Energy Data for AI Models

The development and deployment of artificial intelligence (AI) depend on access to large volumes of high-quality data,[1] as well as on addressing the key barriers to data access and use while ensuring appropriate safeguards, for example in terms of privacy and cybersecurity.[2] As the EU seeks to accelerate the digitalisation of its energy system, understanding the data requirements of AI models and the legal framework governing access to and use of energy data becomes increasingly important.

In this instalment, we explore the type of energy data required to train AI models and provide some reflections on the applicable EU law framework for energy data access and use. First, we discuss the matter of identifying the main categories of data that can support the training and operation of such models, highlighting the diversity of data sources available in the energy sector. Secondly, we emphasise some key considerations regarding the EU legislative framework governing access to and exchange of energy data, with particular attention to the interplay between energy-specific and horizontal data legislation. Thirdly, we discuss if and how, based on this legislative framework, relevant data can lawfully be used for the purposes of training AI models in the energy sector. Finally, we briefly examine how the Commission’s recent proposal to amend the Electricity Regulation seeks to address some of these challenges through a dedicated framework for electricity grid data exchange.

Types of data relevant for AI models in the energy sector

As mentioned above, the development of AI models for the energy sector relies on the availability of large volumes of diverse and high-quality data. It is therefore relevant to explore which data types might be used for AI in energy. While the precise data needs depend on the specific AI use case and application, it is possible to identify a number of broad categories of data that are likely to play an important role.

Models designed for energy sector applications will inevitably depend on energy and energy-related data. Such data can be broadly grouped into the following categories: network data (e.g., structural data of assets), market data (e.g., bids and offers on wholesale markets for energy) and consumer data (e.g., metering and consumption data). Although energy data can be classified in different ways depending on the purpose of the analysis, this categorisation provides a useful framework for understanding the main sources of information available within the energy system.

While these categories represent an important source of data for current AI applications in the energy sector, they should not be regarded as exhaustive. AI models may also rely on complementary data sources outside the energy sector. In particular, external data, such as weather and climate data, can play an important role in tasks such as demand and renewable generation forecasting as well as grid operation.

EU legislative framework to access data for AI models in the energy sector

Once the relevant data types have been identified, a second step is securing access to these data by law. There are few provisions under EU law that explicitly provide for access to data needed to develop energy‑specific AI models. In EU data law, the 2023 Data Act is notable since it provides for access to data generated by connected products to the users of these devices. Other pieces of EU data legislation, including the Free Flow of Non-Personal Data Regulation, the Open Data Directive, and the Data Governance Act (DGA), provide some horizontal supporting measures facilitating data access, for example via data altruism organisations. Taken together, access to energy data is rendered easier by horizontal data law, but it is by no means comprehensive in the context of AI model development. For instance, the product and user-centric approach of the Data Act may not cover all relevant data across the data types identified above, nor does it necessarily guarantee access to these data to actors involved in the development of AI models.[3]

A similar observation can be made with regard to the AI Act. Although the Regulation recognises in its recitals that initiatives such as the Common European Data Spaces can improve data availability and quality for the training, testing, and validation of AI systems, it does not create new rights to access energy data and it does not establish specific mechanisms through which researchers, market participants, or public authorities could obtain such data for the purpose of developing AI models, being mainly focused on establishing requirements for the development, placing on the market and use of AI systems. [4]

In the same way, EU energy law contains only a limited number of provisions that expressly enable the access to energy data for the development and training of energy-specific AI models. Certain sectoral instruments contribute to making relevant datasets available. These include, among others, the Transparency Platform established under Commission Regulation (EU) No 543/2013, the Regulation on Wholesale Energy Market Integrity and Transparency (REMIT), as well as databases and information-sharing mechanisms created under the Renewable Energy Directive (RED III) and the Energy Performance of Buildings Directive (EPBD).[5] Together, these initiatives provide valuable and regularly updated information that could support the training of AI models in the energy domain. However, access to such data remains fragmented. Relevant datasets are distributed across multiple platforms, repositories, and governance frameworks, making their discovery and use more complex. Moreover, some categories of data that could be particularly valuable for AI development, such as detailed metering information, remain largely outside dedicated regimes for this secondary use (or ‘reuse’).[6]

Lawful secondary use of energy data for AI models

Merely having access to data may not be sufficient in order to lawfully use these same data for developing AI models for the energy system. Such data will be often generated and collected for purposes other than training AI models. Therefore, the third step examined in this blog is ensuring that secondary use of energy data complies with applicable law.[7] In EU law, the precise conditions attached to secondary use largely hinge on the characteristics of the data, and especially on whether the data are considered personal or non-personal. In the former case, secondary use must comply with the GDPR. Specifically, using personal energy data, such as household metering data, for the development of AI models must be based on one of the lawful grounds for processing data. In this context, it should be noted that the Commission’s Digital Omnibus Regulation proposal would introduce a new provision (Article 88c) in the GDPR that explicitly recognises the processing of personal data for the development and operation of an AI system as falling under the ‘legitimate interest’ ground for the processing of personal data (as per Article 6(1)(f) GDPR), preserving, however, the principles and safeguards of data protection and providing data subjects with an unconditional right to object to such processing. Non-personal data may also be subject to restrictions on secondary use, for instance due to contractual, intellectual property or trade secrecy reasons. The Data Act, for example, restricts the use of non-personal, readily available data from connected products by a data holder to what is stipulated in the contract with the user of that product (Article 4(13)).

A new framework for the exchange and secondary use of electricity grid data

The development of AI models for the energy sector depends on the identification, availability, accessibility and lawful secondary use of high-quality data. As this instalment has shown, the current EU legal framework provides limited pathways for accessing and reusing relevant energy and energy-related data: fragmentation and legal uncertainty remain. In this regard, the Commission’s Strategic Roadmap for Digitalisation and AI in the Energy Sector and, more specifically, Pillar III-Flagship Action 5, points towards a more coordinated approach, notably through efforts to streamline energy-specific data exchange, facilitate the pooling of energy data for AI training and research, and establish trusted frameworks for the use of AI in energy.

In this context, the legislative proposal for the amendment of the Electricity Market Regulation, which was published by the Commission on 17 July 2026, aims to establish an EU law basis for Flagship Action 5 of the Strategic Roadmap. Regarding the issue of energy data for AI models, it is possible to identify how the Commission has chosen to address the three steps mentioned above. First, with regard to data type, the proposal focuses on electricity grid data. Secondly, regarding data access and exchange, the Commission proposes the establishment of a ‘voluntary secure electricity grid data exchange framework’, coordinated by ENTSO-E and the EU DSO entity, to enable the reuse of such data (proposed Article 18a(5)). Thirdly, regarding the lawfulness of such secondary use, the proposal specifies that the sharing of energy data that are personal data should be avoided if possible. It additionally mandates ENTSO-E and the EU DSO entity to draw up technical and operational measures for compliance of the voluntary data sharing arrangement with relevant EU legislation, and in particular the Data Act, the GDPR and the AI Act. Finally, the Commission would be empowered to adopt implementing acts containing more detailed requirements for the lawful, secure and controlled reuse of data (proposed Article 61(5b)).

Acknowledgements

The Florence School of Regulation gratefully acknowledges the financial support of the European Commission (DG ENER) for conducting the research that led to this blogpost. Views expressed in this blogpost reflect the opinion of individual author(s) and do not necessarily reflect the views of the European Commission.

[1] For instance, according to art 10(3) of the AI Act: ‘Training, validation and testing data sets shall be relevant, sufficiently representative, and to the best extent possible, free of errors and complete in view of the intended purpose. They shall have the appropriate statistical properties, including, where applicable, as regards the persons or groups of persons in relation to whom the high-risk AI system is intended to be used. Those characteristics of the data sets may be met at the level of individual data sets or at the level of a combination thereof’.

[2]  The respondents to the Open Public Consultation to the Strategic Roadmap for Digitalisation and AI in the Energy Sector considered the lack of data interoperability and insufficient use of commonly agreed data exchange standards, the absence of an established data governance framework, and the lack of consistent data access regulatory frameworks to be the first, second and third barrier to the establishment of a Common European Energy Data Space to enable demand-side flexibility and smart energy services. Anna Spoden, prepared for the European Commission, ‘Task 5 Report: Analysis of the Open Public Consultation to the Strategic Roadmap for Digitalisation and Artificial Intelligence (AI) in the Energy Sector’ ENTEC2 Study 2: AI and Digitalisation (2025), p. 22.

[3] There is some uncertainty, for example, regarding the qualification of smart electricity meters as connected products. See Sofia Nicolai and Max Münchmeyer, ‘The EU Data Act and Electricity Consumer Participation in Demand Response and Flexibility Services’ 44 Journal of Energy & Natural Resources Law 41 <https://doi.org/10.1080/02646811.2025.2485725>. A further question in this context is that of who qualifies as a ‘user’ of a connected product, with recent Finnish national guidance specifying that end consumers of electricity do not fall under this category. See TRAFICOM, ‘Oikeus Sähkön Mittauslaitteiston Keräämään Dataan’ (2026) <https://www.traficom.fi/files/media/file/S%C3%A4hk%C3%B6n%20mittauslaitteistot%20ja%20data-asetus_0.pdf>.

[4] As clarified in the second instalment of the Topic of the month ‘foundation models are generally described as models trained on large and diverse datasets that can be adapted to a wide range of downstream tasks. Generative AI typically refers to AI systems capable of generating new content, such as text, images, audio, video, data or code. These definitions suggest that the two terms emphasise different aspects of AI systems. The term foundation models highlights the model’s role as a common basis from which many more specific applications can be developed. Generative AI refers to the system’s primary function; its capability to create new content. This distinction in emphasis is important, but it also makes the relationship between the two categories less straightforward’. This description draws on the definitions of foundation models provided by EESC, ENTEC2, ETIP SNET, Stanford University and IEA.

[5] For example, see art 12(2) REMIT, art 16 EPBD, art 20a(3) of the RED III.

[6] Secondary use of data refers to the processing of data for purposes other than the initial ones for which they were collected or produced, for example, research and innovation purposes and AI models development (see also art 2(2)(e) of the European Health Data Space Regulation).

[7] See, for example, the considerations as to the lawfulness of secondary use of health data in the impact assessment of the European Health Data Space Regulation Commission, ‘Impact Assessment Report Accompanying the Document Proposal for a Regulation of the European Parliament and of the Council on the European Health Data Space’ SWD(2022) 131 final 12–18.