1. Background
Europe’s AI boom is colliding with two constraints: electricity and sovereignty. Generative AI and large‑model training require vast computing power and have triggered a surge in data‑centre construction. Belgium’s grid operator Elia reported that requests from data‑centre operators in Belgium have grown nine‑fold since 2022 and that reserved capacity for 2034 already exceeds double the 8 TWh outlined in national grid plansreuters.com. As AI workloads scale, electricity demand from data centres across Europe is projected to more than double by 2030 and could triple in Belgium by 2035, making up ~10 % of Belgium’s power consumption—far above the current EU average (~2 %)silicon.co.uk. At the same time, Europe is pursuing digital sovereignty, seeking to ensure that AI innovations and sensitive data remain under European jurisdiction. Denmark’s newly launched Gefion supercomputer and the Danish Centre for AI Innovation (DCAI) illustrate this trend: Gefion combines 191 DGX‑H100 systems with 1 528 NVIDIA H100 GPUs and 382 Intel Xeon CPUsdatacenterdynamics.com and, in October 2025, DCAI added 140 PB of storage (80 PB NVMe) to create one of Europe’s first fully sovereign, end‑to‑end encrypted AI infrastructuresprnewswire.com. These developments have profound implications for multiple sectors.
Top picks for "balanc ambition energy"
Open Amazon search results for this keyword.
As an affiliate, we earn on qualifying purchases.
2. Belgium’s Proposed Data‑centre Capacity Caps
2.1 Overview of the Proposal
Belgium’s grid operator Elia told Reuters that it may set an electricity allocation limit for data centres to prevent other industrial users from being squeezed outreuters.com. Under the proposal, data centres would be placed in a separate category with fixed capacity allocationsreuters.com. Flexible connections—where access may be curtailed during grid congestion—would remain possiblereuters.com. The 2028‑2038 federal grid plan will address data‑centre consumptionreuters.com, and Belgium’s energy minister emphasised he would pay close attention to the issuereuters.com.
Data‑centre requests in Belgium have surged dramatically. Elia said requests have increased nine‑fold since 2022reuters.com, and reserved capacity for 2034 is already more than double the 8 TWh foreseen in grid plansreuters.com. Analysts note that data centres already consume ~4 % of Belgium’s electricity, versus the EU average of ~2 %, and could reach 10 % by 2035silicon.co.uk. A Boston Consulting Group study found that electricity consumption by Belgian data centres could triple by 2035silicon.co.uk. The growth is driven by hyperscalers such as Google, which plans to invest €5 billion in Belgian data‑centre expansion to support its AI strategyreuters.com, raising concerns that speculative projects might lock up grid capacity for yearsreuters.com.
Belgium’s government is also contemplating doubling nuclear capacity from 4 GW to 8 GW, either by extending existing reactors or building new onesreuters.com. This underscores the scale of investment required to support future energy‑intensive industries.
2.2 Key Facts About the Belgian Proposal
| Aspect | Details | Sources |
|---|---|---|
| Drivers | AI‑driven data‑centre boom; requests for grid connections up 9× since 2022; reserved capacity for 2034 already >2× planned 8 TWh | Elia via Reutersreuters.com |
| Proposed measure | Create a separate category for data centres and allocate a fixed electricity quota; flexible connections may be limited during congestion | Elia via Reutersreuters.com |
| Rationale | Prevent other industrial users from being blocked; avoid speculative developments hoarding capacityreuters.com | Reutersreuters.com |
| Future planning | Data‑centre consumption to be addressed in 2028‑2038 federal grid planreuters.com; energy minister pledges oversightreuters.com | |
| Energy landscape | Data centres already use ~4 % of Belgium’s electricity; could rise to 10 % by 2035silicon.co.uk; Belgium exploring doubling nuclear capacity to 8 GWreuters.com | |
| Industry investment | Google plans €5 billion investment in Belgian data‑centre expansionreuters.com |
2.3 Market Impact and Vertical‑Specific Effects
a. Hyperscalers & Cloud Providers (Google, Microsoft, AWS)
- Capacity constraints & delays. Fixed quota allocations mean hyperscalers may not be able to secure enough power for expansion, delaying AI‑cloud offerings. The requirement to justify projects could curb speculative build‑outsreuters.com.
- Shift to other regions. Providers may redirect investment to neighbouring countries with more accommodating policies (e.g., the Netherlands, Nordics), altering the European data‑centre geography. Europe’s new AI Factories initiative, which selects multiple sites across the EU to host AI‑optimised systemseurohpc-ju.europa.eu, offers alternative locations with guaranteed sovereignty.
- Efficiency race. Energy caps incentivise investment in energy‑efficient chips (e.g., Nvidia B300, AMD MI300), liquid cooling, and more efficient model architectures. Companies like Mistral (France) emphasise smaller, open‑source models with lower energy consumptionsilicon.co.uk.
- On‑site generation & PPAs. Hyperscalers might invest in on‑site solar/battery systems or long‑term power‑purchase agreements with renewable providers to secure independent capacity. Belgium’s exploration of nuclear expansionreuters.com further highlights the need for long‑term power sources.
b. Colocation & Data‑centre Operators
- Preferential category. A dedicated quota provides clarity but also caps growth, potentially disadvantaging operators planning large‑scale AI campuses.
- Premium pricing & flexible contracts. Scarcity of capacity could drive up colocation prices and encourage flexible “curtailment” contracts. Operators may offer variable pricing based on time‑of‑use or grid congestion.
- Regional competition. Operators in countries with abundant renewable energy (e.g., Denmark, Sweden, Finland) will become more attractive, benefiting from the shift in demand.
c. Industrial & Manufacturing Users (Chemicals, Pharma, Automotive)
- Protected access. Elia’s plan aims to ensure that energy‑intensive AI data centres do not crowd out other industrial usersreuters.com. Manufacturers reliant on electricity for production (chemical plants, steel mills) gain security.
- Opportunity cost. However, AI adoption in manufacturing may slow if local compute capacity is constrained. Firms may turn to sovereign AI hubs like Denmark’s DCAI or other EuroHPC AI Factories for model training.
d. Energy & Utilities
- Grid investment & flexibility. Utilities will need to upgrade transmission lines and integrate more renewables. Flexible connections proposed by Elia encourage demand response and grid balancing.
- Nuclear & renewables push. Belgium’s plan to double nuclear capacityreuters.com points to long‑term investment opportunities. Renewable developers gain leverage as hyperscalers seek green power to secure expansions.
e. AI Startups & Research
- Access barriers. Start‑ups may struggle to obtain power allocations in Belgium, leading them to use cloud providers or sovereign AI hubs elsewhere.
- Innovation in efficiency. Limited capacity fosters research into energy‑efficient AI algorithms and hardware.
f. Government & Regulators
- Policy precedent. Belgium’s approach could serve as a model for other EU countries facing grid congestion. Regulators will need to balance industrial growth, sustainability, and energy security.
- Transparency & fairness. The separate category ensures transparent allocation but risks entrenching incumbents. Policy must encourage innovation while preventing speculative “capacity hoarding.”
3. Denmark’s Sovereign AI Infrastructure – Gefion & DCAI
3.1 Architecture and Key Features
Denmark’s Gefion supercomputer and the Danish Centre for AI Innovation (DCAI) were inaugurated in October 2024 and significantly expanded in October 2025. Key features include:
| Component | Specification/Feature | Sources |
|---|---|---|
| Compute | 191 NVIDIA DGX H100 systems (1 528 H100 Tensor Core GPUs and 382 Intel Xeon Platinum CPUs) interconnected with Quantum‑2 InfiniBanddatacenterdynamics.comnovonordiskfonden.dk; installed by Eviden (Atos) and hosted in a Digital Realty data centre using 100 % renewable energynovonordiskfonden.dk. | |
| Storage & Networking | In Oct 2025 DCAI and WEKA launched a 140 PB storage platform with 80 PB of NVMe flash using WEKA’s NeuralMesh softwareprnewswire.com. This turns Gefion into a full AI factory—users can ingest data, train models and deploy them under one roofprnewswire.com. | |
| Performance & Efficiency | NeuralMesh delivers sub‑millisecond latency and zero‑copy data paths that continuously feed GPUsprnewswire.com. A single‑tier, software‑defined architecture reduces footprint and energy consumption compared with multi‑tier systemsprnewswire.com, lowering total cost of ownership and supporting green AI objectives. | |
| Security & Sovereignty | End‑to‑end encryption; zero operator access to customer data; compliance with EU data sovereignty mandates; independent privacy auditsprnewswire.com. Data is not stored permanently; users control their datanovonordiskfonden.dk. | |
| Access & Users | Initially accessible to select enterprise, research and government customers; broader onboarding planned Q1 2026prnewswire.com. The centre is owned by DCAI A/S (majority by Novo Nordisk Foundation, minority by EIFO) with revenues reinvestednovonordiskfonden.dk. | |
| Use cases | Early pilot projects include weather forecasting (Danish Meteorological Institute) and quantum‑circuit simulationsdatacenterdynamics.com. Flagship collaborations include genomics research for vaccine design and climate modellingdatacenterdynamics.com. |
3.2 Market Impact and Vertical Benefits
a. Regulated Sectors – Healthcare, Life Sciences & Public Services
- Secure, compliant compute. DCAI offers end‑to‑end encrypted infrastructure with full control of dataprnewswire.com and 100 % renewable powernovonordiskfonden.dk, addressing regulatory requirements for health‑care, genomics, financial services and government agencies. Sensitive data, such as patient records or defence information, can be processed locally without leaving European jurisdiction.
- Accelerated R&D. Life‑science researchers can train models for drug discovery, protein design and genomics using high‑capacity computenovonordiskfonden.dk. Novo Nordisk Foundation emphasised that extreme AI power enables breakthroughs in disease diagnosis and treatmentnovonordiskfonden.dk.
- Public‑sector AI adoption. Government agencies (e.g., weather forecasting, transportation planning) gain domestic AI capacity, reducing dependency on foreign hyperscalers and improving national resilience.
b. Research & Academia
- Accessible exascale‑class resources. Gefion provides researchers with an AI‑optimised supercomputer previously unavailable in Denmark, bridging the gap between European universities and global leaders. The centre has been consulted by universities and ministries as the key to unlocking AI‑based researchnovonordiskfonden.dk.
- Collaborative network. Through the EuroHPC AI Factories programme, Gefion is part of an interconnected network. Each factory acts as a one‑stop shop providing AI‑optimised HPC resources, training and expertiseeurohpc-ju.europa.eu, enabling cross‑border projects and fostering standardisation.
c. AI Startups & SMEs
- Domestic AI factory access. Start‑ups that cannot compete for capacity at hyperscaler data centres gain access to sovereign compute with predictable pricing. Differential payment models balance academic and commercial usenovonordiskfonden.dk.
- Data‑sovereignty differentiation. Start‑ups developing models for sensitive European industries can market their services as fully sovereign, potentially winning customers from global competitors.
d. Hyperscalers & Cloud Providers
- Partnership or competition? Hyperscalers may partner with DCAI (e.g., running AI inference endpoints or offering hybrid cloud integration). Alternatively, the sovereign platform competes by offering comparable performance at national scale.
- Hardware providers. The deployment demonstrates the strategic importance of NVIDIA’s DGX systems and WEKA’s storage software. Competitors (AMD, Graphcore, Cerebras) will vie to supply future European AI Factories.
e. HPC & Quantum Research
- Advanced research integration. Gefion will support complex simulation of quantum circuitsdatacenterdynamics.com and HPC‑AI convergence (climate, astrophysics). The platform can also integrate with Nvidia’s CUDA Quantum for hybrid CPU‑GPU‑QPU programmingnovonordiskfonden.dk.
f. Energy & Sustainability
- Green AI showcase. The use of 100 % renewable energy and reduced footprint demonstrates that high‑performance AI can be sustainablenovonordiskfonden.dk. This challenges the narrative that AI must be energy‑intensive and provides a benchmark for other regions.
3.3 Strategic Importance for Europe
Denmark’s sovereign AI infrastructure signals Europe’s shift from dependence on US hyperscalers to domestic AI factories. The EuroHPC Joint Undertaking selected six new AI Factory sites in October 2025 (Czechia, Lithuania, Netherlands, Poland, Romania and Spain) to join the existing networkeurohpc-ju.europa.eu. Each factory will act as a one‑stop shop for AI start‑ups, SMEs and researcherseurohpc-ju.europa.eu. The DCAI model demonstrates how compute, storage, training and data‑sovereignty requirements can be integrated. Other EU countries are likely to emulate this model, creating competition among AI factory providers and potentially reducing the dominance of non‑European hyperscalers.
4. Competitive Landscape & Cross‑Region Implications
4.1 Shift of Investment and Capacity
Belgium’s potential caps on data‑centre power create bottlenecks that may redirect investment to regions with more abundant energy. The Nordics (Denmark, Sweden, Norway) and certain Central‑European countries (Czechia, Lithuania, Romania) offer renewable energy, favourable cooling climates and newly designated AI Factory siteseurohpc-ju.europa.eu. Google’s €5 billion investment plans for Belgium will face scrutiny and may partially shift to countries like the Netherlands or Denmark. As a result, regional competition for AI infrastructure will intensify, with governments offering incentives (green energy, tax breaks) to attract hyperscalers and AI factories.
4.2 Energy Policy and Infrastructure Development
Grid constraints highlight the need for coordinated energy planning. Belgium’s doubling of nuclear capacity (from 4 GW to 8 GW)reuters.com illustrates the scale of investment necessary to support digital ambitions. Other countries might expand renewable portfolios (wind, solar, hydro) or consider nuclear and geothermal options to satisfy AI power demand. The interplay between energy policy and digital policy will shape national competitiveness.
4.3 Digital Sovereignty & Data‑Localisation
Europe is moving toward digital sovereignty, spurred by privacy regulations (GDPR) and the upcoming AI Act. The DCAI/WEKA platform shows that high‑performance AI can be offered within the EU while respecting data localisation. Countries that fail to provide sovereign AI capabilities may lose out on sensitive AI workloads. Conversely, building sovereign AI infrastructures is capital‑intensive and requires sustained public–private investment.
4.4 Vertical Competition
| Vertical | Competitive Dynamics |
|---|---|
| Hyperscalers | Must navigate energy caps in Belgium, invest in green power, and partner with sovereign AI facilities. Could face competition from EU‑backed AI factories offering similar performance with local sovereignty. |
| Data‑centre Operators | Belgium’s cap may slow growth; operators may expand in Nordics/Eastern Europe. Those offering sovereign capabilities gain differentiation. |
| Chip & Hardware Suppliers | NVIDIA dominates initial deployments (Gefion). AMD, Intel, Graphcore and start‑ups will compete to supply future AI factories. Energy efficiency per watt will be a key selling point. |
| Renewable & Nuclear Energy Providers | Increased demand for PPAs from hyperscalers; nuclear expansion in Belgium; opportunities for wind and solar developers in Nordics and Eastern Europe. |
| Regulated Industries | Benefit from sovereign AI compute; competition shifts to quality of AI models and domain expertise rather than compute access. |
| Start‑ups & SMEs | Compete through efficiency and sovereignty; may bypass large clouds by using AI factories; access to high‑performance compute becomes democratized. |
5. Conclusions & Recommendations
Belgium’s proposed data‑centre capacity caps and Denmark’s sovereign AI infrastructure illustrate the trade‑off between AI ambitions, energy constraints and sovereignty. Europe’s data‑centre boom risks overwhelming power grids; Belgium’s nine‑fold surge in data‑centre requests and plans to cap capacity underscore the urgency of energy planningreuters.com. Simultaneously, Europe is building its own AI factories to ensure data sovereignty—Denmark’s Gefion supercomputer with 1 528 GPUs and 140 PB of storage demonstrates that high‑performance AI can be offered domestically with renewable energydatacenterdynamics.comprnewswire.com.
Implications for Verticals:
- Hyperscalers & Cloud Providers: Prepare for stricter capacity allocations in Belgium; invest in energy‑efficient hardware, renewable PPAs and cross‑border data‑centre expansion. Partner with sovereign AI facilities to offer hybrid solutions.
- Data‑centre Operators: Diversify sites across Europe; invest in green energy and flexible connection technologies; consider building sovereign AI capabilities for regulated sectors.
- Regulated Industries & Public Sector: Leverage sovereign AI infrastructures like DCAI for sensitive workloads; engage in early collaborations to secure capacity.
- AI Start‑ups & SMEs: Exploit opportunities offered by EuroHPC AI Factories to access world‑class compute without ceding data sovereignty; develop energy‑efficient models.
- Energy Providers & Policy‑Makers: Coordinate long‑term energy investments (renewables, nuclear, grid upgrades) with digital infrastructure planning; design transparent allocation mechanisms to balance industrial and digital growth.
- Hardware & Chip Vendors: Focus on energy efficiency, lower latency and integration with sovereign AI architectures; engage in EU R&D programs to be part of future AI factory deployments.
Overall, Europe’s approach signals that AI innovation must be grounded in sustainable energy and data sovereignty. Companies that align with these priorities—through efficient technologies, renewable energy strategies and compliance with EU regulations—will gain competitive advantage in the evolving digital landscape.
