Proposed U.S. gas-generating capacity jumped 44% in just eight months. According to an analysis by Environment America Research & Policy Center and Frontier Group, based on Energy Information Administration filings, planned grid-connected gas capacity through 2030 now stands at 60.4 gigawatts, up from 41.8 GW in December 2025.
That figure has nearly doubled since early 2023. The driver the groups identify is surging electricity demand from data centers running AI and other intensive computing workloads, which require continuous, highly reliable power.
What the Numbers Actually Say
60.4 GW is what developers are proposing. It is not what has been built.
The 60.4 GW figure represents developer intentions captured in EIA filings, not completed construction. Developers cancel, delay, downsize, or replace projects before they ever produce a watt.
To put the scale in context: 60.4 GW exceeds the total generating capacity of every individual U.S. state except Texas, California, and Florida, according to the groups’ comparison. Texas alone accounts for 33.2% of proposed additions, which would expand the state’s existing gas capacity by nearly 25%.
Louisiana (6.6 GW), Pennsylvania (4.4 GW), Oklahoma (3 GW), and Mississippi (2.6 GW) account for much of the rest. Together, those five states represent a substantial majority of the proposed pipeline.
Why AI Is in the Room
Gas turbines can be dispatched on demand, and that matters when your power customer never sleeps.
Data centers need uninterrupted power, around the clock, at scale. Gas-fired generation can deliver dispatchable electricity on demand. Solar and wind projects typically require storage, transmission upgrades, and firming resources that take considerably longer to develop, and that gap is where AI data centers are moving.
The 60.4 GW total does not capture the full picture. Environment America and Frontier Group say off-grid gas plants built to serve data centers directly could add tens of gigawatts beyond the grid-connected figure, though that estimate is a projection rather than a measured capacity total.
Off-grid plants aside, the analysis does not establish a one-to-one relationship between every proposed plant and a specific data center. Data-center growth is the major identified driver, not the sole explanation for the pipeline’s expansion.
On the renewable side, planned wind additions have fallen 33.9% since December 2025, according to the groups’ analysis. Solar remains the largest category of expected new generation, but the groups argue it is stalling rather than accelerating to meet rising demand. Perspectives from utilities, data-center operators, and grid regulators on whether renewables can realistically close that gap were not available in the underlying research and require additional reporting.
Projected Health and Climate Costs
The health modeling is a projection, but the policy question it raises is immediate.
Separate modeling research published on arxiv.org estimates that data-center-related pollution could contribute to roughly 600,000 asthma symptom cases and 1,300 premature deaths in 2028. That projection assumes a high-growth scenario, and the same research puts the potential annual public-health burden above $20 billion. Both figures are modeled estimates that depend on assumptions about growth rates, emissions levels, and health effects, not observed counts.
Environment America’s Andre Delattre has said the buildout could place the clean-energy transition at risk and lock in fossil-fuel dependence for decades. Johanna Neumann of the same organization has said new gas plants, delayed coal retirements, and slower renewable growth could harm public health, the climate, and future generations. Both statements reflect the organizations’ positions as paraphrased in available reporting; direct transcripts were not independently verified.
Whether this pipeline materializes depends on permitting outcomes, project financing, and interconnection timelines. The 44% jump tells you where developers are placing their bets right now. Whether those bets pay off is a separate question entirely.
