AI data centers ignite backlash over gas-powered expansion

By mid-2026, the rapid expansion of behind-the-meter AI data center power plants has become undeniable, with at least four facilities constructed in the past 18 months totaling 2 GW of onsite capacity—equivalent to the output of two nuclear power plants. Elon Musk’s xAI Colossus 1 and 2 data centers near Memphis stand out as the largest, operating 1.5 GW of mobile gas turbines mounted on flatbed trailers, making it the world’s largest behind-the-meter data center power installation. This scale of deployment underscores a significant shift as AI operators seek to circumvent slow utility grid connections by building massive, self-sufficient power plants on-site.

Despite this growth, the landscape is marked by stark uncertainties and uneven progress. High-profile projects like Fermi America’s Project Matador have stalled, with satellite imagery revealing months of inactivity due to the absence of tenant commitments and funding, illustrating the precarious nature of many announced developments. Meanwhile, some operators, including xAI, have deployed dozens of unpermitted gas turbines, highlighting regulatory and operational ambiguities that accompany this rapid scale-up. This mix of operational successes and stalled ambitions paints a complex picture of the behind-the-meter trend’s maturation.

Innovation is a hallmark of this emerging sector, as developers employ unconventional and mobile power generation technologies to accelerate deployment and sidestep traditional infrastructure bottlenecks. Strategies include mounting gas turbines on semi-trucks and repurposing jet engines from Boeing 747s, as seen in Crusoe’s $1.25 billion deal with Boom Supersonic for 29 jet-engine turbines and Wärtsilä’s turbine supply for Texas data centers. This inventive approach not only addresses supply chain constraints but also reflects the urgent scramble among AI firms to secure reliable, high-capacity power amid soaring demand.

While skeptics dismiss behind-the-meter data center power plants as hype, the reality is more nuanced; several projects are operational or actively under construction, signaling tangible growth despite the challenges. Analysts who point to stalled projects like Fermi overlook the substantial capacity already online and the ongoing investments by major players such as Oracle’s plan to purchase 2.8 GW of fuel cells from Bloom Energy. This dynamic underscores a sector in flux—balancing bold experimentation and rapid expansion with regulatory hurdles and market uncertainties.

By mid-2026, AI data center developers have exploited regulatory loopholes and federal rollbacks to rapidly deploy large-scale fossil fuel power plants with minimal environmental review and public input. Companies like xAI near Memphis have installed dozens of mobile gas turbines on flatbed trailers, bypassing traditional grid connections and permitting processes, which severely limits opportunities for community engagement and scrutiny. This approach has allowed behind-the-meter power plants totaling over 2 GW capacity to emerge in just 18 months, highlighting how these projects advance despite scant regulatory barriers and growing environmental concerns.

In Texas, regulators have aggressively fast-tracked AI data center power plant construction by repurposing minor source permits—originally intended for small polluters like dry cleaners—to approve massive natural gas facilities with little oversight. Facilities such as Stargate have leveraged these rubber-stamped permits to avoid major air quality reviews, blindsiding local communities like Omaira Garcia’s ranch in Abilene with sudden construction and significant pollution. This loophole has facilitated plans for at least 15 gas plants that could emit over 130 million tons of greenhouse gases annually, equivalent to 35 coal plants, locking in decades of emissions with minimal public awareness or input.

The EPA’s 2026 proposal to delegate public notice and participation requirements for minor source permits to states like Texas threatens to further erode community voices and transparency, especially where regulators such as the Texas Commission on Environmental Quality (TCEQ) are criticized for weak enforcement. While the EPA argues this rule aims to reduce administrative burdens and speed permitting without weakening emission standards, critics including environmental groups warn it could roll back recent reforms designed to increase public input, effectively silencing communities facing pollution from backup diesel generators at data centers like Vantage near San Antonio.

Federal agencies and courts have increasingly framed AI data center power plants as critical national infrastructure, using this designation to limit enforcement of environmental laws and public participation. The Department of Justice’s dismissal of a NAACP lawsuit against xAI’s unpermitted gas turbines, citing national security concerns tied to military operations, exemplifies how successive administrations wield discretionary power to ease construction of favored infrastructure while curtailing legal challenges. This regulatory shift, coupled with nondisclosure agreements between developers and local governments, deepens the opacity surrounding these projects and diminishes community oversight.

Under the Trump administration, federal policies aggressively fast-tracked the approval of 74 methane gas plants dedicated to powering AI data centers, prioritizing fossil fuels while sidelining renewables like wind. This approach included waiving environmental protections and declining to impose nationwide regulations on data centers, effectively granting Big Tech broad authority to develop polluting, water-intensive facilities. In response, state and local governments, along with lawmakers such as Bernie Sanders and Alexandria Ocasio-Cortez, pushed back with moratoriums and regulations, highlighting a stark divide between federal ambitions and localized environmental concerns.

Texas emerged as a flagship example of permissive state-level regulation aligning with federal priorities, with Governor Greg Abbott embracing the AI data center boom as a new 'gold rush.' Exploiting regulatory loopholes, Texas allowed data centers like OpenAI’s Stargate to secure minor air permits—typically reserved for small businesses—bypassing major environmental reviews and public engagement. This regulatory laxity sparked significant community opposition and environmental justice concerns, as residents such as Omaira Garcia were blindsided by construction near their homes, underscoring the social costs of Texas’s industry-friendly stance.

Contrasting sharply with Texas, New York State instituted a one-year moratorium on new AI data centers with power demands exceeding 50 MW, marking the first statewide suspension of environmental permitting for such facilities. Governor Kathy Hochul’s executive order reflected mounting concerns over unsustainable electricity and water consumption, as well as infrastructure strain and rising utility costs borne by communities. While the Data Center Coalition warned this could shift investments and jobs to less restrictive states like Texas and Virginia, some hyperscale operators like Meta sought to mitigate backlash by funding grid expansions and supporting low-income households, signaling nuanced corporate responses to regulatory pressures.

The EPA, under Administrator Lee Zeldin, further accelerated fossil fuel power plant approvals for AI data centers by proposing rules that reduce public participation in permitting processes, such as eliminating federally mandated comment periods for minor pollution sources. This regulatory easing culminated in a July 2026 memo exempting 'islanded' off-grid data center power plants from the Clean Air Act's Acid Rain Program, removing sulfur dioxide and nitrogen oxide emission caps for nearly 60 such facilities. Justified as essential for maintaining U.S. AI dominance and protecting ratepayers from electricity price spikes, these policies drew sharp criticism from environmental groups like the Sierra Club and EDF, who warned that silencing community voices threatens public health and environmental justice. However, the recent Supreme Court decision overturning Chevron deference raises the prospect that courts may soon challenge these agency interpretations.

The proliferation of onsite fossil fuel power plants at AI data centers is driving up energy costs for American consumers by competing directly for natural gas supplies, which are market-traded commodities. Unlike grid-connected facilities, these data centers negotiate gas contracts outside state utility regulators' oversight, enabling them to secure cheaper gas at the expense of other users, thereby inflating prices for homes and businesses nationwide. Moreover, the reliance on inefficient, back-ordered gas turbines exacerbates pollution and undermines environmental regulations, as seen in the xAI lawsuit in Mississippi, where continuous operation of unpermitted turbines has led to significant toxic emissions.

Communities neighboring AI data centers with onsite fossil fuel plants, such as Omaira Garcia’s ranch in Abilene, Texas, face severe air quality degradation and health risks without prior notification or meaningful engagement. Regulatory loopholes in Texas allow these facilities to obtain minor air permits—originally designed for small businesses like dry cleaners—bypassing comprehensive environmental reviews and public scrutiny. This 'small first, big later' permitting strategy has enabled rapid expansion of polluting plants, disproportionately burdening vulnerable populations, including predominantly Black neighborhoods near xAI’s Colossus 2 data center, where cancer risks are already four times the national average.

The environmental footprint of these onsite fossil fuel plants is staggering, with nine planned Texas gas plants linked to AI data centers projected to emit over 130 million tons of greenhouse gases annually—equivalent to the emissions of 35 coal-fired power plants or entire countries. This surge in emissions not only threatens climate goals but also locks in fossil fuel dependency for decades, while simultaneously increasing local pollution of nitrogen oxides, carbon monoxide, and formaldehyde. Despite these impacts, federal exemptions like the EPA’s removal of off-grid AI power plants from the Acid Rain Program caps have left nearly 60 such facilities unregulated for sulfur dioxide and nitrogen oxides, intensifying environmental justice concerns for nearby communities.

The cumulative effect of onsite fossil fuel power plants at AI data centers extends beyond environmental harm to economic strain, as rising pollution and infrastructure demands contribute to higher energy bills for consumers. In New York, for example, Governor Hochul has highlighted community opposition driven by concerns over increased utility costs, noise, and depleted resources. While developers tout economic benefits such as funding local services, residents like Garcia feel 'trapped' by the environmental and financial burdens imposed without transparent communication or adequate mitigation, underscoring the urgent need for equitable energy policies and robust community engagement.

The EPA's 2026 proposals to curtail federally mandated public participation in permitting 'minor source' air pollution permits for data center backup generators have sparked significant backlash from environmental groups and public health advocates. Critics warn that shifting authority to states like Texas and Utah to determine public notice requirements risks silencing community voices and eroding transparency, especially in regions where state regulators such as the Texas Commission on Environmental Quality have been criticized for weak enforcement. This regulatory rollback threatens to undermine recent state-level reforms aimed at enhancing public input and could disproportionately disadvantage communities near industrial sites, as highlighted by Sierra Club's Luis Miranda and Utah’s Division of Air Quality director Bryce Bird, who emphasize the importance of maintaining public comment processes to uphold accountability.

The Department of Justice's novel legal strategy defending xAI's operation of 59 unpermitted gas turbines in predominantly Black communities exemplifies a troubling shift in enforcement of the Clean Air Act. By arguing that citizen suits must be dismissed if projects are deemed political priorities tied to national security and economic interests—citing AI infrastructure's role in military operations—the DOJ effectively shields major polluters from accountability. This approach departs from traditional Clean Air Act enforcement, as the government refrains from prosecuting xAI itself while seeking to block public interest litigation, raising profound concerns about the erosion of public enforcement rights and environmental justice for communities facing cancer risks four times the national average.

The EPA's July 2026 memo exempting off-grid, 'islanded' power plants serving AI data centers from the Acid Rain Program's sulfur dioxide and nitrogen oxides emission caps marks a significant regulatory retreat that undermines longstanding environmental protections. Affecting nearly 60 sites, this exemption removes critical federal pollution limits despite these facilities emitting the same pollutants as traditional coal or gas plants, effectively stripping nearby communities of federal safeguards. EPA Assistant Administrator Aaron Szabo framed the exemption as essential to maintaining U.S. AI dominance and economic prosperity, prioritizing rapid AI infrastructure deployment over environmental enforcement, a rationale that extends beyond AI data centers to other off-grid fossil-fuel generators like bitcoin mining operations, thereby weakening the broader regulatory framework.

Environmental advocacy groups are actively leveraging citizen enforcement under the Clean Air Act to challenge data center operators like Vantage Data Centers and VoltaGrid for allegedly circumventing stricter air permitting requirements through regulatory loopholes. By obtaining 'minor source' permits to avoid 'major source' scrutiny, these operators risk emitting thousands of tons of nitrogen oxides and volatile organic compounds annually in serious ozone nonattainment areas such as San Antonio. These legal challenges underscore growing concerns about the erosion of public accountability and enforcement rights amid EPA proposals to accelerate power generation construction and narrow public participation, highlighting the contentious legal environment surrounding AI data center power generation.

By mid-2026, AI data center operators are pushing the boundaries of clean energy innovation, experimenting with avant-garde solutions like Panthalassa's $140 million wave-powered buoy data centers and Meta's partnership with Overview Energy to harness solar power from space, beaming it down via infrared lasers. These ambitious projects reflect the sector's urgent drive to reconcile soaring computational demands with sustainability goals, even as they remain largely futuristic and unproven at scale.

However, the immediate reality of AI data center expansion reveals a stark tension between visionary clean energy aspirations and the pressing need for reliable power, leading many operators to rely heavily on behind-the-meter gas plants that emit substantial greenhouse gases and local pollutants. Elon Musk’s xAI, for instance, has deployed dozens of unpermitted gas turbines near Memphis, illustrating how short-term infrastructure demands often clash with environmental stewardship and regulatory frameworks.

This growing energy appetite has not gone unnoticed by policymakers, with jurisdictions like Seattle and Flint, Michigan instituting moratoriums on new AI data center developments, while Texas’ governor has urged that new facilities bear the costs of grid interconnection and upgrades. These measures underscore the escalating political and regulatory challenges in balancing the rapid expansion of AI infrastructure with sustainable energy policies, signaling a future where innovation must be matched by rigorous oversight to ensure environmental accountability.