随着人工智能产业进入大规模基础设施建设阶段，其带来的能源需求正引发越来越多的环境争议。最新研究警告称，美国计划建设的AI数据中心可能在未来几年显著推高电力行业碳排放，甚至抵消过去15年来通过能源转型取得的大部分减排成果。
根据彭博社获得的一份报告，环保组织塞拉俱乐部（Sierra Club）预计，美国已规划的AI数据中心将在2025年至2030年间额外产生约6.18亿吨二氧化碳排放，相当于使电力行业相关排放增加42%。如果预测成真，美国电力行业通过减少煤电使用、提升能源效率以及扩大可再生能源布局所取得的15年温室气体减排成果，可能被基本抵消。
报告指出，到2030年，美国电力行业总排放量可能因此升至21亿吨二氧化碳。这一最新预测值比该组织一年半前的估算高出2至4倍，反映出AI基础设施扩张速度远超此前预期。
面对传统电网接入能力不足的问题，许多数据中心运营商正尝试建设独立供电设施，以避免漫长的并网等待时间。虽然这种做法能够减轻居民电网压力，并降低对地区电价的影响，但这些离网能源系统可能额外带来超过4.6亿吨碳排放，其规模大致相当于墨西哥一年的全国碳排放量。
不过，各机构对于未来排放走势的判断仍存在巨大差异。目前最高预测与最低预测之间的差距，甚至超过加利福尼亚州和佛罗里达州电力需求总和。部分研究机构认为，可再生能源的大规模部署将显著缓解问题。例如罗迪姆集团（Rhodium Group）预测，即便在最悲观的排放情景下，美国电力行业2040年的排放水平仍将比2025年低24%。
此外，外界也无法确定当前宣布的所有AI数据中心项目最终都会落地建设。气候研究机构Sightline Climate此前估计，原定于今年启用的美国数据中心项目中，约三分之一至二分之一可能面临延期或取消。
能源供应问题已经开始影响AI产业发展。由于电力基础设施建设跟不上需求增长，部分数据中心GPU部署速度放缓。一些企业购买的高性能计算设备甚至不得不处于闲置状态，等待能够提供足够电力的数据中心投入运行。
报告还提到，亚马逊计划在得克萨斯州建设的数据中心项目将获得一座装机容量7.65吉瓦的天然气发电厂供电。根据获批文件，该电厂每年最高可排放3300万吨二氧化碳，如果达到这一水平，其可能成为美国最大的单一碳排放源之一。
近年来，数据中心已经因用电需求激增、耗水量庞大以及对社区基础设施造成压力而受到越来越多关注。随着AI模型规模不断扩大，未来如何在保障算力增长与实现减排目标之间取得平衡，正成为科技行业、能源企业和监管机构共同面临的重要挑战。
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。
As the artificially intelligent industry enters the stage of large-scale infrastructure development, the resulting energy demand is causing increasing environmental controversy. Recent research warns that the planned AI data centre in the United States may significantly boost carbon emissions in the electricity sector in the coming years, even offsetting most of the reductions achieved over the past 15 years through energy transformation.
According to a report received by Bloomberg, Sierra Club, an environmental organization, predicts that the planned AI data centre in the United States will generate an additional 618 million tons of carbon dioxide emissions between 2025 and 2030, equivalent to a 42 per cent increase in electricity-related emissions. If projections materialize, the 15-year GHG emission reductions achieved by the United States electricity sector through reduced coal power use, improved energy efficiency and an expanded renewable energy landscape may be largely offset.
The report notes that total emissions from the United States electricity sector may rise to 2.1 billion tons of carbon dioxide by 2030. This latest projection is two to four times higher than the organization ' s estimate a year and a half ago, reflecting the fact that AI infrastructure has expanded at a much faster rate than previously anticipated.
Faced with the lack of access to the traditional grid, many data centre operators are trying to build independent power facilities to avoid long network waiting times. While such an approach could reduce the population ' s power grid pressure and the impact on regional electricity prices, these off-grid energy systems could generate additional carbon emissions of more than 460 million tons, roughly the size of Mexico ' s national carbon emissions for one year.
However, there are still significant differences in the way agencies judge future emission trends. The current gap between the highest and the lowest projections exceeds even the aggregate electricity demand in California and Florida. Some research institutions believe that large-scale deployment of renewable energy sources will significantly mitigate the problem. For example, the Rhodium Group predicts that, even under the most pessimistic emission scenarios, emissions in the United States electricity sector in 2040 would be 24 per cent lower than in 2025.
In addition, it is not possible to ascertain that all currently announced AI data centre projects will eventually land. Sightline Climate, a climate research institution, had previously estimated that about one third to one half of the US Data Centre projects scheduled to be launched this year might face delays or cancellations.
Energy supply issues have begun to affect the development of AI industry. The deployment of GPU in some data centres slowed as the construction of electricity infrastructure did not keep pace with the growth in demand. Some enterprises have even had to be idle in the form of high performance computing equipment, pending the operationalization of data centres capable of providing sufficient electricity.
The report also mentions that the Amazon project to build a data centre in Texas will provide power to a gas power plant with installed capacity of 7.65 GW. According to the approved documentation, the plant could emit up to 33 million tonnes of CO2 per year, and if this level is reached, it could be one of the largest single carbon emissions sources in the United States.
Data centres have received increasing attention in recent years as a result of surges in demand for electricity, high water consumption and pressure on community infrastructure. As the AI model expands in size, the future balance between guaranteeing growth in computing power and achieving emission reduction targets is becoming an important challenge for science and technology industries, energy enterprises and regulators alike.
Auto-fetched and translated by AI Atlas. All rights belong to the original author.