Across the United States, a wave of new gas-fired power plants is being built specifically to feed the electricity needs of artificial intelligence data centers, and a growing body of industry analysis suggests many of these facilities are markedly less efficient — and dirtier — than the modern gas plants that supply the broader grid. The pattern has emerged most visibly in Texas, where developers racing to bring power online for hyperscale computing campuses have increasingly turned to simpler, faster-to-build turbine technology rather than the more advanced combined-cycle plants that dominate new gas generation elsewhere in the country.
The distinction matters because not all gas power plants are built the same way. Combined-cycle plants capture waste heat from a gas turbine and use it to generate additional electricity through a steam cycle, substantially boosting the amount of power extracted from each unit of fuel burned. Simple-cycle plants skip that second step, sacrificing efficiency in exchange for lower upfront cost and, crucially, a much shorter construction timeline. For data center developers under pressure to bring massive new computing capacity online within months rather than years, that speed advantage has proven decisive, even though it means burning more gas — and emitting more carbon dioxide and other pollutants — for every megawatt-hour produced.
Why Data Centers Are Choosing Older Technology
The rush to power AI infrastructure has collided with long-standing bottlenecks in the U.S. electricity grid, where interconnection queues for new generation and transmission upgrades can stretch on for years. Rather than wait for utilities to add capacity through conventional planning processes, some data center operators and independent power producers have opted to build dedicated, behind-the-meter gas plants that can be permitted and constructed quickly, sidestepping the longer lead times associated with higher-efficiency combined-cycle turbines and their associated steam infrastructure. Analysts tracking these projects have pointed to a broader trend in which speed-to-power has become the dominant factor shaping technology choices, often outweighing considerations of fuel efficiency, emissions intensity or long-term operating cost.
The result, according to energy researchers, is a growing fleet of power plants built explicitly to serve computing demand that operate at a disadvantage compared with the modern gas fleet, undercutting broader efforts to curb power-sector emissions even as the technology sector publicly emphasizes sustainability commitments. The tension underscores a wider challenge facing the AI boom: the pace of computing demand growth is outstripping the pace at which cleaner power infrastructure can realistically be built, pushing some developers toward faster but more carbon-intensive interim solutions.
For Gulf observers, the episode carries relevance beyond U.S. borders. The UAE and its neighbors are themselves in the midst of an aggressive build-out of data center capacity to support national AI strategies, cloud infrastructure investment and partnerships with global technology firms. Unlike Texas, where developers have leaned on quickly deployable gas turbines, Gulf states have generally paired data center growth with a broader energy mix that includes nuclear power, utility-scale solar and natural gas, alongside continued investment in grid capacity. The American experience nonetheless offers a cautionary reference point for regional planners: as computing demand accelerates, the pressure to prioritize speed over efficiency in power procurement is likely to intensify globally, making the design choices behind new generation capacity — not just its fuel source — a critical factor in determining the true environmental footprint of the AI buildout.
