DATA CENTERS AND AI, FACTS, FIGURES, AND QUESTIONS

The exercise I undertook below, and which I am sharing with you, consisted of trying to understand the various aspects of the data center issue, which is increasingly criticized, primarily in the United States today. I address the impacts in terms of resources, the environment, the economy, finance, and society (the question of employment, etc.), concluding with the need for debate and political choices to control their development. There is a rationale, at least strategic, for hosting data centers, as we will see. In the meantime, it is crucial to understand who will benefit from the massive investments we will be discussing and the repercussions at the national and local levels in Europe, given that these data centers continue to be used and designed primarily for American tech giants.

France has just signed a new mega-investment agreement for data centers on its soil, totaling €45 billion by 2031 (with a total target of €75 billion), with the Japanese tech leader, SoftBank, confirming its position as one of the biggest European host countries and investor in the sector. The main reason is the appeal of readily available, consistent, carbon-free, and relatively inexpensive energy (nuclear).

Reactions to this announcement, made at the Choose France summit on June 1st, have so far been rather timid, or even non-existent in the mainstream media.

Conversely, in the United States, the controversy surrounding these data centers is not new and tends to intensify over time, with a large majority of Americans now opposed to the continuation of these massive investments. In this context, the US government, according to a May 27 article in Le Grand Continent, "fears the emergence of 'anti-technology' extremism."

AI is a huge accelerator of these investments worldwide. Beyond the issues I will discuss below regarding the direct environmental impacts, energy needs, and real economic benefits (beyond those of the tech giants), we risk experiencing a genuine bubble in terms of overinvestment in capabilities. Simultaneously, we risk a financial bubble similar to the one we saw in 1998-2000 during the dot-com bubble.

The announcement of Softbank's investment caught my attention and made me realize that the subject is increasingly present in more specialized media, but only discussed sporadically in our circles and therefore deserved to be delved into in order to better analyze it and understand its implications.

As I began to tackle it, I quickly realized how extremely complex the subject is, like many subjects relating to new technologies and their diverse and varied impacts on the future of our societies and our lives.

In the end, what emerges from what I have been able to find, read and synthesize below is that our societies are facing necessarily political choices regarding the development of these data centers, choices which will also extend to the possible regulation of the end use of our iPhones, tablets and computers, whether at the professional or private level.

The question of who will ultimately benefit from these data centers in the long term remains to be seen. It is clear that American mega-players and other tech giants like SoftBank want to "preempt" French and European electricity capacity and will capture the lion's share of the added value. Data centers installed in Europe are ultimately just electricity distribution hubs serving US generative AI giants, whose profits will not stay in our country. Will the positive local economic effects therefore be limited to those of a few electricity companies and security firms?

To make this synthesis, I relied on multiple sources, including Les Échos, La Tribune, The Atlantic, Le Grand Continent, Le Monde, Le Monde Diplomatique, the website of the water distribution giant, Veolia, the websites of ADEME and RTE, and I will cite these sources throughout the article.

FACTS AND FIGURES

As we know, behind the apparent immateriality of digital technology lie energy-intensive infrastructures: in electricity, but also in water for cooling.

Global demand for data center capacity : According to the Veolia group website, demand for data center capacity is expected to triple by 2030 , due to the rapid expansion of large-scale digital infrastructure, cloud computing and AI.

Number of Data Centers in France : according to a study by ADEME (the Agency for the Environment and Energy Management ) in January 2026, the organization lists 352 data centers in operation.

Electricity consumption :

• According to RTE, their total electricity consumption is 10 terawatt-hours (TWh) per year. As a reminder, 1 TWh (or 1000 GW) is equivalent to 1 billion kWh. In 2025, total electricity consumption in France was 449 TWh. Data centers therefore represent 2.2% of this annual consumption, equivalent to the electricity consumed by 9 to 10 urban areas with more than 100,000 inhabitants for a year.

• According to Hubblo, a SaaS platform for collecting energy consumption data , a typical data center consumes approximately 10 MW of electricity per year. However, given the enormous computing power required by generative AI, the new data centers currently being planned will consume 100 times more electricity each, amounting to nearly 1 GW per year , or the equivalent of the output of a nuclear power plant .

• The projects announced last week concern 7 to 8 GW, in addition to 14 GW already planned.

This would mean a need for 8 to 9 nuclear reactors in France, and a doubling of the share of electricity used in France by data centers to 4% of the total by 2035.

Water consumption : According to the World Economic Forum (WEF), a 1 MW data center can consume up to 25.5 million liters of water per year for cooling alone, equivalent to the daily consumption of nearly 300,000 people. Veolia predicts that by 2030, the total water consumption of data centers and semiconductor manufacturing will equal that of 46 million people, roughly the combined population of the metropolitan areas of New York, Los Angeles, and Paris.

What are they actually used for?

Data centers have become the foundation of many everyday uses: digital public services, professional tools, entertainment platforms, e-commerce, social networks… and artificial intelligence, which has become dominant in just a few years. A structural dependence has taken hold. Admittedly, on a historical scale, this dependence is recent. But it is nonetheless profound. The explosion of digital uses, driven primarily by the development of AI, is leading to a rapid increase in the need for computing and storage capacity.

WHERE ARE THEY BUILT AND WHY?

In Europe: abundant and carbon-free electricity

It was the availability of abundant (France is an exporter), stable, and carbon-free electricity, thanks to EDF's fleet of nuclear power plants, that convinced SoftBank to choose France. The accelerated "fast track" procedure launched in May 2025 by RTE, the electricity transmission network operator, also played a role. It identified eight sites suitable for data centers of over 400 megawatts (MW) that could be connected to the grid more quickly than through a standard procedure. The three sites in the Hauts-de-France region selected by the Japanese group were among those selected.

A strategic and political issue

We will return to this point later, but it is clear that the establishment of data centers on national territory (true for France, but also for other European countries competing for these investments) is more a matter of strategic considerations than economic ones, and above all, a question of "digital independence." Thus, on its website, ADEME titles its analysis on the subject "Digital sovereignty in question."

According to the organization, many services used in France rely on data centers physically located abroad. This reality raises a significant issue of digital sovereignty. Controlling one's data, guaranteeing its security, and limiting dependence on foreign infrastructure requires having processing capacity within one's own territory.

According to RTE 's website , "92% of Western countries' data was stored in the United States at the end of the 2010s. Since then, European rules on the protection of personal data have accelerated the development of hosting capacity in Europe. Major digital players are now seeking to locate their infrastructure as close as possible to their European users. Data centers have thus become a matter of digital sovereignty: European rules for responsible digital practices and ethical uses of artificial intelligence will be more difficult to apply to data centers located outside the European Union ."

Finally, regarding environmental impact, and according to Sébastien Broca's article in Le Monde Diplomatique cited below, the issue is primarily regulatory. "In environmental matters, European law requires economic actors to publish data on the consequences of their activities. The European Energy Efficiency Directive (EED), for example, requires large data centers to implement assessment and monitoring systems in order to record their energy consumption in a shared database." This is not the case, for instance, in the United States.

TECHNOLOGICAL AND ECONOMIC IMPACTS

Job creation?

According to La Tribune (March 4, 2026, updated June 1), in terms of employment, the number of direct jobs remains low, especially compared to the amounts invested or the workforce of former industrial sites. "It's not exactly inspiring: a data center generates on average a few dozen permanent jobs," emphasizes Thésée Data Centers, a company specializing in the field. The figure given is roughly one job per megawatt installed, although this varies depending on the customer profile and the intensity of usage. Stakeholders therefore highlight the indirect benefits, sometimes conflated in their communications with direct jobs. "Operators are primarily large industrial buyers," explains Antoine Fournier, CEO of Thésée. On construction sites, the effect is more visible: "During a construction phase, the number of people mobilized can represent several times the site's permanent workforce." "Therefore, it is with manufacturers and suppliers of electrical materials like Vinci and Schneider that the real impact on employment should be calculated, according to him. But the construction phase doesn't last forever…"

In an article by Matteo Wong in "The Atlantic", "Inside the Dirty, Dystopian World of AI Data Centers", he cites the case of Virginia, a state with a high density of data centers.

Loudoun County, in this state, has the highest concentration of data centers in the world, with 199 already operational and about 30 more under construction. According to a report, 13% of the world's data center capacity is concentrated in the county's 1,350 square kilometers. One particularly dense area is nicknamed "Data Center Alley."

Data centers typically employ only a few dozen people, but their construction has generated a steady stream of jobs . They also contribute nearly 40% to the county budget, funding services such as police, schools, and parks for a population that has been steadily growing since 2010.

Significant risks of failure

The water and electricity consumption of data centers, necessary for cooling the machines, is increasing all the more as service providers do everything possible to avoid what is known in industry as a "total blackout": a general failure due to a power outage, a water leak in the air conditioning system, a computer bug, etc. In 2017, for example, a massive outage at a British Airways data center led to the cancellation of 400 flights and stranded 75,000 passengers at London's Heathrow Airport. A prolonged failure of Amazon servers would pose a serious economic problem in the West.

Enormous operating costs due to this "total blackout" risk

In an increasingly competitive environment, many hosting companies are committed to ensuring their infrastructure is up and running 99.995% of the time, meaning just twenty-six minutes of service downtime per year. To strive for absolute availability, hosting providers are taking numerous precautions. First, they implement redundancy in their power distribution networks. "You end up with two power feeds, two generators, and rooms filled with lead-acid batteries as large as public libraries to ensure continuity between the outage and when the generators take over," explains Paul Benoit of Qarnot Computing. This system is often accompanied by massive logistical support. In short, " no building costs more per square meter than a high-end data center ."

And, as if that weren't enough, hosting providers are also duplicating the data centers themselves. At a conference around 2010, Google engineers reportedly explained that Gmail was duplicated six times, while the general rule is that a cat video should be stored in at least seven data centers worldwide . The industry is therefore plagued by "zombie servers," just as voracious as the others.

Finally, hosting providers oversize their infrastructure to anticipate traffic spikes. As a result, "if a router is operating at 60% of its capacity, that's the maximum," says computer scientist Anne-Cécile Orgerie. A corollary of this excess is a tremendous waste of electricity. An old New York Times investigation (September 22, 2012) revealed that some underutilized data centers could waste up to 90% of the electricity they consumed. At a conference held in late 2019 at Data Centre World (one of the major gatherings of cloud professionals) in Paris, an executive made this astonishing statement: "We realized that data centers were going to consume a third of the electricity in Greater Paris."

A risk of overinvestment and a financial bubble?

According to Matteo Wong's article in The Atlantic cited above, the race for artificial intelligence is expected to be the main driver of the roughly 2% annual growth in US electricity demand, which has stagnated for nearly twenty years. Nationally, this is not a crisis; regionally, however, the situation could be different. Dominion Energy, Virginia's largest electricity provider, forecasts 5.5% annual growth, with overall electricity demand doubling by 2039. By the end of the decade, training the industry's most powerful AI model could require as much electricity as... millions of American households.

In China, hundreds of data centers have been announced since 2023, and more facilities are planned under the ocean and in the desert. China's greatest asset in the AI race lies neither in the talent of its software engineers nor in the number of its data centers, but in its energy abundance: in 2024, the country produced almost as much electricity as the United States, Europe, and India combined.

President Trump declared the country faced an "energy emergency" and insisted on the need to build more power plants for the United States to win the AI race. A senior executive at OpenAI believes the US must leverage all available resources: solar panels, natural gas turbines, and nuclear reactors. Anthropic, OpenAI's main competitor, published a report advocating for streamlined permitting processes for data centers and power plants to remain competitive with China.

However, an internet-related energy crisis never materialized, contrary to what was feared at the time: in the 1990s, while the rollout of fiber optics was underway in Loudoun, energy companies built more coal and gas-fired power plants. "Dig more coal! PCs are coming!" Forbes headlined in 1999. Faced with unmet demand, the country ended up with a surplus of gas-fired power plants and numerous bankrupt energy companies.

The rise of generative AI could also turn out to be a speculative bubble. This technology remains extremely expensive, primarily due to the price of advanced computer chips, and no AI company has yet presented a convincing business model. One avenue for profitability could lie in more efficient algorithms, which would eliminate the need for new gas-fired power plants. And if AI doesn't prove as revolutionary as experts predict, entire sections of data centers could remain unused or unfinished, relics of a future that never materialized.

Record investments, unprecedented in terms of both amount and speed…

The amounts invested in AI are unprecedented—accounting for a third of US growth—and largely concern infrastructure, including data centers. The entry ticket to AI is a trillion dollars. Morgan Stanley anticipates that the sector will see three trillion dollars in growth between 2025 and 2029. Goldman Sachs, meanwhile, estimates that if additional investments were to represent 2 to 3% of GDP, similar to the development of rail and automotive infrastructure, capital expenditures by "hyperscalers" would reach approximately $1.1 trillion in 2027 (representing 45% growth).

To meet these needs, tech leaders are forming circular funding rounds, thereby adding a significant risk of concentrating the burden of these investments. On March 31st, OpenAI finalized an unprecedented $122 billion funding round. Amazon contributed $50 billion, in exchange for which OpenAI committed to $100 billion in orders for use of Amazon Web Services (AWS) cloud computing. Meanwhile, OpenAI and AWS placed massive chip orders with Nvidia , with Nvidia investing $30 billion in OpenAI's funding round. Shareholder relationships are merging with customer relationships, and vice versa, creating an overall pattern that, according to economist Frédéric Lordon, author of "The Crisis Too Far," resembles the Ouroboros: the ring-shaped snake that bites its own tail. A single cut somewhere, and the inner tube collapses completely. (Le Monde Diplomatique, May 2026)

But there is no shortage of potential disruptions. The Ouroboros doesn't operate solely on capital but also on debt. OpenAI has thus achieved the feat of getting its partners to indebt themselves in order to create the infrastructure for its own development, in exchange for commitments to future orders—amounting to $100 billion. The entire structure rests on the profit realization timeframe indicated by Mr. Sam Altman himself regarding OpenAI: 2030. This is based on the implicit assumption of a kind of accelerated takeoff, since revenue would jump from $20 billion today to $200 billion in 2030. Incidentally, HSBC analysts are clearly struggling to discern the profits announced by Mr. Altman for that timeframe, instead predicting a persistent deficit of around $200 billion to be covered, due to astronomical computing costs (accumulating at $1.4 trillion by 2033) and a $620 billion data center rental bill . How can one not question the financial system's capacity to bear such a colossal risk over such a long period? One single missed opportunity in profitability meetings or refinancing stages, and everything falls apart.

For those of us who are a little more familiar with the financial markets, the fact that shareholders of Open AI, Anthropic, and SpaceX (which owns X AI) have decided to sell some of their shares on the market in the coming months (currently in the case of SpaceX), is a good indicator that we are approaching the end of the bull market cycle in the US…

ENVIRONMENTAL IMPACTS

The figures are staggering: the global digital industry consumes so much water, materials, and energy that its footprint is three times that of a country like France or the United Kingdom. Digital technologies now use 10% of the world's electricity and are estimated to emit nearly 4% of global carbon dioxide (CO2) emissions, slightly less than double that of the global civil aviation sector. "If digital companies prove more powerful than the regulatory bodies that oversee them, there is a risk that we will no longer be able to control their environmental impact," warns Jaan Tallinn, founder of Skype and the Future of Life Institute, which focuses on the ethics of technology.

WATER

Consumption figures

According to Les Échos, in France, refering to a report by Arcep cited by France Info, "the volume of water withdrawn by data centers [in France] already reached 681 million liters" in 2023, a volume experiencing sustained growth (+19%) and consisting "almost entirely" of drinking water. To operate, a data center also needs electricity, partly generated by hydroelectric power plants, which also requires thousands of liters of water. According to a European report, this "consumed" volume was 400 million liters of water in France, also in 2023.

Veolia's website explains why water is essential to the operation of data centers. It is at the heart of many cooling systems – chillers, cooling towers, or liquid cooling – and the volumes required are colossal.

According to Le Échos dated May 18, 2026 , a data center can consume up to 19 million liters of water per day, and data centers are becoming increasingly water-intensive. This demand puts significant pressure on local areas, especially in regions already facing water stress or where a large number of data centers are concentrated. In some arid regions, such as Arizona, their needs have even exploded by more than 800%. In the Phoenix area alone, 14 billion liters of water were required in 2025 to provide cooling.

In a context of climate change marked by higher average temperatures and more frequent episodes of drought, the “thirst” of data centers becomes a crucial issue, as their number explodes to meet the ever-increasing demands of AI models seeking computing power.

The rapid development of artificial intelligence and the expansion of associated data centers will only exacerbate this problem. It is estimated that GPT-3, an AI model, consumes 500 ml of water for every 10 to 50 responses.

By 2027, global demand for AI is expected to represent a water withdrawal of 4.2 to 6.6 billion cubic meters, more than 4 to 6 times Denmark's total annual water withdrawal.

Similar observation in France

But the real problem with this water consumption is that it isn't recycled. 70% to 80% of the water used to cool the servers is simply lost through evaporation into the air. According to Veolia's website, the rest is discharged to municipal wastewater treatment plants.

Priority is given to water reuse, while preserving its quality.

According to Veolia , alternatives exist: closed-loop cooling systems, where water circulates continuously and is cooled before being reused. These solutions reduce water consumption while optimizing industrial processes. However, their effectiveness depends on one condition: appropriate water treatment.

Untreated water can cause numerous problems. The presence of suspended solids can, for example, disrupt flow rates and fluid distribution. Particles can become lodged in components such as valves, accelerate corrosion, or lead to sludge formation. Furthermore, biofilms can create insulating layers that reduce heat transfer, while corrosion and erosion from debris can shorten equipment lifespan.

These undesirable effects can be corrected through chemical treatments.

The environmental impact of chemical treatments

Traditionally, water treatment in data centers has relied heavily on the use of chemicals such as chlorine, which pose significant environmental and operational safety challenges. Their handling and storage within the facilities present risks, and the use of biocides or conventional chemicals can generate pollution and harm aquatic ecosystems if not properly controlled.

However, alternative methods and products now exist that are less operationally demanding and less harmful to the environment. Among the most commonly used treatment approaches are water softening, reverse osmosis, and direct nanofiltration.

Efficient water use

For data centers, it's not enough to simply use clean, sustainably treated, and efficient water: the cooling systems themselves must operate efficiently to reduce the environmental impact of their consumption. In this context, closed-loop cooling systems, incorporating wastewater recycling or rainwater harvesting, can play a key role and reduce freshwater use by 50 to 70%.

Adopting this type of circular water management strategy will be essential to address environmental pressures as demand for data centers continues to grow. Indeed, given the astronomical volumes of water consumed for cooling, even small improvements can have a considerable impact.

“By adopting a holistic approach to water management, data centers can save resources throughout their entire operating cycle, where water plays a role at every stage. In a context where sustainability, water scarcity, reuse, and resource efficiency are closely scrutinized, it is essential that data centers adopt a proactive approach.”

In any case, a huge organization…

ENERGY

Enormous needs at a global level

Eric Schmidt, former CEO of Google, addressed US lawmakers in April 2025: "What we need most is energy, in colossal quantities. What we expect from you is electricity in all its forms, renewable, non-renewable, it doesn't matter, the important thing is to have a lot of it and for it to arrive quickly." According to him, the United States will need the equivalent of 67 nuclear power plants by 2030 to power its data centers. However, it takes between 12 and 15 years to build a nuclear power plant. France, on the other hand, already has electricity. That said, our capacity is expected to be increasingly strained by other uses (such as the development of electric vehicles), and trade-offs will undoubtedly be necessary at some point.

Source: AleaSoft

AI's carbon footprint is difficult to measure.

According to Sébastien Broca, a lecturer in information and communication sciences at Paris-VIII University, writing in Le Monde Diplomatique, Silicon Valley does not publish data on the subject. “ChatGPT consumes a lot of energy. However, accurately quantifying the energy footprint of each use is extremely difficult. The electricity consumed to fulfill a particular request depends on a multitude of variables: the artificial intelligence (AI) model used, the complexity of the request, and the data center to which it is directed. As for the carbon dioxide (CO2) emissions generated, they fluctuate depending on the electricity source, which varies from one location to another, but also from one time of day to another. To make matters worse, many of these elements are impossible to know. OpenAI reveals neither the number of parameters in its most recent models, nor the data centers where it processes user requests, nor the energy sources used to power its infrastructure.” The energy footprint of a ChatGPT request is akin to a puzzle designed so that no one can solve it.

The mystery surrounding OpenAI is not an isolated phenomenon. All the major players in AI are now hiding behind the argument of trade secrecy to release as little information as possible about the energy (and, more broadly, environmental) impact of their technologies . After the emergence of ChatGPT in December 2022, companies in the sector drastically reduced the information they shared with the public and independent researchers. The semiconductor giant Nvidia is no more forthcoming. It is impossible to know the carbon footprint of its graphics processing units (GPUs) and their replacement rate, even though they number in the thousands in the large data centers that are accelerating the rise of AI.

The tech giants have nonetheless been forced to admit that they will likely not be able to meet their announced climate goals. According to Microsoft's latest environmental report, its "real" greenhouse gas emissions—that is, those generated by all of its operations, without taking into account offsetting mechanisms—more than doubled between 2020 and 2024! Indeed, the digital industry cannot currently do without carbon-intensive energy sources, such as coal and natural gas. In the United States, many data centers are located in Virginia, West Virginia, and Pennsylvania, states where renewable energy is underdeveloped. Furthermore, recent investments by tech companies in nuclear power—from the reopening of the old Three Mile Island plant to highly speculative fusion projects—leave no doubt about the future growth of their energy needs.

In addition to the deliberate ambiguity surrounding the environmental footprint of their technologies, industry giants are employing a second strategy. Leaders like Sam Altman (OpenAI), Dario Amodei (Anthropic), and Eric Schmidt (former president of Google) claim that AI will ultimately solve the problems associated with climate change.

To reduce AI's energy consumption, we must first quantify it. The young Dutch researcher Alex de Vries runs the website Digiconomist, created in 2014 to document the environmental impact of Bitcoin. He is credited with estimating that ChatGPT consumes ten times more electricity than Google's search engine, a figure subsequently adopted by the International Energy Agency (IEA) after its publication in 2023. In a more recent article, de Vries estimates that AI already requires as much electricity as a country like Switzerland and that it will soon account for 50% of global data center electricity consumption. In France, the think tank The Shift Project, founded by Jean-Marc Jancovici, puts forward comparable figures. Data centers could be responsible for 2% of global greenhouse gas emissions within a few years, with AI alone accounting for nearly 1%.

Canadian researcher Sasha Luccioni excels in these areas. In charge of climate issues at the Franco-American company Hugging Face (which promotes open-source AI against the closed models of major industry players), she demonstrated as early as 2023 that the carbon footprint of AI was twice as large as previously estimated, when all stages of model development were taken into account, from equipment manufacturing to end-use. Ranked by Time magazine in 2024 among the "100 Most Influential People in AI," she now uses her platform to denounce the lack of transparency among large tech companies and advocate for the development of more energy-efficient AI models.

Luccioni bases her commitment on this conviction: if the public knew how much energy each search consumes, she believes, they would be "more reluctant to use AI for superfluous tasks like searching for a country's capital." With this in mind, the idea emerged of creating an "eco-score" for AI, modeled after the energy label given to household appliances, the Nutri-Score on food packaging, or fuel consumption data for vehicles. This, it is believed, would guide consumers toward more sustainable practices.

In matters of climate change or reducing chemical pollution, market regulation through information has failed. There is no reason to believe that AI will be any different, a field where the evolution of uses depends less on informed user demands than on tech players, who are pushing for the integration of their solutions into all economic sectors and everyday tasks.

OTHER IMPACTS ON THE ENVIRONMENT AND LAND

SoftBank's massive investment, which plans to add another 2 GW after the first phase ends in 2031, and other Choose France projects will fuel criticism regarding the environmental and land-use impact of these giant digital factories, already very prevalent in France. Concerns are also being raised about the risk that France's energy resources will be monopolized by these data centers, often owned by foreign capital, to the detriment of other uses, such as the electrification of transportation or industry.

For example, according to Le Monde (May 30, 2026), Google plans to build a massive data center in the Indre department, its first outright operation in France. This has drawn criticism from a group attempting to block its construction. The 195-hectare site has been reserved since July 2025 for its French subsidiary, Tricolore Computing, to eventually house thousands of computer servers. This would be Google's first outright data center in France; until now, the company has leased cloud services from other partners, particularly in the Île-de-France region.

According to Google's projections, the phased construction of its data center in the Indre region will reach full capacity with an electrical output ranging from 880 megawatts to 1.1 gigawatts, roughly equivalent to that of a conventional nuclear reactor, as mentioned earlier. Invited to elaborate on its project in the Centre-Val de Loire region, Google France did not respond to requests from Le Monde. This discretion is a source of tension locally, both regarding employment and the appropriation of land and water resources, which is fueling emerging opposition.

A HIGHLY POLITICAL ISSUE IN THE END

In conclusion, on an extremely complex and ultimately divisive subject, I have decided, in order to maintain a semblance of factual neutrality, to summarize and reproduce the ADEME report on the future of data centers in France and the 5 choices facing our society, a way not to sidestep the issue, but to pose the subject in terms of choices to be expressed by ourselves and our representatives.

But first, a little detour to Ireland…

The Irish example

The professional website ZD Net, in an article published on June 8, 2026, headlines: "Data centers: Ireland is suffocating and imposes 'Bring Your Own Power' on Tech giants."

Faced with energy consumption exceeding that of urban households, Ireland is ending grid connections and forcing cloud giants to deploy their own power sources to continue expanding. After imposing a nearly three-year moratorium banning the construction of new data centers , the government is relaxing its policy, but under a strict condition: the BYOP principle, or "Bring Your Own Power."

From now on, any new site construction or expansion requires the operator to provide its own power plant or sign exclusive supply contracts for new renewable energy sources nearby. Simply connecting to the public national grid has therefore become impossible.

This decision comes as infrastructure concentration reaches a critical threshold for local residents and businesses, notes the Wall Street Journal. Data centers now consume 21% of the country's total electricity production, a proportion that exceeds 50% in the Dublin area and neighboring County Meath.

For the Irish state, the challenge is to maintain the country's attractiveness without provoking public anger or hindering urban development. American technology companies represent a crucial source of revenue, accounting for approximately 22% of the country's tax revenue in 2024. However, this coexistence is becoming problematic in terms of local distribution infrastructure.

And Ireland is not alone in this situation. To such an extent that the Irish situation serves as a laboratory for other regions of the world also facing this growth crisis, from Virginia in the United States to countries like India or Mexico, which are concerned about the pressure exerted on electricity and water resources.

ADEME report, January 2026

Data center electricity consumption: 5 scenarios for tomorrow

Because data centers have become indispensable, ADEME is exploring five possible trajectories for the evolution of data center electricity consumption in France by 2035 and then 2060. Energy efficiency, territorial organization, technological gamble, or headlong rush: these different scenarios present major differences in terms of energy sovereignty and CO2 emissions, both in France and abroad. And the decisions made today will have a lasting impact on the future.

What will data center energy consumption look like by 2060?

To inform the possible choices, ADEME has modeled five scenarios for the evolution of electricity consumption in data centers. These are not predictions, but coherent trajectories, based on different societal choices, uses, and public policies.

Base scenario: if we change nothing

In this scenario, current trends continue. The number of data centers increases sharply, as does their electricity consumption. Artificial intelligence, cloud computing, and the proliferation of digital services are the main drivers of this growth. This trajectory translates into a rapid increase in greenhouse gas emissions and heightened pressure on energy, water, and land resources. In the short term, this development can be considered without major constraints on the French electricity system, which has been a major electricity exporter for several years. However, it appears difficult to reconcile with climate objectives. Indeed, in this scenario, electricity consumption driven by French usage could increase 3.7 times by 2035. And nearly two-thirds of this consumption would take place abroad, in countries where electricity mixes are, on average, much more carbon-intensive than in France.

Scenario 1: Reduce digital demand

This scenario is based on a deliberate commitment to digital restraint. Certain uses are being questioned, particularly those whose social utility is deemed low compared to their environmental impact. The most energy-intensive services, including some AI applications, are subject to strict regulation. This approach allows for a slowdown in growth, followed by a gradual decrease in the long-term electricity consumption of data centers.

Scenario 2: Organize and prioritize uses

Here, data center development is carried out in close consultation with local communities. Sites are regulated, and digital uses are prioritized based on their social, environmental, or health benefits. This approach relies on local synergies, particularly for recovering heat from data centers (which can be used in district heating networks), and on mitigating pressure on water, land, and energy networks. Consumption remains controlled, without any sudden decline.

Scenario 3: Leveraging innovation

In this trajectory, the development of data centers and digital applications, including AI, continues at a sustained pace. The focus is therefore on energy efficiency, technological innovations, and the use of a low-carbon electricity mix to control the impact. Consumption is nevertheless increasing significantly, even if part of the impact is offset by efficiency gains and the expected benefits of digital technology in other sectors of the economy.

Scenario 4: offset the impacts

This scenario is based on a rapid expansion of digital usage, AI, and associated infrastructure, with limited regulation. Technological solutions are primarily deployed to offset environmental impacts. Data center electricity consumption increases sharply, both in France and especially abroad, to meet French consumer demand. This trajectory has a significant consequence: it increases dependence on infrastructure located outside of France.

Thus, the continued growth of digital uses raises a central question: how far can data centers be developed without jeopardizing energy and climate balances, as well as the preservation of resources? It is precisely to shed light on this question that ADEME has modeled these five contrasting trajectories.

Digital technology imposes societal choices

Data centers are neither inherently good nor bad. They constitute infrastructures serving uses that can be beneficial, but also highly resource-intensive. Their future depends on the choices we make collectively and individually regarding digital technology. The ADEME study does not prescribe a single path. However, it shows that the status quo is not neutral. Depending on the choices made, the energy, climate, and territorial impacts can vary considerably.

So, who can take action and how?

Public authorities play a key role. They can regulate the development of data centers, set standards, prioritize certain uses, and integrate these infrastructures into coherent energy and territorial planning, in order to combine increased digital sovereignty with control of greenhouse gas emissions and carbon footprint. Businesses and users also have a decisive role. Economic models, the design of digital services, and the collective relationship with AI directly influence demand. Together, these choices shape the future of digital technology and its environmental footprint.

One final question: Have you heard any political party or elected official at the national level raise these issues to date? If so, I'd be interested to hear about it!