最新研究显示，数据中心在过去两年推动美国燃气发电需求激增。全球能源监测组织发现，与数据中心明确相关的燃气项目需求在2025年达到97吉瓦，较2024年的略高于4吉瓦增长近25倍。总体来看，数据中心已使美国燃气发电需求在两年内接近三倍增长。当前美国电网中燃气装机容量约为565吉瓦，若开发管线中的项目全部建成，将新增近252吉瓦，使燃气电力规模扩大近50%，相当于为数千万美国家庭供电的能量。
扩张的环境影响显著。天然气虽较煤炭清洁，但2022年约占美国能源相关二氧化碳排放的35%。更大的风险来自甲烷泄漏，其在20年尺度上的温室效应约为二氧化碳的80倍；油气生产被估计占全球甲烷泄漏的三分之一，而美国是全球最大天然气生产国。与此同时，监管环境趋松：关键的甲烷监管要求被延后，削弱了对泄漏监测与修复的约束，可能放大新增需求带来的排放后果。
并非所有项目都会落地。全球能源监测组织追踪的项目中，正在建设、最可能完工的规模不足30吉瓦，另有159吉瓦处于前期规划与融资阶段。全球燃气轮机短缺亦是约束因素，全球范围内约三分之二的项目尚未确定设备供应商。尽管如此，即便部分建成也将显著推高排放与能源结构压力，凸显在AI与数据中心扩张不可逆的背景下，效率提升与甲烷减排的重要性。
New research shows data centers have driven a surge in US gas-fired power demand over the past two years. Global Energy Monitor found demand from gas projects explicitly tied to data centers reached 97 gigawatts in 2025, nearly 25 times higher than just over 4 gigawatts in 2024. Overall, data centers have nearly tripled US gas-power demand in two years. With about 565 gigawatts of gas capacity currently on the grid, building all projects in the pipeline would add nearly 252 gigawatts, expanding the gas fleet by almost 50%, equivalent to powering tens of millions of homes.
The environmental implications are substantial. While cleaner than coal, natural gas accounted for about 35% of US energy-related CO2 emissions in 2022. Greater concern comes from methane leaks, which are roughly 80 times more potent than CO2 over 20 years; oil and gas production is estimated to cause a third of global methane leaks, and the US is the world’s largest gas producer. At the same time, regulations are being rolled back, including delays to rules requiring operators to monitor and curb methane leaks, potentially amplifying emissions from new demand.
Not all proposed projects will be built. Of those tracked, projects under construction—most likely to be completed—total just under 30 gigawatts, while another 159 gigawatts remain in preconstruction. A global shortage of gas turbines further constrains timelines, with about two-thirds of worldwide projects lacking confirmed manufacturers. Even partial build-out, however, would materially increase emissions and strain energy systems, underscoring the need for efficiency gains and methane reductions as AI-driven demand persists.