Published in La Tribune, on January 29, 2026
Where will data center hosts and Artificial Intelligence (AI) service providers find the enormous amounts of electricity needed for their regular operations? Several solutions are possible, some of which are already in use: connection to the local grid, off-grid production, or even orbiting. All of these solutions could end up being adopted at the expense of part of the current and future population. Policymakers will therefore have to manage difficult trade-offs between the needs of AI development and the well-being of the population.
Local impact of a grid connection
The International Energy Agency (IEA) estimates that data centers (excluding cryptocurrency mining) currently consume 1.5% of global electricity. This figure may seem insignificant, but it has doubled in 15 years and, with the development of AI, forecasts indicate an increase of more than 10% per year. According to the IEA, an AI-focused data center (DC) consumes as much electricity as 100,000 households, and the largest ones currently under construction will consume 20 times more. Connection to the electricity grid therefore causes a demand shock that requires adaptation of production capacity and the transmission network (see Appendix A of the ENTSOE position paper). This sharp increase in demand creates risks of energy rationing in the short and medium term and requires substantial investment to reduce these risks in the long term. As a result, consumers will face higher electricity prices if current tariff structures and service access rules are not changed.
These effects are already sparking controversy and sector-specific policy decisions aimed at protecting traditional consumers. In the US, for example, the protocol adopted by the Pennsylvania-New Jersey-Maryland (PJM) system operator to accommodate DCs has provoked protests from elected officials who consider the resulting increase in electricity prices for households in the region to be unjustified. In Ohio, the regional regulator has decided to create a specific category of customers requiring DCs to pay 85% of their expected energy consumption in advance in order to finance the necessary upgrades to the electricity grid. In the same vein, in France, in order to meet connection demands in the Marseille region (the world's fifth largest data transit hub), candidate DCs must contribute to the financing of the necessary work (550MW) at a rate of 92.5k€/MW. In Belgium, the grid operator, Elia, is considering capping the electricity consumption of DCs to avoid crowding out other industrial users following the sharp increase in demand from DCs. In Ireland, new DCs will have to have generators or batteries, and the energy produced on their sites will have to be injectable into the electricity grid via the wholesale market. Similar decisions are under discussion in many countries (see, for example, HERE and HERE), including emerging countries. In most cases, the debates also focus on the choice of electricity supply technologies.
The temptation of off-grid supply
Having a reliable energy supply that is independent of market fluctuations, grid congestion, and political tensions related to demand shocks is a solution that appeals to some DC operators. We have already discussed the use of nuclear power in a previous post. Two other technologies are currently being researched: geothermal energy and gas turbines.
Geothermal energy is renewable, non-intermittent, and has a low carbon footprint. In particular, geothermal energy could reduce the cooling costs of DCs, which account for 40% of their energy consumption. However, accessing underground thermal energy requires drilling, the private cost of which depends on the nature of the rock and the depth, and the public cost of which is often underestimated: microseismical events, groundwater pollution, thermal consequences of air or water cooling. Big Tech companies do not have the expertise to undertake this work, but they can sign supply contracts with specialized companies. For example, Star Energy Geothermal, a subsidiary of the Indonesian company Barito Renewables Energy, is in talks to supply energy to data centers if developers build them near its geothermal power plants. Similarly, XGS Energy has signed an agreement with Meta (Facebook, Instagram, WhatsApp) to develop 150 MW of next-generation geothermal energy in New Mexico for the sustainable development of AI.
Turbines are another possible solution. The turbine technology mastered by Boom Supersonic, an American aerospace company developing the Overture supersonic airliner, has attracted the attention of Crusoe, a provider of AI platforms. Crusoe has ordered 29 Superpower gas turbines (totaling 1.22 GW) derived from the core of the Symphony supersonic engine developed by Boom Supersonic. The main advantages of these turbines are that they require no cooling and can operate at ambient temperatures of up to 43°C. A turbogenerator can be installed in two weeks once the foundations have been laid and takes up no more space than a standard shipping container. The main drawback is, of course, the same as with all technologies that burn fossil fuels to generate electricity: emissions of fine particles and carbon dioxide.
And in outer space?
Since terrestrial resources are scarce and their use causes pollution, why not install DCs in space, where solar energy is abundant and cooling is free (on the side not exposed to the sun, of course)? ASCEND (Advanced Space Cloud for European Net Zero Emission and Data Sovereignty) is a feasibility study carried out for the European Commission by a consortium led by Thales Alenia Space. Its objective is to define the development program and associated technology roadmap for a system that will be operational in 2030. Various competitors have entered the race to place DCs in orbit around the Earth, and even on the Moon (see, for example, HERE and HERE). It is therefore likely that technical problems (construction, launch, operation) will quickly find financially acceptable solutions. But the idea that space can be occupied free of charge without negative externalities is as mistaken as the belief held by our ancestors that we could draw on the Earth's subsoil without causing damage. It is easy to imagine future generations fighting to have it recognized that the environment to be protected does not end with the Earth's atmosphere.
Choosing from a menu of options
To meet the energy needs of data centers, it will not be enough to take into account local or spatial technical constraints. The economic and social problems associated with consumption by other industries and households, combined with the effects on the environment, will also be at the heart of debates and controversies that are only beginning. To resolve this multidimensional problem efficiently and fairly, policymakers in the countries concerned must be offered a menu of solutions from which they can choose the one best suited to local socio-economic and technical conditions, and the choices will not be easy.
