributes to smog, ground-level ozone, and fine-particle pollution. It can worsen asthma, irritate airways, reduce lung function, and contribute to cardiovascular and respiratory disease.

Do backup generators operate only during blackouts?

Not always. They also run for testing and maintenance and may operate during grid emergencies or other approved conditions.

What is the Stargate campus?

Stargate is a large AI infrastructure platform involving OpenAI, Oracle, SoftBank, and infrastructure partners. Its flagship data center campus is located near Abilene, Texas.

How many turbines are installed at Stargate in Abilene?

The campus has installed 10 natural-gas turbines.

How many diesel generators are installed there?

It has 62 diesel backup generators under the initial permits.

How much pollution can the current Stargate permits allow?

The permits authorize more than 1.6 million tons of greenhouse gases and more than 1,000 tons of combined harmful air pollutants per year from the installed turbines and generators.

Is Stargate actually emitting 1.6 million tons every year?

That has not been established. The figure is a permitted maximum rather than a confirmed measurement of actual annual emissions.

Are the gas turbines intended to run continuously?

Crusoe says the turbines will ultimately serve as backup power for the data halls rather than continuous primary generation.

Why are they permitted for more operation than expected?

Permits sometimes authorize broader operating limits than companies initially expect to use. Critics argue that these limits still matter because they permit future operation without additional approval.

Does Stargate use pollution-control equipment?

Crusoe says the turbines use selective catalytic reduction systems that sharply reduce nitrogen-oxide emissions.

What expansion is planned?

Developers have sought approval for 41 additional gas turbines and 18 more generators.

How large would the expanded power system become?

The full installation could contain 51 turbines and 80 diesel generators.

Why are minor permits controversial?

Minor permitting systems are intended to streamline common, predictable sources. Critics argue that approving dozens of turbines and generators through several smaller authorizations avoids the level of review appropriate for a major power facility.

Did Texas regulators break the law?

That has not been established. TCEQ says its permits comply with applicable state and federal rules. Former regulators and environmental advocates have questioned whether connected equipment should have been reviewed together.

What is project aggregation?

Aggregation is the regulatory practice of treating connected equipment or development phases as one pollution source rather than dividing them into several smaller permit applications.

What is a sham permit?

The term is used in air-permitting policy for an authorization that appears minor only because the applicant divided or understated a project that was always intended to become a major source. No court has definitively ruled that Stargate’s permits constitute sham permits.

Why are data centers building their own power plants?

Grid connections and new utility infrastructure can take years. Onsite generation allows developers to begin operating sooner and maintain reliable electricity.

Is natural gas cleaner than diesel?

Modern gas turbines usually produce less particulate and sulfur pollution than diesel engines and may emit less carbon dioxide per unit of electricity. They still produce greenhouse gases and local air pollution.

Does natural gas contribute to climate change?

Yes. Burning natural gas releases carbon dioxide, and methane can leak throughout production and delivery.

Could renewable energy power AI data centers?

Yes, but continuous operation requires a combination of renewable generation, storage, transmission, flexible demand, and other firm carbon-free sources.

Can batteries replace diesel generators?

Batteries can cover short outages and grid disturbances. Very long outages may require larger storage systems or another backup source.

Could hydrogen fuel cells be used?

Potentially. Their environmental performance depends heavily on how the hydrogen is produced and whether local pollution is controlled.

What is carbon-free hourly matching?

It means matching a facility’s electricity consumption with carbon-free generation during every hour rather than purchasing enough renewable energy to offset annual consumption in total.

Why is annual renewable matching insufficient?

A data center may consume fossil-generated electricity at night or during low-wind periods even if its annual renewable purchases equal its total yearly use.

Do data centers use large amounts of water?

Many do, particularly when using evaporative cooling. Water use varies significantly by cooling technology and climate.

How does the Abilene Stargate site reduce water use?

Crusoe says it uses closed-loop, non-evaporative liquid cooling and air-cooled chillers, reducing water consumed during heat rejection.

Does that mean the site has no water footprint?

No. Construction, electricity generation, equipment manufacturing, sanitation, and other activities still involve direct or indirect water use.

Are AI data centers harmful to nearby residents?

Potential impacts depend on location, equipment, controls, operating hours, and existing air quality. Concerns can include NOx, particulate pollution, noise, traffic, water use, and cumulative industrial exposure.

Who is most vulnerable to NOx pollution?

Children, older adults, people with asthma or cardiovascular disease, outdoor workers, and residents of already polluted communities may face greater risks.

What is environmental justice?

Environmental justice concerns fair treatment and meaningful participation for all communities in environmental decision-making, particularly populations that have historically carried disproportionate pollution burdens.

Do data centers create jobs?

Yes. They can generate significant construction employment and permanent technical, security, maintenance, and operations jobs, although permanent employment may be smaller than the construction workforce.

Do data centers benefit local tax systems?

They can increase tax revenue, but incentives and abatements may reduce the amount collected. They may also require costly electricity, water, road, and emergency infrastructure.

Is AI itself environmentally harmful?

AI is not environmentally uniform. Its impact depends on model size, usage, hardware efficiency, electricity source, cooling method, and the social value of the application.

Can AI help address climate change?

AI can support grid management, materials research, weather forecasting, methane detection, and energy efficiency. Those benefits must be compared with the emissions created by its infrastructure.

Will more efficient chips solve the energy problem?

Efficiency helps, but total demand may still rise if companies build larger models and expand usage faster than energy savings improve.

What should AI companies disclose?

They should report location-specific electricity use, hourly carbon intensity, water consumption, generator operating hours, actual emissions, expansion plans, and clean-energy sourcing.

What should regulators require?

Potential measures include full-project review, cumulative pollution analysis, public notice, strict operating limits, emissions monitoring, cleaner backup technology, and transparent reporting.

What can communities do?

Residents can monitor permit applications, request public hearings, submit comments, ask for air monitoring, contact local officials, and seek technical or legal support from environmental organizations.

What can individual AI users do?

Users can avoid unnecessary repeated generation, choose efficient services, support clean-energy policy, and demand transparency from technology providers.

Should people stop using AI?

The evidence does not require a universal rejection of AI. It supports using AI more deliberately while demanding cleaner infrastructure and stronger oversight.

What is the central environmental issue behind the AI boom?

The key issue is whether rapidly expanding computational demand will accelerate construction of clean electricity and efficient infrastructure—or lock communities into decades of new fossil-fuel pollution.