Europe’s artificial intelligence (AI) boom risks driving up power bills and public backlash. But almost every watt that feeds data centres emerges again as heat – energy Europe cannot afford to waste. With smart rules, targeted investment, and transparent reporting, server halls could be transformed from contested costs into civic assets: warming homes, cutting carbon, and proving AI’s worth to citizens.

From code to concrete

AI is no longer written only in code; it is being poured into concrete. Across Europe, vast campuses of servers and supercomputers are rising, as the European Commission pushes to make good on its desire to brand the continent an ‘AI power.’

Under the European High-Performance Computing Joint Undertaking, a network of AI factories is taking shape: 15 sites are slated to open by 2026, anchored by 9 new AI-optimised supercomputers. Startups and small and medium-sized enterprises get priority access, while an InvestAI facility worth EUR 20 billion is set to finance up to five ‘gigafactories,’ each designed to host more than 100,000 advanced processors capable of training the largest models.

Computing power – the processing capacity that trains and runs AI models – has now become a strategic resource, embodied in the sprawling architecture of AI data centres. As with semiconductors or critical raw materials, access to clean, reliable power for computing is becoming a question of strategic autonomy as much as of innovation. Once seen as peripheral warehouses, these complexes have become the backbone of the digital economy: financed at speed, scrutinised by markets, and ravenous for electricity.

For the latter, the appetite is formidable. By 2030, Europe’s data centres could draw on the order of 168 terawatt-hours a year — roughly half Italy’s and about a third of Germany’s electricity consumption — pushing the sector toward 5% of European Union demand.

Sensing both the scale of this demand and the political stakes, Commission President Ursula von der Leyen used the 2025 State of the Union to set energy and artificial intelligence in lockstep. Europe’s claim to digital leadership, the President argued, rests not only on innovation but also on securing clean, reliable, and affordable power. The two must advance together — or not at all.

Yet that vision collides with Europe’s energy reality. By late 2024, industrial users were paying around EUR 190 per megawatt-hour – among the highest in the advanced economies, where U.S. and South Korean industrial customers averaged about EUR 75 and EUR 125 per megawatt-hour, respectively.

Industrial bills climb far above wholesale rates. They include not only the cost of energy, but also taxes, grid fees, and the price of hedging against volatility. Wholesale power – the price traded on short-term markets – averaged EUR 85 per megawatt-hour in early 2025. Even that was 30% higher than the year before.

The premium reflects more than market volatility. It is partly the price of Europe’s pivot away from Russian gas, compounded by rising carbon costs, tight network capacity, and the risk premia built into long-term contracts after years of energy shocks.

And the strain will not stay abstract. In Ireland, for instance, regulators warn that households in districts dense with data centres could see energy bills rise by 8 to 21% over five years, as grids are reinforced to meet surging demand.

The largest operators, however, lock in contracts that often insulate them from those very costs. This means families pay more while the biggest players pay less. Every extra terawatt-hour, in other words, is not just an entry on a balance sheet. It is a political fact felt in household budgets.

Such effects vary by region, but they highlight how quickly public perception can turn when infrastructure costs are unevenly shared. For Europe as a whole, the challenge is not whether to build these facilities but how to balance their local pressures with collective gain. That tension will shape Europe’s digital build-out in the years to come.

Europe’s AI drive is not about to slow, nor should it. There is no single answer to the energy burden of data centres: grids, fuels, and politics differ across regions.

Yet one pathway is distinctively European. Almost every kilowatt consumed by the data centres emerges as heat; in a continent laced with district-heating networks, that by-product can be captured and reused rather than wasted, turning a liability into a dividend.

What is missing is not demonstration projects; a pan-European framework that can make ‘heat reuse’ the default rather than the exception is what turns scattered pilots into established practice.

From pilots to patchwork

Heat reuse begins with a simple fact of physics: almost all the electricity consumed in an AI data centre is released again as heat. A data centre with a one-megawatt electrical load produces roughly 8.7 gigawatt-hours of recoverable thermal energy each year — enough to warm 600-700 homes in a northern European climate.

Yet most of that warmth still drifts into the air or seeps into nearby water. Capturing it requires no exotic engineering: pipes channel the flow, pumps lift the temperature to 60-70 °C, and storage tanks balance daily peaks.

Across Europe, the plumbing of these systems is evolving. Smart thermal networks – district-heating systems that operate at lower temperatures and use large heat pumps, thermal storage, and digital controls to capture and distribute waste heat more efficiently – are now making this process easier. These new systems can absorb heat more effectively and help balance supply and demand across entire districts, expanding the range of viable sources and simplifying the integration of data centre heat at scale.

The result can be channelled into district-heating grids as if the data centre were just another supplier. Europe is uniquely advantaged: nearly 19,000 district-heating networks already serve around 80 million people, providing a ready-made socket into which the waste heat from data centres — still too often left unused — can be plugged.

The potential is anything but marginal. Just outside Helsinki, Microsoft and Fortum are building two AI data centres that will cover about 40% of the region’s heating demand, reusing nearly three-quarters of the heat they generate.

Stockholm has gone further, creating an open market. Its ‘Open District Heating’ scheme already delivers more than 100 gigawatt-hours annually — enough for 30,000 apartments – under a transparent tariff that treats server heat as a commodity.

Paris and Amsterdam are experimenting too. In Paris, Telehouse’s TH2 and Equinix’s PA10 data centres are being connected to municipal heating schemes, while in Amsterdam, Equinix AM3 and AM5 already feed waste heat into the city’s district-heating network. Yet these examples come from regions with dense grids, strong utilities, and established heating demand; elsewhere in Europe, conditions for recovery at this scale are far less favourable.

That said, the evidence points one way: what was once dismissed as waste exhaust can instead become a civic asset, capable of heating tens of thousands of homes, cutting carbon by hundreds of thousands of tonnes, and anchoring entire cities in a more resilient energy system; beyond efficiency, these projects point to a subtler form of energy security – reducing reliance on imported gas for heating and linking digital infrastructure to Europe’s own energy flows.

But for every celebrated case, dozens of facilities across Europe still vent their heat. The barrier is not physics; it is governance. Permits drag into bespoke negotiations, exemptions erode obligations, and utilities hesitate when costs fall on them while benefits flow to households. The result is a striking gap between ambition and delivery. Recent Commission analysis finds that only about 1.9% of data-centre heat is actually reused in the EU today – evidence that law alone does not deliver heat.

Europe has not been idle. As part of the Fit for 55 package, the revised Energy Efficiency Directive requires any data centre with more than 500 kilowatts of computing load to report its performance to a central EU database. A delegated regulation adopted in 2024 set out the details, creating a common performance scheme that includes the Energy Reuse Factor as one of the new sustainability indicators for data centres.

The first filings were due in September 2024, with annual updates required each May thereafter. Facilities with a rated input above one megawatt face an even stricter obligation: they must make their waste heat available unless they can demonstrate that reuse is technically or economically infeasible.

On paper, these measures deliver both transparency and obligation. In practice, they have produced more paperwork than pipelines. Thousands of data centres now file reports, including the new generation of AI facilities whose energy demand is rising fastest; yet feasibility is unevenly interpreted, enforcement is patchy, and results remain largely invisible to the public. For AI data centres — few in number but rapidly expanding and far more power-hungry — this lack of consistent implementation risks turning flagship projects into flashpoints.

Germany has moved first, legislating a binding benchmark for new sites: 10% of waste heat must be reused from 2026, rising to 15% in 2027 and 20% in 2028. France, Spain, and Italy have stopped short of quotas, relying on case-by-case reviews that weigh feasibility against cost.

The Nordics enjoy dense heating grids and have no shortage of technical potential, but their schemes remain voluntary. Ireland, constrained by its fragile grid, has authorised only small municipal pilots. The result is not convergence but fragmentation: a handful of leaders, a long tail of late movers, and little sense of a common European benchmark.

The incentives misalign. Households absorb higher grid costs, operators shield themselves through long-term contracts, and utilities hesitate when their investments deliver benefits they cannot capture. Without an EU-wide framework, waste heat risks becoming the emblem of private gain and public strain in an era when AI infrastructure itself is emerging as critical infrastructure, on par with ports, railways, and energy grids.

Europe does not need a new agency or another billion-euro subsidy to make heat reuse routine. The networks exist, the projects have proven themselves, and the physics is immutable. What Europe lacks is predictability, simplicity, and visibility. Turning pilots into pipelines requires translating its legal obligation into real delivery.

From paper to pipelines

Article 26 of the Energy Efficiency Directive already obliges data centres to reuse waste heat unless reuse is technically or economically infeasible. To give that duty substance, Delegated Regulation 2024/1364 established a common performance scheme for large facilities (e.g., data centres), introducing standard indicators such as the Energy Reuse Factor. Yet law alone does not deliver heat. Rules on paper need mechanisms in practice.

Bridging that gap requires practical steps that turn obligation into delivery. Their implementation will depend on local planning cycles, grid capacity, and the willingness of utilities to invest — but the aim is to make the process predictable rather than uniform.

The first step is to require data centres to connect to district-heating networks wherever feasible. Article 26 already creates the obligation, but in practice, it is too often treated as optional. A standard permit clause should make connection the default wherever proximity and computing capacity align. Across the EU, roughly 1,000 large data centres lie within 2 to 3 kilometres of a district-heating network. Together they represent more than 75 terawatt-hours of recoverable thermal energy each year — enough to heat several million homes. AI data centres are still few in number, but their rapid growth makes default connection urgent.

To avoid blunt obligations, the rule should apply only where three clear conditions are met: a district-heating network exists or is already planned; the facility can deliver reliable, year-round heat; and the tariff is competitive with the utility’s most cost-effective low-carbon source. Germany points the way. Its Energy Efficiency Act sets a binding rule of 10% reuse in 2026 and 20% in 2028. A European clause could turn such national experiments into common practice across the Union.

The second step is to make contracts simple, predictable, and fast. At present, every project requires months of bespoke legal drafting, which raises transaction costs, slows delivery, and discourages investment. What is missing is a dedicated contractual instrument: a Heat-Reuse Purchase Agreement (HRPA). Conceived as the thermal counterpart to the power-purchase agreements that scaled renewables, an HRPA would replace ad hoc negotiations with a concise, standard template designed to make heat reuse replicable.

The HRPA would standardise on a small set of standard terms. These include the temperature at which heat is delivered, the metering and verification standards used, the number of guaranteed supply hours, and the tariff structure. By codifying these parameters, the agreement reduces legal uncertainty and provides a shared reference point for both utilities and operators.

For operators, the contract changes the cost structure. Instead of treating cooling in AI data centres as a permanent liability, the HRPA establishes a modest but reliable revenue stream. For utilities, it secures a stable source of low-carbon heat that can be integrated into medium- and long-term network planning. This dual benefit addresses the coordination problem that has so far limited uptake.

Transparency strengthens these effects. Publishing contracts with limited redactions would provide accountability, reassure local stakeholders, and demonstrate the scale of the benefits without revealing commercially sensitive details. In this way, HRPAs would bring the same standardisation logic that enabled power-purchase agreements to waste heat, transforming one-off pilot arrangements into repeatable and bankable practice.

Stockholm illustrates the value of contractual clarity. Since 2014, its Open District Heating programme has allowed data centres to sell surplus heat into the city grid under predefined conditions. By publishing contract terms in advance and treating server heat as a commodity, the scheme created a functioning market with a transparent tariff. The result is more than 100 gigawatt-hours of recovered heat each year and a system that rewards reliability rather than bespoke negotiation.

The third step is to build the bridge between servers and city grids. Many projects stall at the point where raw heat from servers must be converted into a usable supply. This requires specific equipment: an industrial pump to raise outlet temperatures to grid standard (i.e., 70-75 °C), a buffer tank to store and smooth daily peaks, and an exchange station to connect with the district-heating network. These components cost little compared with billion-euro AI data centres yet are decisive in moving from potential to delivery.

That gap – the modest equipment needed to connect servers to city grids – is small but critical. In Odense, Denmark, Meta’s data centre now provides heat to about 11,000 homes, enabled by large ammonia heat pumps that raise outlet temperatures from about 27 °C to grid standard. In Sweden, buffer tanks — large insulated water vessels that store heat and smooth demand swings — allow recovery to continue even when daily consumption fluctuates. Such equipment typically accounts for less than 5% of a data centre’s total investment, but without it, the recovered heat never reaches the grid.

EU support should therefore be narrowly targeted at this bottleneck, financing only the connection equipment and awarding grants competitively. Projects should be assessed against two clear criteria: the amount of usable heat delivered per euro of public spending, and the speed with which the first supply reaches the grid. Public funding should play a catalytic role, covering only a minority share of costs while leveraging private capital.

Existing EU instruments can cover this gap: the Innovation Fund (already financing large heat-pump and district-heating decarbonisation projects); the Social Climate Fund (to connect schools, hospitals, and social housing, lowering bills for vulnerable users); the Modernisation Fund (available in 13 Member States, for network upgrades and renewable heat); and Cohesion Policy funds such as the European Regional Development Fund, which support municipal-scale connection equipment and grid extensions.

The final step is to make the results visible. Transparency without outcomes quickly breeds cynicism, yet Europe already has the data in hand. Every May, thousands of large data centres — each drawing more than 500 kilowatts of computing power — now file annual performance reports. That information should be translated into a public scoreboard that citizens can easily understand.

The ledger should track three indicators: heat delivered to the network, the steadiness of supply, and net carbon dioxide avoided once the electricity consumed by heat pumps is taken into account. Existing projects give a sense of scale: the LUMI supercomputer prevents 12,400 tonnes of carbon dioxide emissions each year, while Fortum’s new data centres are designed to avoid roughly 400,000.

To strengthen accountability, the European Commission’s Directorate-General for Energy (DG ENER) could host the scoreboard, publishing results as open data and verifying them through independent audits. Priority projects should focus on schools, hospitals, and social housing, helping to cut energy costs for essential services and making the benefits visible to households.

Publishing comparable figures, such as tonnes of carbon dioxide avoided, households supplied, and hours of steady heat delivered, would do more than provide information. It would spark competition among cities, push utilities to climb the rankings, and allow citizens to see in plain numbers the value captured from the AI build-out.

None of this requires new institutions or bureaucracies. The infrastructure is already in place: the Energy Efficiency Directive obliges large data centres to report their performance within a central EU database. Converting that system into a public scoreboard would simply extend Europe’s transparency rulebook, consistent with the Aarhus Convention on access to environmental information and the EU’s Open Data Directive.

With open data in machine-readable form, clear metrics aligned with the Energy Reuse Factor, and independent audits, what is now an invisible by-product could be transformed into a visible civic dividend. That dividend can be measured in megawatt-hours of heat delivered, hours of steady supply, and tonnes of carbon dioxide avoided — and experienced by citizens as warmer homes, lower bills, and tangible returns from the AI build-out. In time, it would also strengthen Europe’s strategic autonomy; natural gas still supplies about 30% of household energy, and the EU remains roughly 86% dependent on gas imports, with Russia still around 12% of EU gas imports in mid-2025 — so every unit of recovered heat reduces exposure to imported gas.

New heat reuse networks around AI

The four steps harmonise heat reuse across Europe’s existing district-heating networks. They build directly on what already exists: under the revised Energy Efficiency Directive, large data centres must report their performance and make waste heat available unless reuse is technically or economically infeasible. The challenge is no longer the principle but the practice — turning a legal obligation into a functioning system of connections, contracts, and accountability.

The harder question is what to do when the next generation of AI data centres rises far from existing grids. Europe should not treat its heating map as frozen. Where density and demand align — around towns, hospitals, or university hubs — AI campuses can anchor new loops that extend district heating into nearby areas. Yet this potential is not universal.

In Southern Europe, where heat demand is lower and cooling seasons are longer, reuse options are limited. In industrial zones, competition for waste heat can constrain what reaches municipal networks. Even where infrastructure exists, distance, temperature, and timing determine viability. The aim is not to turn every data centre into a civic anchor but to ensure that where reuse is feasible, public benefit is captured transparently and private advantage remains accountable.

As Mario Draghi argued in his 2024 competitiveness report, Europe’s growth now depends on faster permitting, single-market scale, and investment in enabling infrastructure. Extending heat networks around AI anchors delivers on all three: it links digital expansion to lower bills and visible civic returns, ensuring that AI growth strengthens rather than strains public consent. The spread of smart thermal networks across Europe, highlighted in the 2024 European Commission analysis, confirms that the physical conditions for large-scale heat reuse already exist; what remains is to match them with the right governance and investment framework.

Finance and governance must follow this logic: according to a Joint Research Centre report, successful heat reuse depends on proximity, temperature lift, and network design, hence the case for designating heat-reuse priority zones and publishing connection maps to guide investment. Public money should cover early pipework where it displaces fossil use, while operators finance their own facilities and commit to long-term contracts. Developers should be required to make data centres connection-ready, but the decision to extend networks must remain public, led by municipalities and guided by the EU’s heat-reuse framework under the Energy Efficiency Directive.

Beyond financing and network design, large data centres also raise questions about water use, land footprint, and community impact. These must remain part of the permit process: the aim is not to offset such pressures with warmth, but to make visible what is already there – and to ensure that when data centres do operate, their benefits are shared as clearly as their costs.

This layered model keeps ambition credible. The EU provides the rules and transparency system; Member States embed heat planning in law; cities judge where new grids make sense; and developers ensure connection-readiness. If Europe gets this right, AI data centres will not only consume energy — they will return part of it, turning waste heat into the next civic dividend.