Data centres are becoming the backbone of Europe’s AI ambitions, but also a test of their limits. From Ireland to France, the debate is no longer just about digital growth. It is about energy, climate, public acceptance and who pays for the infrastructure powering the AI age.
In just a few years, data centres have gone from being discreet technical buildings to strategic infrastructure at the heart of Europe’s digital sovereignty. A cluster of recent publications now lets us weigh both their value and their blind spots. KPMG has assessed what they are worth to Ireland; France’s regulator, Arcep, published the fifth edition of its sustainability survey in May 2026; and a working paper from University College Cork, presented to the Oireachtas Committee on Artificial Intelligence in June 2026, models what the next wave of expansion could mean for Ireland’s climate and energy targets. Around these analyses, a noisier public debate has grown up, traced by The Irish Times in Dublin and by the review Le Grand Continent on the American front. Read together, they describe one and the same tension. Demand, supercharged by artificial intelligence, is running up against the physical, environmental and now social limits of our electricity systems almost everywhere. What changes from one country to the next is the scale.
Commissioned by Ireland’s Department of Enterprise, Tourism and Employment, KPMG’s The Value of Data Centres to Ireland is the most detailed attempt yet to pin down what this sector really represents in a national economy. Its central claim is not about the sector’s own footprint, which stays modest, but about its role as enabling infrastructure for the whole digital economy.
That direct footprint is genuinely small. In 2024, construction and operation generated roughly €2.2 billion in gross value added, about 0.4% of the national total, and close to 19,500 jobs, up from 2,478 in 2010. The report does point to a real knock-on effect: every €1 million of direct value added is estimated to generate about €0.75 million more elsewhere in the economy, through suppliers and household spending. The sector has also nurtured Irish expertise that travels well, with some €2.1 billion of exports in 2024 tied to data centre construction by Irish-owned firms, around 40% of the construction exports of Enterprise Ireland clients.
The real argument lies in how much the rest of the economy now leans on this infrastructure. The six sectors most dependent on it (ICT, finance and insurance, health, transport and logistics, trade, and professional services) together account for €239.6 billion in value added, roughly 45% of national value added. On the report’s most conservative scenario, around €89 billion of that, or about 17% of the national total, rests on having latency-sensitive computing capacity physically located in Ireland.
To its credit, the report resists easy advocacy. It accepts that a share of workloads could in principle be served from data centres elsewhere in the EU, while stressing that the latency-sensitive core genuinely needs to sit in Ireland. The country’s appeal rests on a familiar combination of factors: a mild climate suited to cooling, deep subsea-cable connectivity, a skilled English-speaking workforce, direct access to the single market, and a competitive corporate tax regime. None of those is guaranteed forever, which is why Ireland’s position is best read as an advantage to be sustained rather than taken for granted.
Those headline figures deserve careful handling, and they have prompted lively discussion within Ireland itself. In The Irish Times, the columnist Cliff Taylor noted that some of the most striking numbers drawn from the report by Government, around €100 billion in annual value added, 876,000 jobs and €14.6 billion in employment-related taxes, all hang on a single word: enabled. Attributing such volumes of jobs directly to the physical presence of data centres, he argued, can paradoxically weaken the public case by handing critics an easy target. The stronger and quieter point is that, by letting global tech firms build their own facilities here, Ireland creates what KPMG calls a “sticky” investment footprint that helps it hold on to existing investors and attract new ones. Columnist Una Mullally went further in a sharper piece, noting that members of Government themselves seem to read the figures more cautiously: at a separate launch in late May, Minister of State Timmy Dooley referred to the data centre ecosystem as supporting “tens of thousands” of jobs rather than several hundred thousand.
There is a counter-narrative, and it is just as motivated. A joint report from Friends of the Earth Ireland and Beyond Fossil Fuels, The Cost of Data Centres, estimated that the average household may have paid around €360 in additional electricity costs between 2015 and 2023 because of pressure on the grid. The wider context matters here: according to the Household Energy Price Index, Dublin was in May 2026 the most expensive EU capital for electricity, around 52% above the European average, and Eurostat ranked Ireland top of the EU for medium-sized household prices in the second half of 2025. The Minister for Finance Simon Harris cautioned against framing data centres as an easy “bogeyman“, and that caution is reasonable. The fact remains that, in the public mind, the link between large industrial loads and household bills is now part of the debate, and it deserves clear answers.
The healthy reading is that Ireland is doing its own homework in public. A serious consultancy report, two well-argued opinion columns, an environmental counter-study, and ministerial pushback are all part of a single conversation about how to govern this infrastructure well. That openness is itself an asset.
Beyond its economic role, Ireland’s success has also placed real strain on its electricity system. In 2024, Irish data centres used around 22% of the country’s electricity, against roughly 3% on average across Europe, and the contrast with neighbouring economies is striking: data centres take about 2% of electricity in France, 5% in Germany and the United Kingdom, and 6% in the Netherlands. The Irish figure is more than every urban household in the State combined, and the share could climb above 30% within five years on the basis of expansions and connections already approved.
The very things that made Dublin Europe’s second-largest hub, with about 1,150 MW installed (just behind London and ahead of Paris), have also become constraints. Demand is heavily concentrated around the capital, where data centres take up half of regional electricity, far from the wind and solar of the west and south. That weighs on a grid that needs significant investment and deepens a reliance on imports, which covered 80% of energy in 2024. To guard against the risk of large-scale outages, four emergency gas and diesel plants came online in 2024. On the climate side, the sector accounts for between 2.5% and 4.5% of national emissions, uses between 0.5 and 5 million litres of water a day, and leaves the country exposed to EU penalties that could reach €26 billion. This is what led the Commission for Regulation of Utilities (CRU) to impose what amounted to a moratorium on new connections in the Greater Dublin area as early as 2021.
A way forward is now taking shape, and it looks like a doctrine. The CRU’s Large Energy Users Connection Policy of December 2025 relaxes the moratorium but asks new centres to provide dispatchable on-site or proximate generation equivalent to their capacity, and to match at least 80% of their annual electricity demand with new renewable generation in Ireland, with a six-year glide path that can run through corporate power purchase agreements. January 2026’s Large Energy-User Action Plan pushes in the same direction, encouraging centres to sit next to renewable generation. The Government has also committed up to €18.9 billion in grid investment by 2030. Households will carry 55% of that bill, even though they already pay around €360 more for electricity than the Western European average, while the centres enjoy notably favourable tariffs. Europe’s first micro-grid, a 110 MW site west of Dublin inaugurated in March 2026, already runs off-grid on its own gas plant.
The deeper question is what happens if the new framework is used to its capacity. A working paper from University College Cork by Teresa Bonserio and Hannah Daly, presented to the Oireachtas Committee on Artificial Intelligence on 9 June 2026, sets out a careful answer. Drawing on the market intelligence exercise that accompanied the CRU policy, the authors stress that they are not forecasting demand but stress-testing what the framework could enable: up to 5.8 GW of additional data centre capacity by 2040, on top of what is already contracted. To give a sense of scale, that figure is roughly equivalent to the all-time peak demand of the entire Republic, 6.02 GW recorded in January 2025.
The paper’s findings are striking, and they are presented openly as an “illustrative stress-test rather than a projection“. Under the assumptions used, electricity demand attributable to new data centres would reach about 8 TWh in 2030 and 35 TWh in 2040, taking the data centre share of total electricity from 22% in 2024 to 38% in 2030 and 55% in 2040. Total national electricity demand would rise by roughly 160% between 2024 and 2040.
The most counter-intuitive finding concerns renewables. The policy’s matching requirement would call for around 20 TWh of additional renewable generation each year by 2040, on top of what is already needed to decarbonise the rest of the system. Yet because new demand grows faster than new supply during the six-year glide path, the share of renewables in the electricity mix would actually fall, from 77% under the official baseline to 65% in 2030, before recovering to 77% by 2040, still short of the 80% Climate Action Plan target. The authors are careful to distinguish “additionality of renewable capacity” from “decarbonisation“: new wind or solar can be financially additional and still be absorbed by new demand, without reducing fossil generation elsewhere.
The implications for gas and emissions follow from the same logic. Instead of the 44% decline in gas-fired generation projected to 2030, the scenario flips the curve into a roughly 20% increase, and gas input to power generation in 2040 would be around five times higher than in the baseline. Additional electricity-sector emissions reach about 3.3 MtCO₂ a year by 2030, slightly above the entire indicative 2030 ceiling for the power sector (3 MtCO₂), and cumulative additional emissions across the carbon-budget periods to 2040 exceed 60 MtCO₂.
It is worth being precise about what this means. The paper does not predict that 5.8 GW will be built; it shows what the architecture of the current policy permits if industry uptake meets the market intelligence range. As the authors note, slower delivery, lower utilisation rates, faster renewable build-out, or earlier procurement could all reduce the impacts; conversely, faster build-out or annual rather than hourly matching could increase them. The contribution is not a verdict but a framework for explicit assessment of trade-offs, exactly the kind of evidence-based scrutiny that a mature policy debate calls for.
That, in fact, is how the conversation is now framed in Dublin. The Oireachtas committee heard the paper alongside complementary perspectives, including from Professor Andrew Parnell of UCD on the potential of AI to improve weather forecasting and energy management, and the committee’s stated intention is to host “a balanced discussion about both the challenges and opportunities” of AI. A United Nations report in early June described Ireland as a “cautionary example” of how concentrated digital infrastructure can stress local grids. Read alongside the UCC paper, the more useful framing is perhaps simpler: Ireland is conducting, in public and in real time, the policy stress-test that other European countries will need to run in turn.
On 21 May 2026, Arcep published the fifth and most comprehensive edition of its “Sustainable Digital” survey. The data centres it surveyed consumed 2.7 TWh of electricity in 2024, up 12% in a year and around 38% since 2021, with AI doing most of the pushing.
The familiar paradox shows up here: efficiency keeps improving, yet it cannot keep up with the sheer growth in demand. Centres less than ten years old run at an average PUE of 1.30, against 1.52 for older ones, but those gains vanish into the change of scale. Centres commissioned between 2021 and 2023 saw their consumption jump 66% in a single year, and the International Energy Agency reckons the power density of AI servers rose elevenfold between 2020 and 2025. Emissions followed, climbing from 124,000 to 178,000 tonnes of CO₂ equivalent between 2021 and 2024. The map is lopsided as well, with three regions (Hauts-de-France, PACA and Île-de-France) holding close to 90% of capacity and consumption, and Île-de-France alone drawing more than 70% of the electricity used by the centres surveyed.
Two caveats matter. The survey captures only about half of colocation centres, and the hyperscalers (Microsoft Azure, Google Cloud) are not in it yet; their inclusion in the 2027 edition should lift the picture appreciably. Even so, the national share stays modest for now. RTE, the grid operator, expects data centre consumption to roughly triple, reaching somewhere between 15 and 20 TWh by 2030 (around 3% of French electricity) and 23 to 28 TWh by 2035 (about 4%), which remains comfortable against a generating surplus of roughly 20%.
France’s framing reflects its own circumstances, chiefly a large low-carbon nuclear fleet that, for now, turns abundant power into a selling point rather than a worry. At the Paris AI Action Summit, President Macron caught the mood with “plug, baby, plug“, a deliberate answer to the American “drill, baby, drill“, alongside €109 billion in announced private investment. The accompanying drive to simplify planning would give strategic data centres “major national interest” status, smoothing permits and speeding up grid connections. The appetite is real: by May 2026, RTE had received connection requests and reservations approaching 18 GW across some eighty projects, against an installed base of only about 0.7 GW. Many of those reservations are speculative, since recently connected sites draw on average just a fifth of the power they asked for, but the direction of travel is unmistakable, and it brings the same grid and acceptability questions that Ireland is now examining in detail a good deal closer.
A third front is opening, one still largely off Europe’s radar: the social licence to operate. With more than 4,000 data centres, the United States has the world’s biggest fleet, and local opposition is multiplying. As Le Grand Continent reports, the Data Center Opposition Report counted 268 campaign groups in early April 2026, with roughly one new group forming every day, while the US Data Center Moratorium Tracker logged 86 local moratoriums by mid-May, 61 of them active, across 35 states. Most of this opposition works through the courts and through moratoriums aimed at tightening the rules on water use, noise and emissions. The scale can be startling: the single mega-complex Fermi America wants to build in Amarillo, Texas, could emit up to 40 million tonnes of CO₂ equivalent a year, more than Jordan or Croatia.
Le Grand Continent also reports that, according to internal documents, some US security agencies have begun to anticipate the possibility of an “anti-technology” protest movement should the rollout of AI accelerate. Opinion there is divided along party lines: a Pew Research Center survey found that half of Democratic voters see data centres as bad for the environment, compared with fewer than a third of Republicans. For Europe, the lesson is less about American politics than about a more general point. Public acceptance is becoming as much of a constraint as grid capacity, and Ireland, where the conversation is already lively and well-documented, is feeling it earlier than most.
At Union level, two strands are being pulled at once, and the tension between them is deliberate. The first is a drive to build sovereign computing power. The European Commission’s AI Continent Action Plan, presented in April 2025, sets out an ambition to mobilise up to €200 billion to make Europe a leading “AI continent“. Its centrepiece, the InvestAI initiative announced at the Paris summit in February 2025, sets aside €20 billion for up to five “AI gigafactories“, vast clusters of roughly 100,000 advanced processors each, federated with the EuroHPC network of AI factories now being rolled out across the Union. Interest has been strong. An early call for expressions of interest drew 76 submissions covering some 60 sites across 16 member states, and in December 2025 the European Investment Bank joined the Commission to help draw in private money. The forthcoming Cloud and AI Development Act goes further again, aiming to at least triple the EU’s data centre capacity within five to seven years, while pointedly favouring “sustainable” centres and tying simplified permits and public support to energy and water efficiency.
The second strand tightens the environmental rules. Under the recast Energy Efficiency Directive of 2023, operators already have to report standard performance indicators to a European database. In March 2026 the Commission went a step further, registering a draft regulation for a common, Union-wide scheme to rate the sustainability of data centres, a move toward a harmonised label and more transparency. A strategic roadmap for digitalisation and AI in the energy sector, together with a dedicated efficiency package, is expected to nudge operators toward grid flexibility, reusing waste heat and recycling water, all in service of the goal of climate-neutral, highly efficient data centres by 2030, a decade over which the sector’s own share of EU electricity is itself heading past 3%.
Put the two strands together and you have the European balancing act in a single frame: the Union wants more compute and a cleaner footprint at the same time. For member states, the practical question shifts from how much capacity to host to how to host capacity that is genuinely sustainable rather than merely larger. That is precisely the calibration Ireland is examining now, at national scale, and the one France will meet as its own pipeline fills.
Step back from any France-and-Ireland comparison and the stakes are plainly continental, not least when it comes to where the value finally lands. The FLAP-D markets (Frankfurt, London, Amsterdam, Paris, Dublin) still hold most of Europe’s colocation demand. Yet, as the Irish Times points out, the “gigafactories” at the cutting edge of AI remain almost entirely American, which is exactly what the Union’s compute push is trying to change. Constrained on the energy side, Ireland is unlikely to host one, and that captures the situation precisely: it has already supplied a great deal of international capacity without yet securing the top of the AI value chain. The France-Ireland electricity interconnector due in 2028 shows the more cooperative side of the same interdependence, a shared piece of infrastructure that will help take some of the strain off Irish demand.
A few threads run through all of this. Efficiency, on its own, is not enough, because every gain in PUE keeps being swallowed by raw growth in demand. The grid, rather than land, has become the scarce resource. And the question of who pays to reinforce the network, whether the consumer, the State or the operator, will only grow more pressing as AI demand climbs.
In the end, the various Irish, French and European publications point in the same direction. The sector creates real value, even if the headline numbers are arguable, but that value depends on the grid’s ability to absorb it, on policy coherence between energy, climate and digital objectives, and on the public’s willingness to accept it. Ireland moved first, and at an unusual scale, and it is now working through choices that every European country with AI ambitions will face sooner or later. The transparency of its public debate, including rigorous scientific contributions such as the UCC working paper and ministerial caution against framing data centres as a “bogeyman“, is itself part of the answer. The rules being written there, on connection, on-site generation, who pays and how communities are brought along, are the very questions others, France included, would be wise to think about today. The most useful stance is not comparison but cooperation, and the interconnector, the EU’s compute strategy and shared standards all point that way.
Sources: KPMG, “The Value of Data Centres to Ireland” (for the DETE); Arcep, “Sustainable Digital” survey, 5th edition (21 May 2026), via L’Usine Digitale; RTE, explainer on the rise of data centres in France (rte-france.com) and Bilan prévisionnel (electricity outlook) 2025-2035, December 2025; Teresa Bonserio and Hannah Daly, “Data centre expansion in Ireland: the conflict between grid connection policy and national energy and climate targets“, Working Paper, University College Cork / Zenodo (June 2026); Cliff Taylor, “Ireland is caught in a data centre trap and there’s no easy way out“, The Irish Times (6 June 2026); Una Mullally, “More than 870,000 jobs created by data centres? Don’t make me laugh“, The Irish Times (8 June 2026); Caroline O’Doherty, “Data centre electricity demand could exceed entire power usage of Irish Republic“, The Irish Times (9 June 2026); Friends of the Earth Ireland and Beyond Fossil Fuels, “The Cost of Data Centres“; United Nations Institute for Water, Environment and Health, report on the environmental cost of AI’s energy use (June 2026); “Sabotaging data centres?“, Le Grand Continent (27 May 2026), drawing on Wired, the Data Center Opposition Report and the Pew Research Center; European Commission, AI Continent Action Plan (April 2025), InvestAI and the AI gigafactories programme, and the forthcoming Cloud and AI Development Act; recast Energy Efficiency Directive (EU) 2023/1791 and the draft Union-wide data centre sustainability rating scheme (March 2026); Paris AI Action Summit.