The digital revolution also needs energy: who will power AI?

Data centers have increased energy demand, but some researchers are using research to propose alternatives

3 min
Room full of data servers.
À.G.
21/11/2026 - 09:07 h

BarcelonaThe new generation of AI needs more and more electricity and water, and data centers have increased energy demand. While some specialists warn of the risks involved, some researchers use research to propose more sustainable alternatives. “The need for data centers that require a large amount of resources has skyrocketed. But, if they are managed well and integrated well into the areas where the centers are located, the impact should be no different from any other type of facility. The problem is that it is being done so quickly and there is such a great need for data centers that not all the necessary criteria to reduce their impact are being applied,” explains Cristian Fàbrega Gallego, professor in the Department of Electronic and Biomedical Engineering who coordinates the master's degree in renewable energies and energy sustainability at the University of Barcelona.

According to the OECD and Cornell University, for every 100-word text or every conversation with between 10 and 40 prompts (instructions), up to 0.5 liters of fresh water are consumed, or the equivalent of fourteen LED bulbs left on for one hour. Apart from written queries, the creation of images and especially videos with artificial intelligence further increases water consumption. Each AI-generated image consumes between 2 and 5 liters of water, as revealed by a study from the University of California, Riverside.

Cooling consumption is very high because almost all the energy consumed by data centers is dissipated in the form of heat. “Artificial intelligence is not environmentally free, but it is not the monster that it is sometimes presented as either. Its impact will depend on the design of the data centers, where the energy it consumes comes from, or on whether we are capable of better utilizing the resources they use,” analyzes Fàbrega Gallego.

One cannot talk about AI without assessing its energy impact, but some researchers are using research to propose networks and applications that can operate with less energy. On the path to changing things and reducing the energy footprint of AI, two studies from the Open University of Catalonia (UOC), with the participation of researchers Fernando Sevilla Martínez and Laia Subirats Maté, propose alternatives towards a more sustainable, efficient, and affordable AI.

"Energy efficiency must become a central parameter in the design of AI. It is not just about making models faster or with better performance, but about making them sustainable, ethical, and accessible. Designing energy-efficient AI not only benefits the planet, but also allows for the deployment of AI in small devices such as robots and sensors, reducing the operating costs for companies and data centers and improving resilience in environments with limited connectivity or energy," points out Subirats Maté.

The first of the investigations demonstrates that it is possible to develop low-power and high-performance spiking neural networks using economical and accessible components. The second work analyzes in detail how spiking neural networks can reduce the energy consumption of autonomous driving systems.

The research group Software and Service Engineering Group (GESSI) of the Polytechnic University of Catalonia - BarcelonaTech (UPC) is developing GAISSA, a project that has developed a software tool. It is an energy card that evaluates energy efficiency in both the training and inference phases of any machine learning model and is capable of evaluating the footprint of machine learning models, and contributing to improving their efficiency, reducing operating costs, and decreasing computation times. Following the tool's refactoring recommendations, engineers will be able to improve energy efficiency by up to 50%, an aspect that will also allow for a reduction in the economic cost of the projects.

Aware of the environmental impact of this consumption, the Ministry for the Ecological Transition and the Demographic Challenge of Spain implemented an innovative legal framework to promote sustainability in this area. This framework establishes obligations for data centers to be more transparent regarding their energy consumption and operational efficiency.

An AI gigafactory

The Government ratified a few weeks ago the agreement of the Board of Directors of the Catalan Land Institute (INCASÒL) for the acquisition of 20 MW from the company Cetenis SL, an action that allows progress in the deployment of the artificial intelligence gigafactory planned in the economic activity sector El Molló, in the municipal areas of Móra la Nova and Tivissa (Ribera d’Ebre).

The gigafactory promoted by the State in consortium with the Catalan government and a group of private companies led by Telefónica, ACS, and Banco Santander is part of the strategic initiatives driven by Catalonia to position itself in the field of artificial intelligence and advanced digital infrastructures. The project was recognized by the Government as a strategic investment project for Catalonia.

"We are preparing the network so that more energy can reach it and more can also be generated through renewable sources," asserts Sílvia Paneque. The Minister of Territory met with the Secretary of State for Energy, Joan Groizard, to finalize the electrical planning in Catalonia within the state plan. The plan provides for creating new electrical substations and expanding the existing ones to provide service both to current and new consumers, but also to ensure the supply for new infrastructure and business projects that will require high levels of energy consumption.

stats