While France is increasing its announcements on digital sovereignty and artificial intelligence, a blind spot persists. AI is still largely thought of as an innovation policy, sometimes as an industrial policy, rarely as what it has become, namely an territorial infrastructure policy. Electricity, water, land, administrative delays, local acceptability, these parameters, long secondary, now condition the real capacity to deploy AI on scale. This shift is beginning to manifest itself concretely on the ground.
AI comes out of the cloud and into the real world
The illusion of dematerialized AI has come to an end. Foundation models, generative AI and agentic systems rely on heavy, localized, resource-intensive infrastructures. A data center is an energy, hydraulic and land asset, located in a given territory.
“In Eybens, near Grenoble, a data center project must ultimately achieve a power of 200 megawattsa level comparable to the electricity consumption of a medium-sized city. » (France Info / France 3 Alpes, January 2026)
This order of magnitude is enough to tip the debate. AI is no longer just a matter of software or calculation and is becoming a question of networks, local capacities and collective arbitrations.
The territory as a new strategic constraint
Until now, the French discussion has been structured around two pillars: capital (subsidies, public funds, investment announcements) and computing (supercomputers, trusted cloud, national capacities), where the territory remained an implicit support.
It now becomes a limiting factor. Not out of hostility to technology, but through progressive saturation of resources: tension on electrical networks, availability of water, acceptability of settlements, permit deadlines. As these constraints accumulate, the question is no longer “do we need AI?” “, but under what material and territorial conditions can it be deployed.
The French paradox: planning without territorializing
France has structural assets with a relatively carbon-free electricity mix, recognized engineering capacity, and a State capable of directing investments. But it retains a persistent blind spot with the territorial translation of its digital ambitions.
Data center projects are often approached from the angle of attractiveness or sovereignty, rarely from that of the induced local load. The consultation comes late and the public narrative remains unclear on the real costs. Communities find themselves at the end of the chain, required to absorb critical infrastructure that they do not control. This discrepancy creates a silent tension where the State accelerates and the territories suffer.
The environmental footprint, now central
Local concerns are not just subjective perceptions. They are based on now established data.
” In France, 46% of the digital carbon footprint comes from data centersaccording to a study by ADEME and the Electronic Communications Regulatory Authority. » (Study cited by France Info)
This figure causes a major shift in the debate. The environmental impact of digital technology is no longer primarily in uses, but in the infrastructure itself. AI, as an accelerator of calculation and capacity, mechanically amplifies this reality.
The American mirror, without copying it
In the United States, Microsoft has just implemented this phase change by making the expansion of its data centers conditional on explicit coverage of energy and water costs, in order to prevent them from being passed on to local consumers. The message is clear: without territorial acceptability, AI does not scale.
This is not an ideological gesture, but a strategic decision. AI infrastructure has become too visible, too expensive and too critical to remain negotiated in secrecy. France is not there yet, but the weak signals are accumulating. Anne Le Hénanff was also questioned on this subject during her parliamentary hearing.
Who pays the real cost of AI?
The central question is no longer technological. It is political and economic: Who should bear the systemic cost of AI? Companies, by internalizing energy, water and infrastructure? Communities, in the name of attractiveness? The State, via subsidies and national planning?
With the announcement of several dozen new data centers coming to the region, this question is becoming structuring and conditions the very credibility of any medium-term AI strategy.
AI as critical infrastructure
As AI becomes a pillar of competitiveness, economic sovereignty and essential services, it gets closer to traditional critical infrastructures: energy, transport, telecoms, and implies the same requirements: continuity, transparency, acceptability, territorial equity. Digital sovereignty can only be built through constant dialogue with the territories.
