同一AI请求在德州产生的碳排放是瑞典的8.5倍

瑞典AI服务商Berget AI与斯德哥尔摩环境研究所等机构联合发布了开源方法，用于计算每项AI请求的碳排放。1这一方法的核心发现表明，地区电网的能源结构对AI应用的环保影响存在显著差异。1

根据这套方法的测算，运行相同的AI模型请求在不同地区产生的排放量差异巨大。1以Kimi K3模型的典型请求为例，在瑞典运行时产生约0.4克CO₂，而同样的请求在德州产生3.2克，在德国产生2.5克，在爱尔兰产生1.6克。1德州的碳排放量达到瑞典的8.5倍。1

该方法已在GitHub上以MIT许可证开源发布，遵循绿色软件基金会的AI碳排放强度指标规范。1Berget AI自2025年以来已开始报告其每项API请求的CO₂排放量。1这一举措旨在促进AI服务商透明公布排放数据，帮助用户选择碳足迹更低的服务选项。1

国际能源署预计数据中心的电力消耗将在2030年前翻倍至约945太瓦时，其中AI是主要的增长驱动力。1同时，开源模型处理的AI流量占比也在上升，从2026年4月的11%增至6月的29%。1

Berget AI, in collaboration with the Stockholm Environment Institute and other organizations, has released an open-source methodology for calculating the carbon footprint of individual AI requests, revealing significant regional disparities based on electricity grid composition 1. A typical request using the Kimi K3 model produces approximately 0.4 grams of CO₂ in Sweden, compared to 1.6 grams in Ireland, 2.5 grams in Germany, and 3.2 grams in Texas—making Texas emissions 8.5 times higher than Sweden 1.

The initiative addresses growing environmental concerns as the International Energy Agency projects data center electricity consumption will double to roughly 945 terawatts-hours by 2030, with artificial intelligence serving as a primary driver of this increase 1. Berget AI has been reporting CO₂ emissions for each API request since 2025 1. The open-source method, released under an MIT license on GitHub and following the Green Software Foundation's SCI for AI specification, aims to enable AI providers to transparently disclose emissions data and allow users to select lower-carbon alternatives 1.

The methodology reflects a broader shift in the AI industry toward sustainability awareness, as open-source model processing has grown from 11 percent of AI traffic in April 2026 to 29 percent by June 2026 1.