Environmental Regulation and AI: Energy Consumption, Reporting, and Sustainability

EU AI Act Environmental Provisions

The EU AI Act addresses AI's environmental impact through several provisions. Article 53(1)(a) requires providers of general-purpose AI (GPAI) models to draw up and keep up-to-date technical documentation including information on the energy consumption of the model's training and, where known, testing and fine-tuning, including the computational resources used. Annex IV specifies that technical documentation for high-risk AI systems must include information about the computational resources used and known or estimated energy consumption. Annex XI requires GPAI model technical documentation to include the computational resources used for training, known or estimated energy consumption, and, where applicable, the use of energy-efficient hardware. These provisions create disclosure obligations but do not set consumption limits.

Corporate Sustainability Reporting and AI

The Corporate Sustainability Reporting Directive (CSRD, Directive 2022/2464) requires large companies and listed SMEs to report sustainability information according to European Sustainability Reporting Standards (ESRS). ESRS E1 (Climate Change) requires disclosure of Scope 1, 2, and 3 greenhouse gas emissions. For companies developing or deploying AI at scale, energy consumed by AI training and inference (typically Scope 2 for owned infrastructure, Scope 3 for cloud compute) falls within the reporting scope. Companies must disclose their climate change mitigation targets and actions, which should include measures to reduce AI-related energy consumption. The first CSRD reports (for fiscal year 2024) began in 2025, with phased application to smaller companies through 2028.

Data Centre Energy Efficiency Requirements

AI training and inference rely on data centre infrastructure subject to the recast Energy Efficiency Directive (2023/1791) Article 12, which requires member states to collect and publish data centre energy consumption data. The Delegated Regulation on data centre sustainability (Commission Delegated Regulation 2024/1364) requires data centres with 500 kW or more of installed IT power to report Power Usage Effectiveness (PUE), Water Usage Effectiveness (WUE), the share of renewable energy, waste heat reuse, and temperature set points. AI workloads are among the most energy-intensive data centre applications: training a large language model can consume thousands of MWh. Organizations operating AI infrastructure must factor these reporting obligations into their operational planning.

Environmental Impact Assessment for AI

While no jurisdiction currently requires a standalone environmental impact assessment (EIA) for AI systems, the intersection of AI deployment with existing EIA requirements is growing. Large-scale data centre construction for AI training is subject to EIA under the EU Environmental Impact Assessment Directive (2011/92/EU) when thresholds are met. The EU Taxonomy Regulation (2020/852) provides criteria for environmentally sustainable economic activities; AI services must assess alignment with taxonomy criteria for climate change mitigation and adaptation. Some member states are considering extending environmental assessment requirements specifically to large-scale AI training operations.

Green AI: Efficiency Measures and Standards

Industry initiatives and emerging standards address AI energy efficiency. ISO/IEC 23053 (Framework for AI Systems Using Machine Learning) includes considerations for computational efficiency. The Green Software Foundation's Software Carbon Intensity (SCI) specification provides a methodology for measuring software carbon emissions applicable to AI systems. Practical efficiency measures include: model compression (distillation, pruning, quantization) reducing inference energy by 50-90%; efficient hardware selection (purpose-built AI accelerators over general-purpose GPUs); workload scheduling to maximize renewable energy utilization; and training optimization techniques (mixed-precision training, efficient architectures) that reduce training compute. These are not yet legally mandated but demonstrate reasonable environmental practices.

Practical Compliance Steps

Organizations developing or deploying AI should: (1) measure and record energy consumption for all AI training runs, including compute hours, hardware specifications, and energy source; (2) include AI-related energy consumption in CSRD Scope 2/3 emissions reporting; (3) assess data centre infrastructure against Energy Efficiency Directive reporting requirements; (4) evaluate AI workloads against EU Taxonomy criteria for sustainable activities; (5) implement model efficiency techniques to reduce energy consumption per inference; (6) document energy consumption in AI Act technical documentation for GPAI and high-risk systems; (7) set reduction targets for AI-related emissions as part of climate transition plans; and (8) select cloud providers that report PUE, renewable energy share, and carbon intensity of their AI infrastructure.

This article is for informational purposes only and does not constitute legal advice. Regulatory requirements change frequently — verify current rules with official sources. Built by Sawai Gyoseishoshi Office, Hiroshima, Japan.