The scale of gas-fired generation aimed at powering data centers in the United States is expanding rapidly. According to my analysis of the latest data from Global Energy Monitor (GEM), the announced capacity of such power plants has surged from 97 GW at the end of 2025 to more than 189 GW by mid-2026. This is unprecedented growth: as early as the beginning of 2024, the figure barely exceeded 4 GW, meaning that over two and a half years the volume has increased more than 40-fold, and in just the last six months it has nearly doubled.
In this statistic, I include announced projects, facilities at the pre-construction stage, and plants already under construction. However, it is important to emphasize: a significant portion of this capacity may never be realized. This is not a forecast, but rather an indicator of market demand.
Captive Power Plants: A New Standard for Hyperscalers
The key driver of the boom is the spread of the behind-the-meter model, where a power plant is built directly next to a data center and operates bypassing the main grid. This approach allows developers to avoid multi-year queues for connection to overloaded power systems. As GEM analyst Jenny Martos rightly notes, "increasingly, the expansion of gas-fired generation in the United States is directly linked to the construction of data centers—one can no longer be discussed without the other."
As early as January, I recorded nearly 252 GW of all gas-fired generation under development in the United States, of which about 97 GW was intended directly for data centers. If this entire portfolio is realized, installed capacity will grow by almost 50%, and capital expenditures will exceed $416 billion. The new data shows that it is precisely the segment tied to data centers that continues to grow most dynamically.
Political Context and Autonomy
The shift toward self-supplied energy coincided with a change in White House policy. In March, Amazon, Google, Meta, Microsoft, OpenAI, Oracle, and xAI committed to independently building or purchasing new energy sources for their data centers, as well as paying for grid infrastructure. The goal is not to shift costs onto ordinary consumers. By July, the program had expanded to include utility companies and state authorities, gathering more than 200 additional participants.
Captive generation provides speed, but strategically it is a double-edged sword. Gas turbines are a scarce commodity: the order backlogs of the largest manufacturers are already stretched into 2030, and for two-thirds of the projects in the global GEM database, the equipment supplier has not even been identified. As Martos notes, "there remains much uncertainty here: financing, local opposition, moratoriums, turbine supply constraints." Even partial realization of these plans will have long-term consequences, as gas plants are designed for decades of operation.
Geopolitical Contrast: The United States vs. China
The growth of gas-fired generation allowed the United States to overtake China in total capacity of projects under development back in 2025: 252 GW versus 153 GW for the PRC. The difference in approaches is striking: Chinese data centers are located in regions with an abundance of solar and hydroelectric power, which is tied to state policy on developing renewable generation and reducing energy dependence. The United States, meanwhile, chooses gas for the sake of speed and stability, but pays for it with additional emissions and the risk of excess infrastructure if the projected demand from AI fails to materialize.
Texas: The Epicenter of the Boom
The trend is especially pronounced in Texas. In January, I counted 80.6 GW of gas capacity under development in the state, of which about 40 GW was directly linked to powering data centers. By July, developers were actively using streamlined procedures to approve captive plants. At that time, about 300 data centers were operating in the state, with roughly another 200 under development. A telling example is Amazon, which purchased a site in Pecos County for a campus with its own gas plant, GW Ranch, with a capacity of 7.65 GW—that is 35 turbines for autonomous power supply before connection to the main grid.
My expert assessment: the gas-fired generation boom for AI is a temporary solution to a structural problem. Betting on fossil fuels creates long-term commitments that could become a burden if AI growth rates slow down. Investors and operators should closely monitor the ratio of capacity commissioning speed to actual demand; otherwise, we risk getting new "ghost" assets in the energy sector.