GPAI Energy Consumption and Environmental Reporting Requirements (2026) | MmowW

Why Energy Reporting Matters for GPAI

Training large-scale AI models requires substantial computational resources, and the associated energy consumption has become a significant regulatory concern. The EU AI Act addresses this by requiring GPAI model providers to measure and report energy consumption as part of their technical documentation. This obligation aligns with the EU's broader sustainability commitments under the European Green Deal and the Climate Law.

Article 53(1)(d) establishes the baseline requirement for all GPAI providers to include energy consumption information in their technical documentation. Annex XI Section 2(f) specifies that this must include the known or estimated energy consumption of the model during training, and where applicable, during inference.

Baseline Energy Reporting: All GPAI Providers

What Metrics Must Be Reported

Under Annex XI, all GPAI model providers must document the following energy-related metrics:

• Total energy consumed during model training, expressed in kilowatt-hours (kWh) or equivalent units
• Computational resources used, including hardware type, quantity, and training duration
• Total compute measured in floating point operations (FLOPs)
• Known or estimated energy consumption during inference, where this information is available
• Carbon emissions associated with training, where measured or reasonably estimable

Where exact measurements are not available, providers must provide reasonable estimates based on the hardware used, training duration, and data centre energy profiles. The methodology used for estimation should be documented alongside the reported figures.

Measurement Standards and Methodology

The EU AI Act does not prescribe a single measurement standard for energy reporting. However, the GPAI Code of Practice, developed under Article 56, is expected to provide guidance on acceptable measurement methodologies. Providers are encouraged to use established frameworks such as the ML CO2 Impact tool, the IEA data centre energy methodology, or equivalent approaches.

Key considerations for measurement include:

• Distinguishing between the energy consumed by the training process itself and the overhead energy of data centre operations (cooling, networking, storage)
• Accounting for multi-tenant environments where hardware is shared across multiple workloads
• Documenting the energy mix of the data centres used, including the proportion of renewable energy
• Reporting on the geographic location of training infrastructure, as energy carbon intensity varies by region

Additional Obligations for Systemic Risk GPAI Models

GPAI models classified as presenting systemic risk under Article 51 face enhanced obligations. Article 55 requires providers of these models to conduct and document model evaluations that include assessment of systemic risks, which may encompass environmental risks associated with the model's energy footprint at scale.

For systemic risk models, providers should document:

• The environmental impact of training, including total carbon footprint with scope 2 emissions (from purchased electricity) and scope 3 emissions (from the hardware supply chain) where available
• Energy efficiency measures implemented during training, such as mixed-precision training, model distillation, or efficient architecture choices
• Projected energy consumption of the model during deployment at scale
• Any measures taken to mitigate environmental impact, including use of renewable energy or carbon offsets

Alignment with EU Sustainability Goals

The GPAI energy reporting requirements exist within the broader context of EU sustainability regulation. The European Green Deal commits the EU to climate neutrality by 2050, and the energy consumption of AI systems has been identified as a growing area of concern in the European Commission's communications on AI and sustainability.

The Corporate Sustainability Reporting Directive (CSRD) may also require large companies to report on the environmental impact of their AI operations as part of their sustainability disclosures. While the CSRD and AI Act operate independently, organisations subject to both should consider aligning their reporting approaches to avoid duplication and ensure consistency.

Practical Steps for Compliance

Organisations should take the following steps to prepare for energy reporting compliance:

Looking Ahead

Energy reporting for AI systems is an evolving area of regulation. The AI Office is expected to provide further guidance on measurement standards and reporting formats as part of the GPAI Code of Practice. Providers who establish strong measurement practices now will be better positioned to adapt to future requirements.

The European Commission has also signalled interest in developing more detailed environmental standards for AI systems as part of future regulatory iterations. Early investment in energy measurement and efficiency will serve as both a compliance measure and a competitive advantage.

Start your AI compliance journey to track energy reporting requirements alongside your full GPAI compliance obligations.

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.