AI vs. Net-Zero Emissions: The Physical Energy Limits of Europe's Digital Expansion

Introduction: The Collision of Two Exponential Curves

Europe is currently navigating a profound structural collision between the rapid acceleration of artificial intelligence infrastructure and the physical realities of its energy transition. In mid-2026, the European Data Centre Association articulated a stark warning that captured the attention of policymakers worldwide: Europe must prioritize its artificial intelligence infrastructure over immediate climate ambitions, or risk ceding technological sovereignty to international competitors such as China¹. The core of this argument, advanced by industry leaders, rests on the assertion that current grid expansion timelines, renewable energy deployments, and next-generation nuclear projects are progressing too slowly to support the unprecedented power demands of advanced computational models². Consequently, industry lobbyists have controversially suggested a temporary return to carbon-emitting gas-fired power generation as a bridging solution to power the European Union's ambitious digital goals¹.

This tension highlights a fundamental trilemma for European policymakers. They must balance the imperative for economic growth and digital sovereignty against legally binding net-zero climate commitments and the physical constraints of legacy energy grids⁴. The European Commission’s recent European Technological Sovereignty Package, which includes the Cloud and AI Development Act, seeks to triple the continent's data center capacity by the early 2030s². However, energy officials emphasize that companies wishing to profit from the artificial intelligence boom must actively integrate into the clean energy transition through renewable procurement and waste heat recovery⁸.

The resolution to this trilemma will not be found in simple policy declarations, but rather in a complex synthesis of advanced thermodynamics, high-density hardware engineering, grid-level power electronics, and adaptive regulatory frameworks. This analysis explores the intersecting domains of artificial intelligence energy demand, cooling technologies, grid stability, and the emerging pivot toward localized nuclear baseload generation, providing a comprehensive overview of how Europe might navigate the physical limits of the digital age.

The Quantitative Landscape of Artificial Intelligence Energy Demand

To understand the friction between digital infrastructure and climate goals, one must first examine the scale and specific nature of artificial intelligence-driven electricity consumption. Unlike traditional data centers, which primarily manage data storage, web hosting, and general-purpose cloud computing, artificial intelligence data centers are architected around dense clusters of specialized accelerators, such as Graphics Processing Units, designed for continuous, high-performance parallel processing⁹.

The International Energy Agency and the Kiel Institute estimate that global data center electricity consumption reached approximately 415 terawatt-hours in 2024, representing roughly 1.5 percent of total global electricity use¹⁰. Driven predominantly by the rapid adoption of generative artificial intelligence, this demand is projected to more than double by 2030¹⁰. In Europe, projections suggest that data center consumption could reach between 98.5 and 168 terawatt-hours by the end of the decade⁴. To contextualize this volume, 168 terawatt-hours is roughly equivalent to the entire annual electricity demand of a nation like Poland, representing up to 5 percent of overall European Union consumption⁴.

The disparity in energy consumption between traditional internet infrastructure and artificial intelligence inference is stark. A single query on an advanced generative model requires nearly ten times the electrical energy of a conventional search engine query¹¹. Furthermore, the training of foundational models requires thousands of processors operating continuously for months, consuming gigawatt-hours of electricity in a single cycle¹². For example, the energy consumption for training successive generations of large language models has risen dramatically, with recent iterations consuming over 50 gigawatt-hours—equivalent to a significant fraction of a major metropolitan area's annual electricity use¹¹.

Scenario modeling for European grid stability underscores that electricity demand is highly dependent on infrastructure expansion, regulatory design, and energy efficiency scaling¹⁵. Under constrained development models, demand might be limited, while unchecked expansion without efficiency improvements could drive demand to extreme highs¹⁵.

Demand Scenario	2030 Projected EU Consumption	System Characteristics	Primary Constraints
Constrained Growth	45 - 80 TWh	Rigid grid moratoria; delayed interconnections; strict fossil-fuel bans.	Grid bottlenecks; high regional imbalances; administrative friction.
Sustainable Corridor	90 - 115 TWh	Coordinated regulatory design; adaptive capacity planning; high efficiency gains.	Requires proactive grid investments and flexible siting ahead of demand.
Unchecked Expansion	145 - 168 TWh	Rapid hardware deployment outstripping renewable integration; reliance on bridging fossil fuels.	High carbon intensity; severe stress on local reserve margins; public pushback.

Table 1: Projected 2030 European Artificial Intelligence Data Center Electricity Demand Scenarios based on macroeconomic and grid constraint modeling⁴.

A critical insight emerges from this data: the challenge is not merely the aggregate volume of electricity required, but its spatial and temporal concentration. Data centers tend to cluster in established hubs—such as the Frankfurt, London, Amsterdam, Paris, and Dublin corridors—where local transmission networks are already operating near their adequacy limits¹⁵. In jurisdictions like Ireland, data centers consumed an unprecedented 22 percent of the nation's total metered electricity in 2024, surpassing the consumption of all urban households combined²⁰. This extreme concentration has led to severe grid strain, prompting regulatory interventions and de facto moratoria on new grid connections in the Greater Dublin area²³.

The Hardware Bottleneck: Thermodynamics of Advanced Cooling

As computational workloads intensify, the physical architecture of the data center must evolve. The limiting factor in advanced computation is no longer simply semiconductor logic, but thermal management. Traditional central processing unit servers operate at power densities of roughly 7 to 10 kilowatts per server rack¹¹. In contrast, state-of-the-art servers containing multiple high-performance graphics processors routinely demand between 30 and 100 kilowatts per rack, with next-generation silicon pushing toward 120 kilowatts¹².

When electrical energy performs computational work, nearly the entirety of that energy is converted into low-grade thermal energy due to electrical resistance within the microscopic transistors²⁷. Managing this extreme heat flux requires a fundamental shift away from traditional forced-air cooling. Air has a relatively low specific heat capacity and poor thermal conductivity, meaning it lacks the physical ability to efficiently dissipate the heat generated by dense processor clusters without exorbitant energy penalties from high-velocity fans and massive chiller plants²⁸.

The industry standard for measuring this efficiency is Power Usage Effectiveness. This metric is determined by dividing the total energy consumed by the data center facility by the energy consumed strictly by the computing equipment²⁵. A lower value, approaching the theoretical ideal of exactly one, indicates high efficiency, meaning zero energy is wasted on overhead operations. Traditional air-cooled facilities average a Power Usage Effectiveness of 1.5 to 1.6, meaning that for every kilowatt of computing power, an additional 500 to 600 watts are expended on cooling and auxiliary systems²⁵.

To accommodate high-density workloads and reduce auxiliary power penalties, the industry is rapidly adopting advanced liquid cooling technologies. Liquid coolants, specifically water or engineered dielectric fluids, possess a heat capacity orders of magnitude greater than air, allowing them to absorb and transport thermal energy far more efficiently³².

Direct-to-Chip and Microfluidic Innovations

Direct-to-chip cooling systems circulate a liquid coolant through cold plates mounted directly atop the processing units. By bypassing the thermal resistance of bulk air, direct-to-chip systems can capture the majority of the heat generated by a server at the source. This approach can drop a facility's Power Usage Effectiveness to the 1.10 to 1.25 range while enabling the stable operation of processors exceeding 1,000 watts²⁸.

However, as thermal design power continues to scale, even standard cold plates encounter limitations due to the thermal interface materials—such as thermal pastes and metal heat spreaders—separating the silicon die from the cooling block²⁸. This bottleneck has driven significant research into microfluidic cooling. By etching microscopic channels directly into the back of the silicon chip or the underlying interposer layer, coolant can be delivered directly to localized hotspots within the processor architecture itself³⁴.

Recent prototypes developed by hyperscale cloud providers, which integrate artificial intelligence algorithms to optimize bio-inspired, vein-like channel designs, have demonstrated heat removal capabilities up to three times greater than traditional cold plates³⁴. This efficiency reduces the maximum temperature rise within the silicon by over 60 percent, allowing operators to run servers hotter and faster—a practice known as overclocking—without risking thermal degradation³⁴. While microfluidics promises to radically lower thermal resistance, widespread deployment faces severe hurdles. It requires custom-built silicon, adding immense complexity to semiconductor fabrication, and introduces contamination risks, as microscopic impurities can easily clog the micrometer-scale fluid pathways³⁵.

Two-Phase Immersion Cooling

For ultra-high-density deployments, single-phase and two-phase immersion cooling represent the thermodynamic frontier. In these systems, entire server chassis are submerged in a thermally conductive, electrically non-conductive dielectric fluid³⁷. In two-phase immersion, the engineered fluid is designed with a boiling point closely matching the optimal operating temperature of the processors. The fluid boils upon contact with the hot components, utilizing the latent heat of vaporization to absorb massive amounts of thermal energy. The resulting vapor rises, condenses on a secondary water-cooled heat exchanger at the top of the tank, and drips back into the bath³⁷. Immersion systems eliminate the need for server fans entirely and rely on passive thermodynamic cycles, reducing the Power Usage Effectiveness to near-theoretical limits of 1.02 to 1.05³³.

Cooling Technology	Typical Efficiency Metric (PUE)	Target Rack Density	Primary Mechanism of Heat Removal
Traditional Air Cooling	1.50 - 1.80	< 15 kW	Forced convective air flow over finned metal heat sinks.
Rear-Door Heat Exchanger	1.25 - 1.45	15 - 30 kW	Liquid-cooled coils capture exhausted air heat at the rack rear.
Direct-to-Chip Liquid	1.10 - 1.30	30 - 100+ kW	Coolant circulated through cold plates mounted directly on processors.
Microfluidic Cooling	< 1.15 (Projected)	> 100 kW	Micro-channels etched directly into silicon packaging for internal flow.
Two-Phase Immersion	1.02 - 1.05	50 - 100+ kW	Submersion utilizing the latent heat of vaporization of dielectric fluids.

Table 2: Comparison of Data Center Cooling Technologies, Efficiency Metrics, and Thermal Mechanisms¹¹.

Waste Heat Recovery: Synergies and Thermodynamic Constraints

The mandate to recover waste heat from data centers presents both a massive societal opportunity and a complex thermodynamic engineering challenge³⁹. Theoretically, data centers represent a vast, untapped thermal resource; capturing and redirecting this heat could dramatically lower municipal reliance on fossil fuels for residential heating⁴². Recognizing this, the European Union's updated Energy Efficiency Directive obligates data center operators to assess and implement waste heat utilization where economically and technically feasible³⁹.

Member states are interpreting these directives with varying degrees of stringency. Germany’s Energy Efficiency Act provides a rigorous framework, mandating that new data centers commencing operations from July 2026 must achieve an Energy Reuse Factor of at least 10 percent, scaling to 20 percent by 2028²⁶. This means a significant proportion of the total energy input must be captured and exported as thermal energy.

The practical implementation of heat reuse, however, is governed by the laws of thermodynamics—specifically the temperature of the available heat source and the requirements of the heat sink. Traditional air-cooled data centers emit exhaust air at relatively low temperatures, typically between 25 and 35 degrees Celsius²⁶. Conversely, most existing municipal district heating networks in Europe require supply temperatures between 70 and 90 degrees Celsius to effectively heat residential blocks²⁶.

Bridging this thermal gap requires the deployment of industrial-scale heat pumps. A heat pump consumes electrical energy to drive a compressor, elevating the temperature of the recovered heat to a usable level⁴². The efficiency of this process is measured by the Coefficient of Performance, which describes the ratio of useful heating provided to the electrical energy required to drive the pump⁴⁴. This coefficient is highly dependent on the "temperature lift"—the difference between the source temperature and the required output temperature. A smaller temperature lift results in a higher Coefficient of Performance, meaning less electrical work is required to upgrade the heat⁴⁴.

The advent of liquid cooling inadvertently synergizes with waste heat recovery mandates. Liquid cooling loops, particularly those running close to the thermal limits of the silicon, generate return coolant temperatures of 50 to 60 degrees Celsius²⁷. This higher-grade waste heat significantly reduces the necessary temperature lift required by the heat pump. Consequently, the Coefficient of Performance improves dramatically, rendering the economic and environmental case for heat recovery much more viable²⁷. In some modern low-temperature district heating networks, 60-degree coolant can even be utilized directly without the need for an intermediary heat pump⁴².

Despite technological feasibility, systemic integration remains a profound barrier. Successful heat recovery requires spatial proximity between the data center and the district heating network, massive capital investments in insulated subterranean pipework, and synchronization between the constant thermal output of the data center and the highly seasonal heating demands of municipal populations²⁶.

Grid Stability: Virtual Inertia and Frequency Response

Beyond baseline energy consumption, the integration of hyperscale infrastructure introduces profound challenges to electrical grid stability. Advanced computing clusters exhibit unique behavioral characteristics on the power grid. Unlike traditional industrial loads, which draw power at relatively consistent rates, data centers running massive machine learning models experience extreme power variations. When a massive training run is initiated, the load can ramp from near zero to hundreds of megawatts in a matter of seconds⁴⁷.

This rapid load ramping fundamentally stresses the frequency control mechanisms of the power grid. In a traditional power system, alternating current frequency is stabilized by the physical, mechanical inertia of massive rotating turbines inside fossil-fuel or nuclear power plants⁴⁸. If a sudden imbalance occurs between supply and demand, the kinetic energy stored in these spinning masses acts as a buffer, slowing the rate at which the grid frequency drops and allowing operators time to adjust generation⁴⁸. As grids decarbonize and transition toward inverter-based renewable energy sources like wind and solar, the aggregate mechanical inertia of the system decreases, rendering the grid highly susceptible to rapid frequency deviations caused by sudden load spikes⁴⁸.

Data centers interact with the grid exclusively through power electronic converters, which lack physical inertia and exhibit instantaneous response dynamics¹¹. If unmanaged, the synchronous activation of high-density clusters could induce severe under-frequency events, potentially triggering cascading blackouts.

To mitigate these risks, researchers and grid operators are exploring advanced flexible active power control systems. By utilizing the data center's own uninterruptible power supply batteries, or by dynamically adjusting the power draw of the servers themselves, operators can simulate mechanical inertia—a concept known as virtual inertia control⁴⁸. Furthermore, the computational workloads offer a unique form of demand-side flexibility. By slightly curtailing processing speeds, pausing non-time-sensitive batch training tasks, or introducing dummy workloads in response to grid frequency signals, data centers can function as virtual power plants²⁴. This allows them to provide primary frequency regulation services to the grid, actively balancing supply and demand rather than merely acting as passive consumers²⁴.

Regulatory Frameworks: Sovereignty vs. Sustainability

The friction between rapid digitalization and energy constraints has catalyzed a wave of regulatory action across the European Union. In June 2026, the European Commission introduced the European Technological Sovereignty Package, a sweeping legislative framework designed to reduce the continent's structural reliance on foreign technology providers⁶. At its core is the Cloud and AI Development Act, which mandates stringent sovereignty requirements for public-sector data and critical infrastructure⁶.

The Cloud and AI Development Act establishes a multi-tiered assurance framework to evaluate the sovereign integrity of digital service providers⁶.

Assurance Level	Core Regulatory Requirements	Typical Applicability
Level 1	Data residency strictly within European Union borders.	General public administration; non-sensitive workloads.
Level 2	Independence from third-country legal jurisdictions; software supply chain transparency.	Sensitive administrative data; healthcare and financial records.
Level 3	European Union ownership and control; personnel citizenship requirements.	National security-adjacent functions; defense procurement.
Level 4	Full transparency over supply chain; absolute immunity from third-country interference.	Critical sovereignty functions; classified equivalent workloads.

Table 3: The Cloud and AI Development Act Union Assurance Framework⁶.

At the highest tiers, cloud providers must demonstrate full European Union ownership, control, and immunity from third-country legal jurisdictions, effectively creating structural barriers for foreign-domiciled hyperscale providers operating sensitive workloads⁶. While this legislation aims to triple European data center capacity and secure digital independence, it simultaneously amplifies the energy trilemma by demanding that this massive build-out occurs strictly within the geographical and energetic confines of the European grid⁷.

Parallel to the sovereignty push are aggressive environmental mandates. The recast Energy Efficiency Directive requires all data centers with an installed IT power demand of 500 kilowatts or greater to publicly report comprehensive sustainability metrics to a European database⁴⁰. This includes granular data on power utilization, temperature set points, water usage, and the share of renewable energy consumed⁴⁰.

A deeper analysis of these emerging regulations reveals a systemic vulnerability. Rigid mandates, such as fixed build-out moratoria or absolute efficiency targets, often fail to account for real-time grid conditions¹⁵. Adaptive regulatory frameworks, which condition project approvals on dynamic local grid adequacy rather than blanket restrictions, have been shown to maintain grid stability while permitting significantly higher connected loads¹⁶. The tension lies in the realization that while the European Union desires rapid, sovereign technological expansion, its regulatory apparatus frequently imposes compliance timelines and permitting cycles that operate on a much slower temporal scale than digital capital deployment¹⁵.

The Nuclear Renaissance and Financial Shifts

Recognizing the fundamental limitations of intermittent renewable sources—such as wind and solar—in powering uninterruptible, 24/7 computational infrastructure, the technology sector is executing a strategic pivot toward nuclear power¹⁹. Advanced foundational models require absolute baseload stability; even minor voltage fluctuations or micro-outages can corrupt weeks of continuous training data, incurring massive financial and temporal losses¹². To achieve true energy sovereignty and decouple their operations from constrained public utility grids, hyperscale cloud providers are transitioning from passive energy consumers to active co-developers of dedicated nuclear generation¹⁹.

The Promise of Small Modular Reactors

The primary vehicle for this nuclear pivot is the Small Modular Reactor. Defined generally as reactors producing less than 300 megawatts of electrical output, these systems are designed to be factory-fabricated in standardized modules and assembled on-site⁵⁷. This modularity theoretically circumvents the prohibitive capital costs, bespoke engineering challenges, and decade-long construction delays associated with legacy gigawatt-scale light water reactors⁵⁷. Furthermore, advanced Generation IV designs, such as high-temperature gas reactors and molten salt reactors, offer inherent passive safety features and higher thermal efficiencies⁵⁹.

The European Commission has officially endorsed this pathway, launching the European Industrial Alliance on Small Modular Reactors and targeting the deployment of the first commercial units by the early 2030s⁵⁷. Projections incorporated into the Commission's Nuclear Illustrative Programme suggest that modular reactor capacity in the European Union could reach up to 53 gigawatts by 2050⁵⁷.

Despite the enthusiasm, significant hurdles remain. Many reactor concepts exist only in the early design phases and lack regulatory approval within the European Union⁵⁸. The promise of cost reduction relies entirely on achieving economies of multiples—the mass production of standardized units—a manufacturing paradigm the civil nuclear industry has yet to prove it can achieve⁵⁸. Nonetheless, the unprecedented electricity demand from the technology sector has fundamentally altered the financing landscape, providing the guaranteed long-term power purchase agreements necessary to attract private capital to nuclear startups¹⁹.

The World Bank Policy Reversal

The macroeconomic landscape for nuclear energy experienced a seismic shift in June 2025, when the World Bank formally ended its decades-long ban on financing nuclear power projects⁶². The institution had not financed a nuclear project since supporting Italy's first plant in 1959⁶². Recognizing that global decarbonization and developmental goals cannot be met exclusively through intermittent renewables, the institution announced partnerships with the International Atomic Energy Agency to support both the life extension of existing reactors and the deployment of modular technologies⁶².

This policy reversal acts as a powerful market signal. Historically, the International Finance Corporation—the private-sector arm of the World Bank—maintained an exclusion list prohibiting investment in radioactive materials, which served as a de facto standard for global commercial banks⁶⁵. By removing the institutional stigma against nuclear financing, the World Bank effectively de-risks the sector, paving the way for regional institutions like the Asian Development Bank and vast pools of private capital to re-enter the nuclear space⁶⁵. For the digital infrastructure industry, this unlocks the potential to build high-density computing hubs in emerging markets, leapfrogging fossil-fuel dependency and establishing clean baseload power globally⁶⁴.

Geopolitical Repositioning: Redrawing the Digital Map

Within Europe, the pursuit of reliable, low-carbon electricity is reshaping the geographic distribution of digital infrastructure. Legacy hubs are facing severe grid saturation and regulatory pushback, causing development to migrate toward regions with structural energy advantages¹⁵.

France, generating over 60 percent of its electricity from a legacy fleet of 56 nuclear reactors, is aggressively positioning itself as the epicenter of European artificial intelligence⁶⁶. By offering guaranteed access to abundant, low-carbon baseload power, the French government has attracted staggering investments. In 2026, SoftBank Group committed 45 billion euros to construct three next-generation data centers in the Hauts-de-France region, designed to draw 3.1 gigawatts of power directly from the national grid⁶⁶. This initiative aims to establish an integrated technology valley, explicitly leveraging nuclear reliability to draw global capital away from constrained regions⁶⁶.

Similarly, Portugal is emerging as a critical nexus for digital infrastructure, capitalizing on its abundant solar and wind resources, simplified permitting processes, and strategic position as a landing point for trans-Atlantic subsea cables linking Europe, Africa, and the Americas⁶⁹. In late 2025, Microsoft announced a 10 billion dollar investment to construct a massive artificial intelligence complex in Sines, Portugal⁶⁹. Housing tens of thousands of advanced processors and powered entirely by renewable energy, the Sines facility exemplifies the hyperscale strategy: locating massive computational workloads at the precise intersection of high-capacity data transit and sovereign, renewable energy generation⁶⁹.

Conclusion

The warning issued by the European Data Centre Association—that Europe must choose between its digital ambitions and its climate goals—presents a false dichotomy born of the friction between rapid technological innovation and slow-moving public infrastructure¹. A regression to gas-fired baseload generation would fundamentally compromise the continent's net-zero trajectory, effectively exchanging climate security for short-term computational speed.

Conversely, strict regulatory moratoria and rigid efficiency mandates that fail to account for the physical realities of thermal management and grid dynamics will inevitably stifle the digital economy. Such stagnation would result in a loss of technological sovereignty, forcing Europe to rely on international rivals for foundational digital capabilities⁴.

The path forward requires a paradigm shift in how digital infrastructure is conceptualized. Data centers can no longer be engineered as isolated electrical loads; they must be integrated as dynamic, responsive nodes within the broader energy ecosystem. The deployment of direct-to-chip and immersion cooling technologies enables unprecedented compute density while simultaneously upgrading waste heat to temperatures suitable for municipal district heating networks²⁷. Furthermore, through advanced power electronics and intelligent workload flexibility, computational clusters can provide critical frequency regulation and virtual inertia services to grids that are increasingly reliant on intermittent renewables⁴⁸.

The integration of advanced nuclear technologies, catalyzed by shifting global financial policies and the World Bank's historic reversal, offers a long-term solution to the baseload requirement, though near-term constraints demand the optimization of existing low-carbon grids⁵⁹. Ultimately, reconciling Europe's artificial intelligence ambitions with its climate mandates relies not on choosing one over the other, but on an unprecedented, cross-sectoral alignment of semiconductor engineering, grid-level thermodynamics, and adaptive policymaking.

Works cited

1. Post by @politico.eu - Bluesky, https://bsky.app/profile/politico.eu/post/3mokeyibnw22d

2. Europe must choose between AI and climate goals, data center lobby says - POLITICO Pro, https://subscriber.politicopro.com/article/eenews/2026/06/22/europe-must-choose-between-ai-and-climate-goals-data-center-lobby-says-00967296

3. Europe will have to compromise on environmental values to make progress in AI, experts say | EcoPolitic, https://ecopolitic.com.ua/en/news/europe-will-have-to-compromise-on-environmental-values-to-make-progress-in-ai-experts-say/

4. AI ambitions vs. energy reality: Europe faces a strategic gap - Kiel Institute, https://www.kielinstitut.de/publications/news/ai-ambitions-vs-energy-reality-europe-faces-a-strategic-gap/

5. Running into a debacle: The mismatch between the EU's AI ambition and its energy planning. - Kiel Institute, https://www.kielinstitut.de/publications/running-into-a-debacle-the-mismatch-between-the-eus-ai-ambition-and-its-energy-planning-19644/

6. EU Tech Sovereignty: Cloud Concentration Risk and the Compliance Cascade - Lab Space, https://labs.cloudsecurityalliance.org/research/eu-tech-sovereignty-cloud-ai-enterprise-risk-v1-0-csa-styled/

7. Proposal for the Cloud and AI Development Act (CADA) | Shaping Europe's digital future, https://digital-strategy.ec.europa.eu/en/library/proposal-cloud-and-ai-development-act-cada

8. EU tells Big Tech to align AI data centres with climate goals or stay away - TNW, https://thenextweb.com/news/eu-tells-big-tech-to-align-ai-data-centres-with-climate-goals-or-stay-away

9. Artificial Intelligence and Energy Overconsumption: Data Center Electricity Demand, Cooling Burdens, and Regional Sustainability Constraints - SCIRP, https://www.scirp.org/journal/paperinformation?paperid=150890

10. Energy demand from AI – Energy and AI – Analysis - IEA, https://www.iea.org/reports/energy-and-ai/energy-demand-from-ai

11. Electricity Demand and Grid Impacts of AI Data Centers: Challenges and Prospects - arXiv, https://arxiv.org/html/2509.07218v3

12. What Are the Power Requirements for AI Data Centers?, https://www.hanwhadatacenters.com/blog/what-are-the-power-requirements-for-ai-data-centers/

13. Global energy demands within the AI regulatory landscape | Brookings, https://www.brookings.edu/articles/global-energy-demands-within-the-ai-regulatory-landscape/

14. Running into a debacle: - Kiel Institute, https://www.kielinstitut.de/fileadmin/Dateiverwaltung/IfW-Publications/fis-import/89312b7a-273a-4ecb-8481-fa0576d8f06f-Running_into_a_debacle.pdf

15. AI & energy in Europe: What scenarios reveal about stability through 2030, https://blog.se.com/sustainability/2025/12/16/ai-energy-europe-scenarios-reveal-stability-2030/

16. AI and Energy in Europe - University of Twente Research Information, https://research.utwente.nl/en/publications/ai-and-energy-in-europe/

17. AI & Energy in Europe - University of Twente Research Information, https://research.utwente.nl/files/510803310/AI_Energy_in_Europe_Pubished_version.pdf

18. The data center boom: Navigating growth and sustainability - GRESB, https://www.gresb.com/the-data-center-boom-navigating-growth-and-sustainability/

19. Top 10 Nuclear Powered Data Center Projects in Europe in 2025, https://enkiai.com/data-center/top-10-nuclear-powered-data-center-projects-in-europe-in-2025/

20. Data Centres Metered Electricity Consumption 2024 - Central Statistics Office, https://www.cso.ie/en/releasesandpublications/ep/p-dcmec/datacentresmeteredelectricityconsumption2024/

21. Data centres consumed 22pc of Ireland's electricity in 2024 - Silicon Republic, https://www.siliconrepublic.com/enterprise/data-centres-cso-survey-2024-electricity

22. Key Findings Data Centres Metered Electricity Consumption 2024 - Central Statistics Office, https://www.cso.ie/en/releasesandpublications/ep/p-dcmec/datacentresmeteredelectricityconsumption2024/keyfindings/

23. Data Centre's in Ireland: The State of Play - IIEA, https://www.iiea.com/blog/data-centres-in-ireland-the-state-of-play

24. Written submission from Schneider Electric (DCU0060) - h t t p s : / / c o m m i t t e e s . p a r l i a m e n t . u k, https://committees.parliament.uk/writtenevidence/164831/pdf/

25. Data Center Rack Power Costs: A Condensed Analysis - Nlyte Software, https://www.nlyte.com/blog/data-center-rack-power-costs-a-condensed-analysis/

26. Why Germany's Energy Efficiency Act makes waste heat recovery a national priority - Blog, https://www.cundall.com/ideas/blog/why-germanys-energy-efficiency-act-makes-waste-heat-recovery-a-national-priority

27. Data Centre Waste Heat for Building Heating: A Comparative Energy Analysis in Italy - MDPI, https://www.mdpi.com/2071-1050/18/12/6061

28. Direct-to-Chip Cooling Implementation: Reducing PUE Below 1.2 - Introl, https://introl.com/blog/direct-to-chip-cooling-pue-below-12-implementation

29. 2.3 | Cooling Technologies - SUSTAINABLE DATA CENTERS ROADMAP, https://icef.go.jp/wp-content/themes/icef_new/pdf/roadmap/2025/08_CHAPTER%20%E2%85%A1%20%E2%80%93%203.%20COOLING%20TECHNOLOGIES.pdf

30. What Is Power Usage Effectiveness (PUE) in Data Centers? - Cadence Blogs, https://community.cadence.com/cadence_blogs_8/b/data-center/posts/what-is-power-usage-effectiveness-pue-in-data-centers

31. Energy consumption in data centers: air versus liquid cooling - Eaton, https://www.eaton.com/us/en-us/markets/data-centers/data-center-cooling/efficiency/energy-consumption-in-data-centers-air-versus-liquid-cooling.html

32. What is High-Density Computing and the Importance of Cooling? - Nautilus Data Technologies, https://nautilusdt.com/news-updates/what-is-high-density-computing/

33. Data Center Cooling 2026: Liquid Immersion vs Air — TCO & Performance, https://energy-solutions.co/articles/sub/data-center-cooling-liquid-immersion-vs-air

34. AI chips are getting hotter. A microfluidics breakthrough goes straight to the silicon to cool up to three times better. - Microsoft Source, https://news.microsoft.com/source/features/innovation/microfluidics-liquid-cooling-ai-chips/

35. Is Microfluidic Cooling Ready for Data Centers?, https://www.datacenterknowledge.com/cooling/is-microfluidics-ready-to-cool-the-next-generation-of-data-centers-

36. Microsoft tames intense chip heat with liquid cooling veins, designed by AI and inspired by biology - GeekWire, https://www.geekwire.com/2025/microsoft-tames-intense-chip-heat-with-ai-designed-liquid-cooling-veins-inspired-by-biology/

37. What are Data Center Cooling Systems? - Ansys, https://www.ansys.com/simulation-topics/what-are-data-center-cooling-systems

38. Liquid cooling | Netceed, https://netceed.com/wp-content/uploads/2025/06/Netceed_Liquid_Cooling_Whitepaper.pdf

39. Data center policies in the EU - Danfoss, https://www.danfoss.com/en/industries/buildings-commercial/shared/data-centers/data-center-policies-in-the-eu/

40. The EU's Energy Efficiency Directive and Its Impact on Datacenters - September 2025 - Covington & Burling LLP, https://www.cov.com/-/media/files/corporate/publications/2025/04/the-eus-energy-efficiency-directive-and-its-impact-on-datacenters.pdf

41. Data Center Compliance Guide: Laws & Regulations - PowerUP, https://www.powerup.at/knowledge/data-center/compliance-data-center/

42. Don't Waste The Waste Heat Opportunity - Data Center POST, https://datacenterpost.com/dont-waste-the-waste-heat-opportunity/

43. Using Waste Heat Generated from Data Centers: - ReImagine Appalachia, https://reimagineappalachia.org/wp-content/uploads/2026/02/Catching-Heat_Using-Waste-Heat-Generated-from-Data-Centers.pdf

44. Heat Pumps Move to the Centre of District Energy: How Electrified Heat Is Reshaping the Decarbonisation Agenda - HPT, https://heatpumpingtechnologies.org/heat-pumps-move-to-the-centre-of-district-energy-how-electrified-heat-is-reshaping-the-decarbonisation-agenda/

45. Data Center Heat Reuse: Regulatory Requirements, Typical Use Cases and Optimal Planning - Velasolaris, https://www.velasolaris.com/en/data-center-heat-reuse/

46. Three success factors for the future of German data centres - KPMG International, https://kpmg.com/de/en/insights/business-performance-and-resilience/three-success-factors-for-the-future-of-german-data-centres.html

47. AI Data Centers and Energy Regulation: When Traditional Frameworks Meet Unprecedented Demand Growth - Edinburgh University Press Journals, https://www.euppublishing.com/doi/10.3366/gels.2025.0141

48. Event-Triggering Virtual Inertia Control of PV Systems with Power Reserve - Aalborg Universitets forskningsportal, https://vbn.aau.dk/ws/portalfiles/portal/475879834/Event_Triggering_Virtual_Inertia_Control_of_PV_Systems_with_Power_Reserve.pdf

49. Grid-friendly Power Control for Smart Photovoltaic Systems - Aalborg Universitets forskningsportal, https://vbn.aau.dk/ws/files/331671478/Unmarked_Manuscript.pdf

50. A Hybrid Deep Learning and Model Predictive Control Framework for Wind Farm Frequency Regulation - MDPI, https://www.mdpi.com/2071-1050/17/18/8445

51. Frequency regulation service provision in data center with computational flexibility | Request PDF - ResearchGate, https://www.researchgate.net/publication/333558551_Frequency_regulation_service_provision_in_data_center_with_computational_flexibility

52. Strengthening Europe's tech sovereignty - European Commission, https://commission.europa.eu/news-and-media/news/strengthening-europes-tech-sovereignty-2026-06-03_en

53. What the new EU tech sovereignty package means for businesses | Proton, https://proton.me/business/blog/eu-tech-sovereignty-package

54. Why the EU's digital dreams begin with lower electricity prices | Think Tank Europa, https://thinkeuropa.dk/en/brief/2026-01-why-the-eus-digital-dreams-begin-with-lower-electricity-prices

55. EU Energy Efficiency Directive: Essential guidelines for Data Centers compliance - Socomec, https://emea.socomec.com/en/news/eu-energy-efficiency-directive-what-new-rules-mean-data-centres

56. Minimum performance standards for EU data centres - Energy - European Commission, https://energy.ec.europa.eu/resources/preparatory-studies/minimum-performance-standards-eu-data-centres_en

57. Commission unveils strategy to bring Europe's first SMRs online by the early 2030s - Energy, https://energy.ec.europa.eu/news/commission-unveils-strategy-bring-europes-first-smrs-online-early-2030s-2026-03-10_en

58. Böll EU Brief 01/2026 | Small modular reactors – smaller regulation?, https://eu.boell.org/en/small-modular-reactors

59. Nuclear Small Modular Reactors (SMRs) Market 2026-2046: Technologies, Players, Benchmarking, Forecasts - IDTechEx, https://www.idtechex.com/en/research-report/nuclear-small-modular-reactors-smrs-market/1153

60. Small modular reactors - Energy - European Commission, https://energy.ec.europa.eu/topics/nuclear-energy/small-modular-reactors_en

61. Nuclear Small Modular Reactors (SMRs) Market 2026-2046 | 700 GW Forecast, https://www.futuremarketsinc.com/the-global-nuclear-small-modular-reactors-smrs-market-2026-2046/

62. World Bank ends ban on funding nuclear energy, https://www.world-nuclear-news.org/articles/world-bank-agrees-to-end-ban-on-funding-nuclear-energy

63. World Bank Lifts Longstanding Ban On Nuclear Energy Financing In Historic Shift - NucNet, https://www.nucnet.org/news/world-bank-lifts-longstanding-ban-on-nuclear-energy-financing-in-historic-shift-6-4-2025

64. A Global Energy Reset: World Bank Ends Nuclear Ban, Unlocks Billions for Clean Power, https://anthropoceneinstitute.com/2025/08/word-bank-nuclear-band/

65. Nuclear Renaissance: How MDBs Are Rewriting Global Energy Finance for a Low-Carbon Future, https://www.nuclearbusiness-platform.com/media/insights/mdb-nuclear-energy

66. How France's Hauts-de-France Region Hopes to Become an “AI Valley” Following SoftBank's Massive Investment Pledge - Maddyness, https://www.maddyness.com/uk/2026/06/21/how-frances-hauts-de-france-region-hopes-to-become-an-ai-valley-following-softbanks-massive-investment-pledge/

67. Cloud Company Signs Agreement To Use Nuclear Energy For One Of World's Largest Supercomputers In France - NucNet, https://www.nucnet.org/news/cloud-company-signs-agreement-to-use-nuclear-energy-for-one-of-world-s-largest-supercomputers-in-france-2-2-2025

68. SoftBank Group to Build 5 GW of AI Data Center Capacity in France, https://group.softbank/en/news/press/20260531_0

69. Microsoft Announces $10 Billion AI Data Center Investment in Portugal | MLQ News, https://mlq.ai/news/microsoft-announces-10-billion-ai-data-center-investment-in-portugal/

70. Microsoft powers into Portugal with €8.6bn AI data-centre hub - Capacity, https://capacityglobal.com/news/microsoft-powers-into-portugal/

71. Microsoft announces €8.6 billion investment in Sines data centre - Portugal Resident, https://www.portugalresident.com/microsoft-announces-e8-6-billion-investment-in-sines-data-centre/