AI energy consumption and data centre rating under spotlight

The rapid growth of both AI and data centres in Europe, as elsewhere, is placing increasing demands on energy resources and raising concerns about their environmental footprints. 

In the case of artificial intelligence, the AI act, which was introduced in June 2024 setting out rules on the development and use of AI, contains provisions on energy consumption and transparency.  

Among these, providers of general purpose AI models are required to document the known or estimated energy consumption of their models as part of their technical documentation obligations. 

However, obtaining a comprehensive picture of the energy efficiency and carbon footprint of AI models and systems is considered to require collecting data across multiple layers of the AI lifecycle. 

These include the computational resources (e.g. FLOPs and precision, power draw, energy use, or load factors), electricity consumption (using software or hardware-based measurement approaches) and hardware details (e.g. models and configurations, location of computation, data centre overheads) along with any further relevant information (e.g. metadata, measurement methods, assumptions, or energy supply information). 

The objective of the survey, part of a broader study to measure and foster energy efficient and low emission AI in the EU, is to assess the appropriateness and practical applicability of a methodological approach with input from stakeholders who would be affected by the AI Act provisions. 

In particular, the focus is on the availability and accessibility of data required to assess the energy consumption of AI models and systems during both the development (training) and operational (inference) stages, as well as on the suitability of different AI performance indicators.  

The ultimate goal is to develop a robust and practical industry-informed framework for measuring AI energy consumption and efficiency. 

Data centre rating 

Data centres growth is of concern, not only for their energy consumption but also for other resources such as water.  

A regulation on a European database on data centres and a reporting scheme for the collection of information and KPIs to be used for the EU rating scheme was introduced by the European Commission in March 2024, marking a first step towards this scheme. 

Now the Commission has put forward a further regulation setting out a proposed common EU-wide rating scheme with energy and water efficiency classes based on power usage effectiveness (PUE) and water usage effectiveness (WUE) indices calculated according to a set down methodology. 

The PUE and WUE are defined respectively as the total energy consumption and total water input of the data centre as a ratio of the energy consumption of the IT equipment. 

An electronic label for automatic issuance from the European database on data centres also is proposed, with pertinent indicators including a colour coded rating, similar to appliance energy efficiency rating labels. 

The aim of the rating scheme is to increase transparency about the energy use of data centres and to become a resource that will inform better policy making and procurement of more sustainable digital assets and services across Europe.  

The scheme also is expected to allow comparison between data centres and to promote new designs or appropriate efficiency interventions in new or existing data centres, including the use of renewable energy and re-use of waste heat in nearby facilities and heat networks. 

Feedback on the data centre rating consultation is due by 23 April, with adoption scheduled for Q2 of 2026. 

Responses to the survey consultation on the AI energy consumption and emissions proposals is due by 10 May.