The rapid expansion of artificial intelligence technology is driving a surge in new natural gas power projects across the United States, with potential emissions equivalent to those of entire nations. Air permit documents reveal that just 11 data center campuses could collectively emit over 129 million tons of greenhouse gases annually. This aggressive energy push directly challenges established climate targets, according to environmental watchdogs.
The drive to power burgeoning AI operations has propelled tech companies to pursue an unconventional strategy: building their own power plants. This approach, known as "behind-the-meter" power, allows data centers to bypass the traditional electric grid, circumventing long waits for utility connections and public backlash over rising energy costs. It represents a fundamental shift in how large-scale computing infrastructure secures its energy needs.
This trend has seen explosive growth. Research from Global Energy Monitor, a nonprofit tracking oil and gas, indicates that the pipeline for data-center-specific gas power in the U.S. swelled from just 4 gigawatts in early 2024 to nearly 100 gigawatts by the start of 2026. This is a dramatic escalation.
Michael Thomas, founder of clean energy research firm Cleanview, monitors gas permits for data centers nationwide and views this as a dangerous acceleration of emissions. "It's almost like we thought we were on the downside of the Industrial Revolution, retiring coal and gas, and now we have a new hump where we’re going to rise," Thomas told WIRED, expressing deep concern. In Memphis, Tennessee, xAI, a prominent AI company, ignited local controversy by installing gas turbines at its Colossus 1 data center campus to power its Grok AI development. Community members in the nearby low-income Black neighborhood protested, worried about air pollution.
Environmental Protection Agency approved the turbines last year, regulators later granted a permit for an xAI affiliate's second campus, Colossus 2, in Southaven, Mississippi, despite widespread local opposition. The NAACP filed a lawsuit against xAI last week, alleging illegal operation of the turbines. What this actually means for your family is a tangible threat to air quality and local health outcomes.
Permit applications for both the Memphis and Southaven Colossus campuses project that the turbines at each site could generate more than 6.4 million tons of carbon dioxide equivalents annually. Combined, these two sites alone could produce emissions roughly equal to those from over 30 average-sized natural gas plants. That is enough energy to power 1.5 million homes.
Meanwhile, Microsoft is reportedly exploring a power purchase from a Chevron-backed natural gas project in West Texas. That single project's permit suggests it could emit over 11.5 million tons of greenhouse gases each year, surpassing the annual emissions of the entire country of Jamaica. Melanie Nakagawa, Microsoft's chief sustainability officer, stated the company employs a varied approach to energy, using diverse solutions for reliability while investing in carbon-free electricity.
She noted that dedicated onsite energy infrastructure might be part of this strategy in certain regions, particularly where grid limitations hinder rapid deployment. This suggests a pragmatic acceptance of gas power in the face of infrastructure challenges. The policy says one thing about green commitments, but the reality often requires different solutions in the short term.
These emissions figures are derived directly from publicly available air permit documents in state databases, as well as materials collected by Cleanview and the Environmental Integrity Project’s Oil and Gas Watch database. It is important to note that permitted emission numbers often represent a maximum theoretical output, assuming continuous operation at full capacity. Actual greenhouse gas emissions from traditional, grid-connected power plants are typically lower.
Alex Schott, communications director for Williams Companies, which is building three behind-the-meter power plants in Ohio for Meta, confirmed this. Schott stated to WIRED that internal modeling shows actual emissions could be "potentially two-thirds less than what's on paper."
However, the unique operating demands of data centers may narrow this gap. Energy researcher Jon Koomey estimates that while efficient grid-connected gas plants might emit 40-50 percent of their permitted numbers, data center emissions could align much more closely with permit models. This is because data centers maintain consistent, high power requirements and do not typically adjust to fluctuating customer demand.
A November permit application for a data center by AI company Crusoe, involved in three projects reviewed by WIRED, described its facility as "unlike a traditional power plant" that must "respond to the demands of a constantly varying grid." It explicitly stated that "the power requirements do not vary significantly" at the data center. Andrew Schmitt, Crusoe’s senior director of communications, told WIRED that the company views gas as a necessary "bridge—not the destination—as we work to build AI infrastructure that meets the scale of demand while expanding access to innovative forms of energy over time."
Koomey also pointed to a global shortage of the most efficient gas turbines, partly due to the data center boom. This shortage prompts some developers to opt for less efficient models, which must run longer and, consequently, generate more emissions. "[Data center operators’] belief is that the value being delivered by the servers is much, much more than the cost of running these inefficient power plants all the time," Koomey observed. This business calculus prioritizes computational power over immediate environmental impact.
Among the largest potential contributors to emissions are projects associated with the Stargate Project, a multi-company AI initiative originally for OpenAI. Stargate campuses are emerging across states including Texas, New Mexico, Ohio, and Wisconsin. Permit documents for just three Stargate-affiliated gas projects – one near Abilene, Texas, and two for Project Jupiter in New Mexico – indicate a combined potential to emit over 24 million tons of greenhouse gases each year.
Hormuz Closure Strands 20,000 Seafarers; Global Oil Markets Face Crisis
Aaron McLear, an OpenAI spokesperson, affirmed the company's commitment to protecting ratepayers and building infrastructure for U.S. AI leadership. McLear added that where near-term natural gas is essential for reliable power, OpenAI collaborates with partners to employ modern, efficient generation while also accelerating clean power and grid modernization.
Julia Allyn Fishel, an Oracle spokesperson, informed WIRED that a "modification" to the Project Jupiter application is underway, which is expected to "materially lower emissions." The new estimates have not yet been made public by the New Mexico Environment Department. Fishel emphasized Oracle's commitment to covering its own energy costs and implementing optimal energy solutions for each community, ensuring ratepayers' bills and grid reliability remain unaffected by their AI data centers. A fourth gas plant on the main Stargate campus in Abilene, being built by Crusoe for Microsoft, could permit over 7.8 million tons of carbon dioxide equivalents annually.
Microsoft declined to comment on this specific project. Even larger potential carbon footprints exist. Outside Amarillo, Texas, Fermi is constructing what it calls the "President Donald J.
Trump Advanced Energy and Intelligence Campus," aiming for 17 gigawatts. Fermi continually promotes its use of "clean" natural gas. However, documents show the maximum emissions for its two combined gas projects could reach over 40.3 million tons of CO2 equivalents each year.
This figure exceeds the yearly emissions of all power sources in the entire state of Connecticut. Five hours south, near Fort Stockton, Pacifico Energy is developing a 7.2-gigawatt data center campus, powered by a gas project permitted to emit over 33 million tons of greenhouse gases annually. Pacifico did not respond to a request for comment.
These figures illustrate the sheer scale of the energy demands. Major tech companies, many of whom have made public pledges to reduce carbon emissions, acknowledge that the AI infrastructure buildout is hindering their environmental goals. The scale of these gas projects demonstrates how readily a few fossil fuel plants can tip the balance against sustainability efforts.
For example, Meta is linked to three behind-the-meter gas projects in Ohio, with permit documents suggesting a maximum potential of 5.5 million tons of CO2 equivalents annually. Meta's 2025 sustainability report claims a reduction of 23.8 million metric tons in greenhouse gas emissions since 2021. Yet, if the Ohio projects emit even half of their permitted levels, that would erase more than 10 percent of Meta's stated emissions reductions over four years.
Meta declined to comment. These Ohio projects are not Meta's only fossil fuel ventures. Most major AI companies building behind-the-meter power also engage with utilities for grid-connected power plants.
Meta has an agreement with utility Entergy to power its Hyperion data center in Richland Parish, Louisiana. A new gas plant being built by Entergy for Meta could emit nearly 5.2 million tons of greenhouse gases annually. Earlier this month, Meta announced it would fund seven new natural gas plants, totaling over five gigawatts, to serve both its data centers and Entergy customers.
These facilities, the announcement stated, are designed with future carbon capture capabilities, which could reduce some emissions. The policy says one thing about future solutions, but the reality is more gas plants today. In March, several companies linked to these projects signed the Ratepayer Protection Pledge, a nonbinding agreement sponsored by the Trump administration, which encourages AI companies to "build, bring, or buy" power generation for data centers.
Experts told WIRED that this pledge was largely symbolic, with neither the White House nor tech companies holding significant control over policies that truly lower consumer electric bills. This highlights a disconnect between political gestures and tangible policy impact. Last month, three Senate Democrats questioned several leading tech companies, including OpenAI, Meta, and Fermi, about data center emissions.
Fermi, in its response to lawmakers, urged support for nuclear energy and its campus's inclusion in foreign nuclear investment deals. The company also claimed its behind-the-meter power was exempt from greenhouse gas emission reduction regulations since it would not connect to the electric grid. "Clean natural gas is fundamental to the energy transition and is the logical bridge to nuclear for a nation that cannot afford to wait," Fermi's letter stated. The company did not clarify if it would retire its gas turbines once planned nuclear capabilities become operational.
It is improbable that every proposed gas facility will ultimately be constructed; an air permit does not guarantee building. Neither Fermi nor the GW Ranch facility, which represent the two largest potential emitters on this list, currently have a client. On Friday, Fermi announced its CEO would step down immediately, though he remains on the board, and called for the company to be sold.
Stocks plunged over 20 percent, and the company’s CFO also departed. The Stargate project has seen strategy shifts, with OpenAI pausing a planned data center expansion in the UK this week. Turbine shortages, rising labor and construction costs, and energy market disruptions in the Middle East could all create obstacles for AI companies developing their own power sources.
Many of these companies are also actively pursuing renewable energy and nuclear power solutions for their data centers. Despite potential roadblocks, Michael Thomas views behind-the-meter gas power as a persistent trend for data centers, carrying serious climate implications. "The thing that has kept me up at night, and is starting to really worry me," Thomas shared, "is what happens if this gets 10 times bigger?" The sheer scale of proposed projects suggests this is a question demanding urgent attention. Observers will watch closely to see if tech giants reconcile their aggressive expansion with their stated sustainability commitments, or if the race for AI dominance continues to accelerate fossil fuel consumption.
The coming months will show whether community opposition and regulatory scrutiny can slow this rapid buildout. - The AI industry's rapid growth is driving a significant increase in new natural gas power projects. - These "behind-the-meter" power plants could generate vast greenhouse gas emissions, potentially exceeding some national totals. - Tech companies face criticism for these projects, which appear to contradict their public carbon reduction pledges. - The debate continues over whether permit-based emission estimates accurately reflect real-world output for data centers. This trend represents a critical juncture for climate action and the future of energy policy. For working families, particularly those in communities near these planned facilities, the implications are direct: potential impacts on local air quality and the long-term cost of living.
The push for AI dominance is clashing with environmental goals, creating a complex challenge that affects everyone. Policymakers, industry leaders, and local communities will need to navigate this conflict, balancing technological advancement with environmental stewardship. The decisions made now will shape energy landscapes and air quality for decades to come.
Key Takeaways
— - The AI industry's rapid growth is driving a significant increase in new natural gas power projects.
— - These "behind-the-meter" power plants could generate vast greenhouse gas emissions, potentially exceeding some national totals.
— - Tech companies face criticism for these projects, which appear to contradict their public carbon reduction pledges.
— - The debate continues over whether permit-based emission estimates accurately reflect real-world output for data centers.
Source: WIRED
