← 論文一覧に戻る

企業GHGインベントリのためのAI推論排出・資源係数

AI Inference Emission and Resource Factors for Corporate GHG Inventories (原題)

Llopis, Guillermo

Zenodoデータセット2026-09-15#AI×ESGOrigin: Global経営インパクト: 調達リスク対象セクター: cross_sector
DOI: 10.5281/zenodo.22767475
原典: https://zenodo.org/records/22767475

🤖 gxceed AI 要約

日本語

SOMA AIが公開するAI推論の排出・資源係数データセットのv2版。組込ハードウェアと学習負荷の償却項を追加し、フランス・フィンランドの電力グリッド行、商用モデルクラス対応表、整合性チェッカーを新たに収録した。企業がAI利用をScope 3カテゴリ1として算定するための係数を、地域・モデルクラス別に提供する。

English

Version 2 of SOMA AI's open dataset of AI-inference emission and resource factors for corporate GHG inventories. It adds embodied-hardware and training-amortisation terms, France and Finland grid rows, a commercial model-class mapping table, and a consistency checker. The factors let companies quantify AI usage under Scope 3 Category 1 by region and model class.

Unofficial AI-generated summary based on the public title and abstract. Not an official translation.

📝 gxceed 編集解説 — Why this matters

日本のGX文脈において

SSBJ基準・有報でのScope 3開示が進む中、生成AI利用の排出を算定する係数は日本企業のインベントリ作成に直接使える。特にデータセンター立地別の係数は、クラウド利用が多い日本企業のカテゴリ1算定精度を高める。

In the global GX context

As ISSB/CSRD and SEC climate rules push Scope 3 Category 1 granularity, AI inference is an emerging blind spot. This dataset supplies region- and model-class-specific factors, supporting comparable disclosure of AI-related emissions across global value chains.

👥 読者別の含意

🔬研究者:Provides an open, versioned factor set for studying AI's lifecycle emissions and Scope 3 attribution methods.

🏢実務担当者:Enables sustainability teams to estimate AI-inference emissions in Scope 3 Category 1 using region- and model-class factors.

🏛政策担当者:Illustrates a disclosure-infrastructure gap: AI usage emissions lack standard factors, relevant to ISSB/SSBJ guidance development.

📄 Abstract（原文）

Version 2 (2026-09-15) of the SOMA AI-inference emission and resource factor dataset. Associated paper: Llopis, G. (2026), Accounting for AI Inference in Corporate GHG Inventories: A Four-Tier Methodology for Scope 3 Category 1 Reporting, arXiv:2606.10660, under review at Sustainable Production and Consumption. v2 adds embodied-hardware and training-amortisation terms (lifecycle_factors.csv), France and Finland, a commercial model-class mapping table (model_class_mapping.csv) and a consistency checker (tools/check_factors.py). The concept DOI 10.5281/zenodo.20443585 always resolves to the latest version. Changes since v1: - Carbon rows completed for us-west-2 (classes B and C), us-south-central (classes B and C) and us-south (all classes); the grid rows existed since v1, the factor rows did not. - New grid rows eu-west-3 (France, 0.0533 kg/kWh) and europe-north1 (Finland, 0.0813 kg/kWh), Ember Yearly Electricity Data 2023 accessed 2026-09-12, with carbon and water rows for all three classes. - Water rows for Oregon and US South (US-average WUE/EWIF, estimated), France and Finland (European estimate, estimated). Texas rows (Li et al. 2025 confirmed) were already present. - us-south-central provider examples corrected: AWS us-east-2 is Ohio (RFCW), not Texas; the row now cites Google us-south1 (Dallas) and Azure southcentralus. - tools/check_factors.py recomputes every H100-central carbon row and every water row from the grid and energy tables and checks the redundant columns; it is run before every deposit. - Two v1 rows corrected by the checker: us-east-1 class B carbon 0.04380 to 0.04378, and Sweden class C water total 1.254 to 1.253. - Product factor version: soma-ai-ef-2026.09-lifecycle-v2. Licence CC BY 4.0. Contact: [email protected].

🔗 Provenance — このレコードを発見したソース

🔔 こうした論文の新着を逃したくない方は キーワードアラート に登録（無料・3キーワードまで）。

gxceed は公開メタデータに基づく研究支援データセットです。要約・翻訳・解説は AI 支援で生成されています。 最終的な解釈・検証は利用者が原典資料に基づいて行うことを前提とします。