The United States has entered a new phase in the global power race, where the speed of artificial intelligence infrastructure is increasingly shaping the fuel mix built around it. Global Energy Monitor’s latest assessment shows the US moving ahead of China in gas-fired generation under construction, reversing a pattern that held for years as China expanded its electricity system alongside industrial growth. The shift matters because gas plants can provide the firm power that large AI campuses demand around the clock, particularly when grid connections, transmission upgrades and renewable projects cannot arrive on the same schedule.

The scale of the pipeline also suggests that the AI boom is influencing generation investment far beyond the physical boundaries of data centers. US gas capacity in development has climbed from 252 gigawatts at the beginning of the year to 378 gigawatts, putting roughly one-third of the global development pipeline inside the country. If developers complete the entire portfolio, the US gas fleet could expand by about two-thirds, representing more than $647 billion in capital expenditure according to Global Energy Monitor’s analysis.

Data Centers Are Rewriting the Generation Pipeline

The crucial development is not simply that American utilities are ordering more turbines; it is that the rapid expansion of AI data centers is driving a sharp increase in gas-fired capacity intended to serve their growing electricity demand. About half of the additional US gas capacity under development has a direct connection to planned data center demand, according to the analysis, tying generation decisions to an infrastructure cycle that has moved at extraordinary speed. AI training and inference facilities can consume electricity at a scale that is prompting developers to pursue dedicated generation, particularly when grid connections and transmission upgrades cannot arrive quickly enough to meet projected demand.

That pressure creates an incentive to build generation close to the demand rather than wait for slower grid expansion. The result is an unusual coupling between computing capacity and fossil-fuel infrastructure, with decisions about GPUs, substations, turbines and fuel supply increasingly influencing one another. For infrastructure investors, the important question therefore moves beyond whether a data center can obtain power and toward whether the power architecture supporting it will remain competitive over the life of the computing asset.

China’s Former Gas Advantage Has Reversed

China’s position makes the reversal more consequential than a simple US construction surge. For years, China had more gas-fired capacity under construction than the US, but that position has now reversed as American projects surged during the first half of 2026. Global Energy Monitor now sees that relationship changing sharply, with American projects under construction reaching roughly twice the capacity of those in China. When projects that developers have announced or placed in pre-construction stages enter the calculation, the US pipeline approaches three times China’s level.

Jenny Martos, a project manager at Global Energy Monitor, described the change directly: “Six months ago, China had more gas plants under construction but that has now flipped.” Her observation captures a broader shift in the geography of incremental gas investment, where the immediate catalyst increasingly comes from computing rather than conventional industrial demand. That distinction matters because AI loads can arrive in concentrated bursts, potentially creating generation requirements that conventional electricity forecasts would have distributed across a much longer period.

The Emissions Equation Is Getting Harder

The environmental consequence becomes clearer when the generation pipeline is considered as an AI infrastructure strategy rather than a collection of independent power projects. Gas burns more cleanly than coal at the plant level, but a massive expansion of gas capacity still creates long-lived fossil infrastructure and exposes electricity customers to fuel-market volatility. One estimate cited in the referenced analysis indicates that gas plants associated with the data center expansion could raise US power-sector emissions by as much as 20%. Martos warned about the duration of that commitment: “There has been an enormous surge in datacenter proposals powered by gas in the past year and the climate implications of that is huge.

Building all of this gas for AI locks in decades of pollution and it is also locking in dependence on a volatile fuel cost, which will get passed down to rate payers.” That risk extends beyond carbon accounting because a turbine can remain economically relevant for decades even as the GPUs it was built to support become obsolete much sooner. The infrastructure mismatch could leave utilities and customers carrying generation costs long after the original AI demand assumptions have changed.

Turbine Constraints Are Adding Another Risk

The AI buildout has also exposed a less visible bottleneck inside the gas-generation supply chain: turbine availability, with leading manufacturers reporting rising order backlogs and multi-year lead times. The rapid expansion of data centers has contributed to a surge in gas-turbine demand, with leading manufacturers reporting rising order backlogs and multi-year lead times. Developers are increasingly turning to simpler-cycle gas turbines because they can start and stop more rapidly than combined-cycle units, making them better suited to the sharp power swings associated with AI workloads, although they operate less efficiently and produce more emissions. The problem illustrates why AI power procurement cannot stop at a signed interconnection agreement or a headline capacity figure.

A megawatt of nominal generation capacity does not tell investors how efficiently that capacity converts fuel into electricity, how quickly equipment can arrive or what operating costs will look like under changing gas prices. Elon Musk’s xAI illustrates the scale of the shift toward dedicated gas generation, with Reuters reporting that the company installed 59 natural-gas turbines for its Colossus 2 data center project. In a market where computing demand is expanding faster than industrial supply chains can adapt, the constraint can migrate from chips to transformers, turbines, transmission equipment and ultimately fuel infrastructure.

America’s Energy Narrative Is Changing

The US gas surge also complicates a political argument that has treated Chinese fossil-fuel expansion as evidence that American climate action cannot materially change global emissions. China remains the world’s largest carbon emitter, but its rapid deployment of renewable energy is altering the comparison at the same moment that the US is accelerating fossil-fuel infrastructure around AI. The International Energy Agency expects US spending on coal- and gas-fired power plants to exceed China’s for the first time in decades, adding another dimension to the changing energy relationship between the two economies. That does not mean China has abandoned fossil fuels or that the US has stopped adding clean generation. America is increasingly treating AI as a strategic industrial load that requires rapid firm power, while China’s broader energy buildout gives renewable deployment a much larger role alongside conventional generation.

The Political Push Is Reinforcing the Buildout

The Trump administration has positioned data centers as a major economic opportunity and has supported faster development as part of its broader push to expand US AI infrastructure. President Donald Trump has argued that data centers could become an economic engine comparable with major legacy industries and has encouraged states to use tax incentives to attract projects. In recent remarks, Trump said: “Frankly, communities that don’t take a datacenter, they’re making a mistake because there’s plenty of communities that want them.” He added: “Because it means jobs, and it means tremendous tax revenue coming into the towns and cities that take them.” Those incentives can accelerate development, but they also complicate the calculation around who pays for supporting infrastructure. A data center can deliver jobs and tax revenue locally while simultaneously requiring substantial generation, transmission, water and road investment that extends beyond the project’s immediate footprint.

AI Power Could Outlive the AI Hardware

The deepest issue in the gas buildout is the possibility that electricity infrastructure will have a much longer economic life than the technology creating the demand. AI hardware evolves on compressed replacement cycles, while gas turbines, substations and transmission assets can operate for decades. If utilities construct generation around today’s exceptionally strong AI growth assumptions, tomorrow’s computing architecture could alter the amount, location or timing of electricity demand before those plants reach the end of their useful lives. That creates a form of infrastructure lock-in that does not appear on a GPU procurement sheet. Developers can replace accelerators, migrate workloads and redesign server halls, but the surrounding power system cannot make those adjustments at the same speed. The strategic advantage will therefore belong to operators that treat power flexibility as an asset rather than assuming today’s AI load forecast will remain permanent.

The Real US-China Race Is Now About Power Architecture

The US overtaking China in new gas generation marks more than a change in construction statistics; it signals how profoundly AI has begun to influence national energy strategy. America is building a generation pipeline in which a substantial share of new gas capacity is being developed in anticipation of sustained data-center demand, alongside other electricity needs. China’s simultaneous expansion of clean energy makes the comparison even more revealing, because the two largest technology and economic powers are increasingly taking different routes toward satisfying rising electricity demand. The answer will shape emissions, electricity prices, grid investment and the competitiveness of American data centers long after today’s GPU cycle has passed. The defining infrastructure decision of the AI era may ultimately be less about how many chips America can deploy than how much permanent energy infrastructure it builds to keep them running.