The rapid growth of computing power for artificial intelligence is radically transforming the energy landscape of the United States. According to my analysis of the latest data from Global Energy Monitor (GEM), the announced capacity of gas-fired power plants aimed at directly powering data centers has surged from 97 GW at the end of 2025 to an impressive 189 GW by mid-2026. For context: as recently as early 2024, this figure barely exceeded 4 GW. Thus, in just two and a half years, the volume of specialized projects has grown more than fortyfold, and over the last six months alone, it has nearly doubled.
This statistics includes both announced initiatives, projects at the pre-construction stage, and power plants already under construction. It is important to emphasize: a significant portion of these announced capacities may never be realized, adding an element of uncertainty to the overall picture.
Autonomous Generation as the New Standard
The key driver of this boom has been the behind-the-meter model, which involves building power plants directly adjacent to data centers. This approach allows operators to bypass overloaded power grids, connections to which take years in some regions. As GEM analysts note, the expansion of gas generation in the U.S. is now inextricably linked to the development of data centers—these two processes have become a single whole.
Notably, back in January, the total portfolio of gas projects in the country stood at about 252 GW, of which more than a third (97 GW) was intended specifically for data centers. If this entire volume were realized, the installed gas generation capacity in the U.S. would grow by nearly 50%, requiring capital expenditures exceeding $416 billion. However, new data shows that it is precisely the segment linked to data centers that continues to expand at the fastest pace.
Political Context and Market Risks
This shift coincided with a change in White House policy. In March, the largest technology giants—Amazon, Google, Meta, Microsoft, OpenAI, Oracle, and xAI—committed to independently powering their data centers and paying for grid infrastructure so as not to shift costs onto ordinary consumers. The program, expanded in July to include utility companies and state authorities, attracted more than 200 additional organizations, confirming the systemic nature of the transition.
Nevertheless, I urge viewing the 189 GW figure not as a forecast but as an indicator of market demand. The realization of these projects faces serious obstacles: a shortage of gas turbines, with order backlogs scheduled through 2030, as well as financing issues and local opposition. For two-thirds of the projects in the GEM global database, the equipment manufacturer is not even identified. Even partial realization of these plans will have long-term consequences, as gas plants are designed for decades of operation.
Geopolitical Contrast and the Texas Phenomenon
Interestingly, the U.S. has overtaken China in the total capacity of gas projects under development, demonstrating fundamentally different approaches to powering AI. Chinese data centers gravitate toward regions with surplus solar and hydroelectric power, following state policy to reduce energy dependence. The U.S., by contrast, chooses the speed and stability of gas, sacrificing environmental concerns and risking excess infrastructure if projected AI demand fails to materialize.
The epicenter of the boom has become Texas, where about 80.6 GW of gas capacity is under development, of which 40 GW is directly linked to data centers. The state already uses streamlined procedures for approving such plants, and Amazon recently acquired a site for the GW Ranch campus with its own 7.65 GW power plant featuring 35 turbines, planning to use autonomous generation until connection to the main grid.
My conclusion: We are witnessing a fundamental shift in U.S. energy strategy, where gas is becoming a temporary but critically important bridge for the AI revolution. However, the bet on fossil fuels creates long-term risks both for climate goals and for economic efficiency if the industry's growth rates slow down.