AI Needs Power Now, and Natural Gas Is Winning the Race to Keep the Servers Running

Technology companies still promise a future powered by renewables and nuclear energy. The immediate competition is being won by natural gas because AI data centers need dependable electricity faster than transmission lines, reactors and utility-scale clean-energy projects can be completed.

By Karla Alvarado Follow

Published at 5:48 p.m. Eastern Time

The artificial intelligence boom is rapidly becoming a natural gas boom.

Across the United States, data-center developers, electric utilities and technology companies are pursuing new gas-fired power plants to satisfy an electricity appetite that is growing faster than the country’s power infrastructure can comfortably manage.

Solar energy is inexpensive and can be installed quickly. Wind power remains a major source of new electricity. Battery storage is expanding, and technology companies have announced high-profile agreements involving nuclear reactors.

Yet natural gas is gaining the clearest near-term advantage because it can provide large amounts of electricity around the clock, whether the sun is shining or the wind is blowing.

Global Energy Monitor reported in August that proposed and developing American gas-fired capacity connected directly to data centers had nearly doubled in only six months. The organization identified 189 gigawatts of announced, pre-construction or actively constructed gas capacity intended primarily to serve data-center demand.

That is not a forecast from decades in the future. It is a measure of how the AI infrastructure race is already changing American energy planning.

The broader pipeline of proposed U.S. gas generation reached 378 gigawatts, a 50 percent increase since the beginning of 2026. According to Global Energy Monitor, the United States now accounts for roughly one-third of all gas-fired generating capacity being developed worldwide.

Not every project will be built. Many are preliminary proposals without financing, completed permits or confirmed equipment orders. But the direction is unmistakable.

AI companies want electricity immediately, and the existing grid cannot always deliver it. Natural gas is becoming the industry’s preferred shortcut.

Why AI consumes so much electricity

Traditional data centers store files, operate websites, process payments and deliver streaming services. AI-focused facilities perform those tasks while also running thousands of specialized processors used to train and operate increasingly complex models.

Training a large AI model involves repeatedly moving vast quantities of data through graphics processors or other accelerators. Once the model is released, users continue to consume electricity every time they submit a prompt, generate an image, translate a document or request a video.

This second category, called inference, is becoming an increasingly important source of power demand as AI tools move into search engines, smartphones, workplaces, vehicles and customer-service systems.

AI equipment is also unusually concentrated. A large campus can require hundreds of megawatts, while the most ambitious projects are measured in gigawatts. One gigawatt is comparable to the output of a large conventional power plant and can supply hundreds of thousands of homes, depending on consumption and operating conditions.

The International Energy Agency reported that electricity use by data centers increased 17 percent in 2025. Electricity demand from AI-focused facilities grew even faster, according to the agency’s 2026 data-center update.

The challenge is not simply how much electricity these facilities use over an entire year. It is that operators expect power to remain available every second.

A training cluster worth billions of dollars cannot easily sit idle during a cloudy afternoon, a windless evening or a period of grid congestion. Even a short interruption can disrupt computing operations and damage the economics of an expensive facility.

That need for constant, controllable electricity gives natural gas its advantage.

Gas can run when technology companies need it

Gas-fired power plants are dispatchable. Operators can instruct them to produce electricity when demand rises, and some designs can increase or decrease output relatively quickly.

Renewable energy is different. Solar panels generate electricity according to sunlight. Wind turbines depend on weather conditions. Batteries can store excess electricity and release it later, but the duration and scale of storage needed to support a massive data center through extended periods of low renewable generation can be expensive.

A balanced system can combine solar, wind, batteries, transmission and demand management. In many locations, that combination can supply a substantial portion of a data center’s needs.

The difficulty is timing.

Technology companies are racing to install computing equipment within one or two years. Major transmission projects can take considerably longer because they require route approval, land acquisition, environmental reviews and construction across multiple jurisdictions.

New nuclear reactors promise firm, low-carbon power, but conventional plants have long development timelines and high capital costs. Small modular reactors could eventually serve data centers, although many designs still face regulatory, manufacturing and commercial challenges.

Existing nuclear plants are valuable, and companies including Microsoft, Amazon and Google have signed agreements intended to support nuclear generation. Those contracts may shape the energy system during the 2030s. They do not solve every power shortage in 2026.

Gas plants are familiar to utilities, regulators and construction companies. The United States also possesses extensive gas production, pipelines and power-generation expertise.

When a developer asks what can be built quickly and operate continuously, gas often becomes the practical answer.

Texas is at the center of the buildout

No state illustrates the trend more clearly than Texas.

Global Energy Monitor identified approximately 122 gigawatts of gas-fired capacity in development in the state, an increase of more than 41 gigawatts in six months. Approximately 77 gigawatts of the Texas pipeline is intended to power data centers directly.

Texas offers several attractions for AI infrastructure developers.

The state has abundant land, a large natural gas industry, relatively accommodating permitting rules and an independent electricity market. It is also home to rapidly expanding solar and battery-storage industries, allowing developers to propose mixed systems that combine gas generation with renewable electricity.

However, Texas also demonstrates the risks of moving faster than the grid.

The state’s population and economy are growing, while industrial facilities, cryptocurrency mines, electrification and data centers compete for new power. Extreme heat can push air-conditioning demand to dangerous levels, and winter storms can disrupt gas production and power plants.

Adding private generation beside data centers may reduce the burden on the public grid during normal operations. It can also create new competition for gas supplies, water, equipment and pipeline capacity.

The question is not simply whether Texas can build enough power. It is who pays for the infrastructure and who receives priority when the system is under stress.

Louisiana is becoming another gas-powered AI hub

Meta’s enormous Hyperion data-center project in Louisiana has become one of the clearest examples of gas and AI developing together.

The project is expected to grow into a multigigawatt computing campus. Utility Entergy Louisiana has pursued new gas generation and transmission infrastructure to serve the development, with Meta expected to cover major project-related costs.

Supporters say the arrangement can attract investment and construction activity without forcing existing customers to finance the full expansion.

Critics remain concerned about long-term utility costs, air pollution and the possibility that infrastructure built for a technology company could eventually leave ratepayers exposed if projected demand fails to materialize.

This concern is spreading nationally.

The U.S. House of Representatives was expected to consider the bipartisan Ratepayer Protection Act on September 16. The legislation would direct state utility regulators to examine whether large electricity users, including data centers, should bear the additional infrastructure costs required to serve them.

Former Federal Energy Regulatory Commission member Allison Clements called the proposal a modest but important signal that Congress is focused on protecting consumers, according to Reuters.

The legislation reflects an emerging political reality. AI electricity demand is no longer only a technology or energy issue. It is becoming a household affordability issue.

The turbine shortage could slow the gas victory

Natural gas may be winning the race, but it faces a serious supply-chain problem.

A large combined-cycle gas plant requires complex turbines that only a small number of global manufacturers can produce. Orders have surged, creating multiyear delivery schedules and higher equipment prices.

GE Vernova, Siemens Energy and Mitsubishi Heavy Industries are among the major suppliers benefiting from demand. The same backlog that strengthens their businesses can delay the AI projects counting on their equipment.

Developers are responding by considering smaller aeroderivative turbines and reciprocating gas engines. Those systems can often be manufactured and installed more quickly than a large combined-cycle plant.

Global Energy Monitor found that gas-engine capacity in development more than doubled during the first half of 2026, rising from 31 gigawatts to 67 gigawatts. Engine capacity associated specifically with data centers more than tripled to 45 gigawatts.

The speed comes with a tradeoff.

Smaller turbines and engines can be less efficient than modern combined-cycle facilities and may produce more emissions per unit of electricity. Some are better suited for occasional or emergency operation than for continuous service.

A data center that installs rapidly deployable gas equipment to escape a grid-connection delay may therefore receive power sooner while creating a more polluting electricity source.

The climate contradiction

Many of the same technology companies pursuing gas-fired power have promised to reduce their carbon emissions.

Google has sought around-the-clock carbon-free energy. Microsoft has pledged to become carbon negative. Amazon and Meta have signed enormous renewable-energy contracts.

The AI boom is making those commitments harder to achieve.

Burning natural gas produces less carbon dioxide than burning coal for the same amount of electricity, but it still releases significant greenhouse gases. Gas production and transportation can also leak methane, which has a powerful warming effect over shorter time periods.

A large new gas plant may operate for 30 years or longer. Infrastructure installed to solve an immediate data-center shortage could therefore continue producing emissions well after clean alternatives become more available.

Some developers have proposed pairing gas plants with carbon-capture equipment. The idea is to collect carbon dioxide before it enters the atmosphere and store it underground.

Carbon capture may reduce emissions, but large power-sector installations remain expensive and operationally complicated. Capture systems also consume energy, require pipelines and cannot address all upstream methane leakage.

Companies should not receive the environmental benefit of a carbon-capture promise until the equipment is financed, built and consistently operating at the claimed capture rate.

Gas is winning the short race, not necessarily the long one

The current surge does not mean natural gas will dominate every part of AI’s energy future.

Renewables remain attractive because their operating costs are low and they do not require fuel. Batteries are becoming less expensive and can help data centers manage short periods of grid stress. Advanced geothermal energy may provide constant clean electricity in suitable locations.

Nuclear energy has gained renewed support because it offers low-carbon power around the clock. Technology companies are financing reactor restarts and supporting advanced reactor developers. If those projects succeed, nuclear generation could capture more of the AI market in the next decade.

Data centers may also become more flexible.

Google, Nvidia and Emerald AI announced the AI Energy Management Alliance on September 16, bringing together technology companies, utilities and power producers. The coalition wants regulators to reward data centers that can reduce their electricity consumption during periods of grid stress.

Google says it has committed 1 gigawatt of flexible demand through utility agreements, according to Axios.

Some computing tasks can be delayed or shifted to another location without affecting users. Batteries and on-site generation can also reduce grid demand temporarily.

If flexibility becomes measurable and enforceable, it could reduce the number of new power plants required. It could also allow data centers to connect to the grid more quickly without receiving unlimited electricity during the system’s most difficult hours.

The challenge will be proving that companies can reduce demand when utilities need them to do so, not merely when it is financially convenient.

Communities are beginning to resist

The political backlash against data centers is growing because residents increasingly associate the facilities with higher utility bills, water consumption, construction noise and air pollution.

An AP-NORC and University of Chicago Energy Policy Institute poll cited by Axios found that 84 percent of Americans were concerned about the effect of data centers on local electricity prices.

Roughly four in five Democrats and three in four Republicans supported requiring data-center developers to pay for the grid improvements needed to serve them.

New York Governor Kathy Hochul has proposed requiring developers to invest at least $1 million in surrounding communities for every megawatt of utility demand associated with a new data center. The recommendation followed a state moratorium on large new facilities.

These policies show that access to electricity is becoming a social license issue.

A project may possess land, investors and computer equipment but still fail if the surrounding community believes it will raise household rates or worsen pollution.

Natural gas generation can make the conflict more intense because it converts an otherwise quiet digital facility into an industrial energy site with pipelines, engines, turbines and smokestacks.

Many proposed plants will never be built

The enormous gas-development figures require context.

More than three-quarters of the global gas-project pipeline remains in an early stage, according to Global Energy Monitor. Many projects lack a confirmed turbine supplier or scheduled starting year.

During the first half of 2026, approximately 45 gigawatts of proposed capacity had its expected start date delayed.

Projects can disappear because of high interest rates, equipment shortages, local opposition, gas-pipeline limitations, environmental litigation or changing demand forecasts.

AI developers may also overestimate future computing demand. If models become more energy-efficient or investment slows, some planned data centers may never need the electricity currently being requested.

Jenny Martos, a project manager for Global Energy Monitor’s oil and gas plant tracker, warned that distinguishing realistic projects from speculative announcements has become extremely difficult.

The safest conclusion is not that the United States will build all 378 gigawatts in the development pipeline. It is that gas has become the default proposal whenever the AI industry encounters an electricity constraint.

What winning really means

Natural gas is winning AI’s power race because it currently offers the combination the industry values most: scale, reliability, familiarity and relative speed.

It is not winning because it is the cheapest choice in every location. It is not winning because it has no environmental consequences. It is winning because the AI sector has created a deadline that other energy technologies and the existing grid often cannot meet.

That victory could prove temporary.

If transmission construction accelerates, batteries become cheaper, nuclear projects reach operation and data centers learn to shift their computing demand, gas may lose part of its advantage.

If those alternatives remain delayed, the United States could lock in a generation of fossil-fuel infrastructure built not for homes or factories, but for machines training and operating artificial intelligence.

The central policy question is no longer whether AI will consume enormous amounts of electricity. It is whether the country can supply that power without making households pay more, weakening grid reliability or abandoning climate commitments.

For now, natural gas is answering the industry’s call faster than its competitors.

The bill for that speed will arrive later.

Reporting and interview disclosure

This article was independently assembled from government information, energy-market data, corporate announcements and attributed interviews published by Reuters, Axios, the International Energy Agency and Global Energy Monitor.

Consumerlite News