Fracked Gas Powers AI Push in Gas-Addicted Texas

AI may seem weightless, but its power source isn’t. Underneath your chatbot and video generator is a network of gas wells, combustion turbines and scraped scrubland — especially in Texas — where tech companies are building their own small power grid linked directly to fracked natural gas.

The approach is straightforward and audacious: Bring your huge data centers to where the fuel is plentiful, then line up special-purpose generation so you don’t have to compete in line with regular folks on a creaky grid. The product is an AI boom physically rooted in the Permian Basin and reflected by bulldozers, flares and new neighborhood disputes.

The Data Center Rush to Oil Country Accelerates

According to the Wall Street Journal, Poolside, an AI coding startup, plans to build “Horizon”, a data center compound exceeding 500 acres in size that is located just west of Texas — almost two-thirds the size of Central Park. The facility will pull natural gas from the Permian Basin to produce its own electricity, a direct feed into racks of Nvidia chips supplied by CoreWeave containing over 40,000 GPUs. The project is aiming for about 2 gigawatts of computing power, which is more than the electric capacity of the Hoover Dam — but it will be produced by burning gas.

Hydraulic fracturing is the norm, not the exception, in the Permian. A co-location of compute and fuel can provide lower delivered-energy costs and fewer transmission delays. It also locks in AI’s growth to a fossil supply chain that communities and climate scientists have been denouncing for a decade.

Gas-Fired AI Moves In Next Door to Texas Towns

Its other campuses have continued to grow since then, but plans for its flagship Stargate campus in Abilene epitomize the repositioning. The site would need about 900 megawatts spread across eight buildings and contain a gas-fired plant with turbines similar to those on warships, the Associated Press reported. Company executives have not minced words: “We’re burning gas to run this data center.” The on-site plant, operators say, is backup and most of the electricity comes from the local grid which, in West Texas, is a mix of natural gas with substantial wind and growing solar.

Neighbors say the change feels sudden, and personal. One resident told the AP that bulldozers had obliterated her mesquite shrubland, and that jackhammers and floodlights have supplanted the silence she sought by arriving there. Water anxiety is heaped on top: with reservoirs low and outdoor watering banned, Oracle promises that each building’s closed-loop cooling will take only about 12,000 gallons a year to sip from the well after gulping a million as an initial fill. But Shaolei Ren, a professor at the University of California, Riverside, points out that closed-loop systems consume more electricity and thus have more “indirect” water use at power plants.

Louisiana and Beyond Follow the Same Playbook

Meta is planning to build approximately $10 billion in data center capacity in Louisiana’s Richland Parish that will need two gigawatts of power just for computation, people familiar with the site selection say. The utility, Entergy, is lining up three large gas plants totaling 2.3 gigawatts to serve the load, fed by fracked gas from the Haynesville Shale. Local residents have also raised concerns about ongoing construction and the Dumb & Dumber-ic cost of new gas infrastructure.

Meta also announced a $1.5 billion facility in El Paso, approximately one gigawatt of capacity paired with 100% clean and renewable energy — serving as another demonstration that alternatives exist, even in the Southwest. The xAI Memphis buildout also hooks up with regional gas pipelines including Texas Gas Transmission and Trunkline Gas Company, which are tied to fracked (BP) gas supplies.

The National Stakes And The Policy Tailwind

Industry executives say they need to build gigawatts of new power at breakneck speed to match the building frenzy in China of generation and nuclear plants. That framing has landed. A new executive order fast-tracks gas-powered AI campuses by cutting through environmental reviews, offering economic benefits and opening up federal lands to gas, coal and nuclear — pointedly excluding renewables.

The result is to render gas the default bridge for AI’s load growth, even as cleaner alternatives mature. For Texas in particular, where ERCOT has reported multiple all-time demand peaks above 85 gigawatts, building must-run gas plants for data centers could exacerbate grid strain and local emissions.

The Grid Is More Flexible Than We Use It

An analysis by Duke University found that U.S. utilities, on average, draw only about 53% of their available capacity. If data centers dialed demand back by 50% for just a couple of peak hours a year, utilities could handle an additional 76 gigawatts of load — which is what data centers are projected to require within the decade, MIT Technology Review has written. That implies that many of those new gas plants are not truly needed, if the industry instead adopts flexible demand.

By coupling flexible compute with targeted transmission upgrades, storage and renewables, AI’s demand could be met more quickly than the new fossil infrastructure; all at a lower long-term cost and overall water usage.

Local Emissions And Seismic Risk Are Real

Those on-site gas turbines generate carbon dioxide and nitrogen oxides; depending on controls, NOx can be limited to the single-digit ppm via selective catalytic reduction, but the pollution is always local. In Texas, air permits under the supervision of the state environmental agency largely dictate whether these plants are classified as minor or major sources — determinations that help set the table for local air quality.

The footprint of fracking extends beyond the fence line. A typical Permian horizontal well can consume 10 to 20 million gallons of water and produce large amounts of briny “produced” water. Other Division researchers and the Texas Seismological Network have connected induced earthquakes near Midland and Odessa to wastewater disposal. By securing demand, AI campuses reinforce the economics of more drilling and more disposal.

Who Pays When the Music Stops on AI Power Buildout

Analysts caution about a circular dependency: AI developers want cloud providers who require chipmakers who need foundries who need power — and then some. The Financial Times has identified risk that if AI demand disappoints, utilities and customers face stranded gas plants and higher bills. There, Meta has agreed in a fixed-term contract to underwrite the costs of Entergy; CoreWeave and Poolside are said to have arranged a 15-year agreement. What may follow once those contracts expire is an open — and costly — question.

What a Better AI Buildout Might Prioritize

Data centers can also commit to measures that reduce their grid and environmental impact:

• Commit to hour-by-hour curtailment
• Co-locate with high-capacity wind and solar plus storage
• Purchase firmed clean power
• Invest in transmission rather than private gas plants
• Employ hybrid or dry cooling
• Treat brackish and reclaimed water
• Make publicly available verified environmental impact analysis with genuine community consent

AI doesn’t have to mean fracked gas. But today in Texas, the most prevalent model is bulldozers and turbines and pipelines. If the industry doesn’t change course, the price — measured in carbon, water and dollars — will be paid by the people who live closest to these massive server farms and help run everyone else’s AI.