Amazon’s Massive New Texas Data Center Power Plant Could Rank Among the U.S.’s Largest Greenhouse Gas Polluters
Tech giant Amazon has purchased an under-construction massive off-grid power plant in Texas that will supply energy to one of its data centers, a facility that could become one of the single largest sources of greenhouse gas emissions in the United States. Its enormous size is not the only reason it will carry such a heavy climate cost, however: the plant will run on far less efficient gas turbines than standard utility power plants, a design choice that is quickly becoming the norm for new data center projects across the country. This trend could have far-reaching consequences for the environmental footprint of artificial intelligence and the future of the entire U.S. power grid.
The Texas plant is legally permitted to release up to 33 million tons of greenhouse gases every year. The facility will rely exclusively on simple-cycle gas turbines, which generate power by mixing fuel and air in a single combustion chamber, releasing most of the produced heat as waste. Most grid-connected natural gas plants use combined-cycle units, which add a secondary steam turbine to capture that waste heat and convert it into extra electricity. This design means combined-cycle plants produce far fewer emissions for every unit of energy they generate.
While permitted emissions caps for natural gas plants are almost always set higher than a facility’s actual annual output, even if Amazon’s plant runs at half its permitted limit each year, it would still produce more greenhouse gas pollution than 78 average-sized U.S. natural gas plants, according to U.S. Environmental Protection Agency analysis.
In an email to WIRED, Amazon spokesperson Margaret Callahan defended the project, noting the on-site generation will not drive up electricity costs for Texas households. She added that the site will also integrate solar power and battery storage, and that the company remains committed to hitting its target of net-zero carbon emissions by 2040.
For data center developers, simple-cycle turbines have grown increasingly popular because they deliver flexible power that can be installed far faster than more efficient alternatives. Permit applications show at least three new plants being developed in Ohio by oil and gas firm Williams for Meta will also run on simple-cycle units. In Abilene, Texas, energy firm Crusoe is building a plant with 39 simple-cycle turbines to power a Microsoft data center on its Stargate campus. Permit documents for Google’s Goodnight data center, another Texas project developed by Crusoe, list 20 simple-cycle turbines as the facility’s core power source. Even xAI’s Colossus 1 and Colossus 2 data centers in Memphis rely on dozens of simple-cycle turbines for power. Every one of these projects is permitted to release millions of tons of greenhouse gases annually. In statements to WIRED, both Williams and Crusoe confirmed their gas plants meet all local and federal regulatory requirements and are fitted with modern emissions control technology.
Britt Burt, senior vice president at energy market intelligence firm Industrial Info Resources, says he did not expect data center developers to lean this heavily on simple-cycle turbine technology. “It’s not the most efficient way to use a gas turbine,” he explained. But for companies racing to bring new data centers online to meet booming AI demand, installing a set of simple-cycle turbines is far faster and cheaper than waiting for the extra equipment required for combined-cycle plants, especially amid multi-year backlogs for turbine orders and core components. “Right now, because of the way the supply chain is for turbines, they’re having to go with what they can get their hands on,” Burt said.
The unique energy demands of AI data centers also align with the key strengths of simple-cycle turbines. With fewer moving parts, simple-cycle units can start up and shut down far more quickly than larger combined-cycle plants, which fits the volatile energy needs of AI model training that requires sudden, massive spikes in power on demand. In fact, the stress of frequent on-off cycling has been so intense that many data center operators have reported turbines breaking down prematurely from the strain.
“If you have a highly fluctuating load, it's better to have 30 things that can individually turn on and off than to have one giant thing that doesn't like to be turned on and off,” said Stephen Lynch, a mechanical engineering professor at Penn State and director of the university’s Center for Gas Turbine Research, Education, and Outreach.
Simple-cycle turbines also have a major practical benefit for developers: they generally do not require water for cooling, a growing flashpoint for communities near new data centers already facing water scarcity concerns. Burt estimates that a 250-megawatt combined-cycle plant can use roughly 75,000 gallons of water per hour for cooling. The permit application for Amazon’s new Texas plant specifically cites simple-cycle technology’s lack of water consumption as an “environmental benefit.”
Currently, the federal government and a growing number of state governors from both major political parties are encouraging data center power providers to build their own on-site generation to avoid straining the main grid and driving up residential electricity costs. Some companies have pledged to eventually connect their private power plants to the public grid. Callahan noted Amazon’s Texas plant is “designed to transition to grid-connected service as interconnection timelines allow.” Williams similarly told WIRED earlier this month it may consider connecting its Ohio plants to the grid in the future.
Existing simple-cycle plants, known as peaker plants, are only used to quickly supply extra power to the main grid during periods of peak demand. But the current boom in new private off-grid (or islanded) power plants built with inefficient simple-cycle turbines—most of which are designed to operate for 30 years or more—risks creating a parallel, fossil fuel-powered grid that will generate millions of tons of excess greenhouse gas emissions for decades to come.
If the AI boom cools or tech companies’ energy demand drops unexpectedly, these massive inefficient gas plants could be left without a purpose. “We call that a stranded asset,” Burt said with a chuckle.
Update: August 14, 2026, 3:05 PM EDT: WIRED updated this article to correct the spelling of Amazon spokesperson Margaret Callahan’s name.
