根据美国环保组织塞拉俱乐部（Sierra Club）的一项分析，到2030年，美国电力行业二氧化碳排放量可能增至约21亿吨，较2025年水平高出约42%。这一增幅相当于将美国电力行业自2010年代以来逐步实现的减排成果基本抹去。
推动这一变化的核心因素是AI基础设施建设热潮。随着大型语言模型训练、推理服务以及云端AI应用迅速普及，数据中心对电力的需求正以前所未有的速度增长。然而，清洁能源和电网扩容速度未能同步跟上这些需求。
分析指出，越来越多科技企业因为无法及时获得电网接入资格，开始绕开传统公共电网体系，自建发电设施以满足数据中心运行需求。这些被称为“表后发电”（Behind-the-Meter）的项目大多依赖天然气等化石燃料，其产生的排放量通常不会直接体现在公共电网排放统计之中。
报告预计，到2030年，由数据中心开发商建设或承包运营的离网发电设备，每年可能排放超过4.6亿吨二氧化碳。仅这一部分新增排放规模，就接近墨西哥全国一年的碳排放总量。需要指出的是，这一比较针对的是相关离网发电设施的排放，而非全部数据中心活动产生的排放。
造成企业转向自建发电的重要原因之一是电网接入流程缓慢。许多大型项目需要等待多年才能完成并网审批，而AI产业的发展节奏远快于传统能源基础设施建设周期。为了满足项目上线时间要求，开发商越来越倾向于部署天然气涡轮机等可快速建设的发电系统，并将其直接接入数据中心。
即便不考虑离网发电问题，连接公共电网的数据中心项目同样会对排放增长带来显著影响。分析认为，到2030年，这部分项目可能额外增加约1.54亿吨二氧化碳排放，相当于在基准水平上再提高约10%。换言之，无论数据中心选择离网还是并网供电，其能源消耗都将给美国减排目标带来明显压力。
业内人士指出，AI产业正在暴露美国能源基础设施的诸多瓶颈。随着越来越多超级计算集群和大型数据中心投入建设，如何在满足算力需求的同时保证清洁能源转型，将成为未来数年美国能源政策面临的关键课题。
分析同时指出，推动更多新能源项目落地、缩短电网接入等待时间，以及建立更严格的新增能源配套机制，被认为是缓解相关排放增长风险、避免AI产业扩张冲击气候目标的主要途径。
本文由 AI图鉴自动抓取并翻译，版权归原作者所有；如需查看原文请点下方链接。
According to an analysis by the Sierra Club of the United States Environmental Organization, CO2 emissions from the United States electricity sector could increase to about 2.1 billion tons by 2030, about 42 per cent above the level in 2025. This increase amounts to a fundamental elimination of the emission reductions that the United States electricity sector has been achieving since 2010.
At the heart of this change is the boom in AI infrastructure. With the rapid spread of large language model training, reasoning services and cloud-based AI applications, the demand for electricity in data centres is increasing at an unprecedented rate. However, the pace of clean energy and grid expansion has not kept pace with these needs.
The analysis indicated that a growing number of science and technology enterprises were beginning to bypass traditional public grid systems and build their own power generation facilities to meet the operational needs of data centres because they were unable to qualify for grid access in a timely manner. Most of these projects, known as Behind-the-Meter, rely on fossil fuels such as natural gas, and the emissions generated are not usually directly reflected in public grid emissions statistics.
The report projects that, by 2030, more than 460 million tons of CO2 could be released annually from off-grid power generation plants built or contracted by data centre developers. This part of the new emission scale alone is close to Mexico ' s total carbon emissions for a year. It should be noted that this comparison is made for emissions from off-grid power generation facilities rather than from all data centre activities.
One of the important reasons for the shift of enterprises to self-build power generation is slow grid access. Many large-scale projects have to wait for years to be completed and approved online, and the development of the AI industry has been faster than the conventional energy infrastructure construction cycle. In order to meet the project timing requirements, developers are increasingly inclined to deploy rapidly constructed power generation systems, such as gas turbines, with direct access to data centres.
Even without considering off-grid power generation, the data centre project, which connects the public grid, has a significant impact on emission growth. The analysis suggests that by 2030, this component of the project could add about 154 million additional tonnes of carbon dioxide emissions, equivalent to an additional 10 per cent increase in baseline levels. In other words, whether the data centre chooses to be off-grid or grid-connected, its energy consumption will put obvious pressure on the United States reduction target.
Industry has noted that the AI industry is exposing many bottlenecks in the United States energy infrastructure. As more supercalculations clusters and large data centres are built, how to meet computing needs while ensuring a clean energy transformation will be a key issue for United States energy policy in the coming years.
At the same time, the analysis noted that promoting more new energy projects to land, reducing waiting times for grid access and establishing more stringent new energy packages were considered to be the main ways to mitigate the associated risks of emission growth and avoid climate shocks from the expansion of AI industries.
Auto-fetched and translated by AI Atlas. All rights belong to the original author.