For much of the past decade, Europe’s data center market developed around a relatively small group of established metropolitan hubs. Frankfurt, London, Amsterdam, Paris and Dublin, the traditional ‘FLAPD’ markets, offered the customer concentration, connectivity and cloud ecosystems needed during the growth phase of public cloud.
Those markets will continue to grow, but the infrastructure requirements associated with AI are changing the geographical balance. Large-scale AI training and inference require significantly more power, greater rack density and increasingly specialised cooling. At the same time, grid congestion is making it harder to deliver new capacity in some of Europe’s largest existing markets. The result is not the end of the traditional hubs, but broader geographical diversification across European, in which power availability, carbon intensity and delivery certainty carry much greater weight.
The scale of the underlying demand helps explain why this is happening. The International Energy Agency’s Energy and AI report estimates that data centers consumed around 415 TWh of electricity globally in 2024, equating to approximately 1.5% of worldwide electricity consumption. Its base case projects this rising to around 945 TWh by 2030, slightly more than double the 2024 level. The IEA identifies AI as the most important driver of that growth, alongside continued demand for other digital services.
Forecasts on this scale should be treated with a degree of caution. Improvements in chip efficiency, software and cooling could reduce the energy required for individual tasks, while the pace of AI adoption difficult to predict accurately. Even allowing for this uncertainty, the next phase of digital infrastructure development is placing enormous demands on power systems and transmission grids.
That matters because data centers cannot be considered separately from the grid that serves them. A facility cannot operate until sufficient power is available and connected. In some of Europe’s established hubs, the time required to secure a grid connection is now considerably longer than the time needed to build the facility itself.
Research published by Ember in June 2025 estimated that grid connections in established European data center markets were taking an average of seven to ten years, with some projects potentially waiting as long as 13 years. Ember also projected that European data center electricity demand could rise by around 150% between 2024 and 2035. Over the shorter period to 2030, it forecast faster growth in lower-congestion markets, including the Nordics and parts of southern Europe, where demand could increase by around 110%, compared with 55% across the five traditional hubs.
These figures help to explain why the geography of compute is becoming more distributed. Data center customers cannot plan AI deployment around a connection that may not arrive for most of a decade. They need locations in which land, power, connectivity and permitting can be brought together on a commercially relevant timetable.
The answer is not to move every workload to the same new location. Different forms of compute have different requirements. Some applications depend on proximity to users, financial markets or local data sources. Others, particularly large, power-intensive training and high-performance computing workloads, can be located further from major cities if the connectivity, security and regulatory conditions are right.
This creates a stronger role for parts of Northern Europe. The Nordic countries combine sophisticated digital infrastructure with power systems that are already substantially decarbonised. According to Nordic Energy Research, 96% of Nordic electricity is fossil-free. That regional figure inevitably contains differences between individual countries and market zones, but it demonstrates the structural advantage available when intensive computing is located within a predominantly low-carbon power system.
At Verne, we understood the importance of location well before the current acceleration in AI. Our first campus was developed in Iceland because its renewable electricity, climate and operating environment offered a strong foundation for high-performance computing. Verne’s Nordic data centers in Iceland and Finland are powered by 100% renewable energy and are designed for HPC, AI and other resource-intensive workloads.
This is what we mean by Natural Intelligence. It is the practical use of local conditions, engineering experience and human judgement to improve how infrastructure performs and utilises natural resources. In a cooler climate, for example, a facility can reduce its dependence on energy-intensive mechanical cooling. In a low-carbon power system, the electricity consumed by the IT equipment carries a lower associated operational footprint. Neither advantage removes the need for efficient design, but both create significant efficiency gains.
The engineering challenge is also changing at rack level. Traditional enterprise racks often operated at relatively modest densities, while the latest AI systems require well over 100 kW per rack and rising. At the same time individual GPUs can produce up to 50 times more heat than conventional CPUs.
These densities have direct implications for cooling. Air cooling systems remains appropriate for many workloads, but it becomes progressively more difficult to move sufficient heat away from the chips as power density rises. Liquid cooling brings the coolant closer to the source of the heat and can support far denser computing environments.
The point is not that every data center must switch all of its cooling systems from air to liquid, or that one cooling method is universally superior. Many facilities will operate hybrid environments for years ahead. The relevant question is whether the building, power system and cooling architecture can accommodate changing customer requirements without unnecessary complexity or disruption. Verne has already adapted existing facilities in Iceland and Finland to support closed-loop cooling systems for liquid-cooled deployments, building on infrastructure and operational expertise developed for high-density computing.
France has a different, but equally important role in this emerging geography. It combines one of Europe’s largest digital markets with an unusually low-carbon electricity system. According to French transmission system operator RTE’s 2025 Electricity Review, 95.2% of French electricity generation in 2025 was low-carbon, and the carbon intensity of domestic generation averaged 19.6 gCO₂e per kWh. RTE placed the equivalent European average, including the EU, Switzerland, Norway and the UK, at 175 gCO₂e per kWh.
This does not mean that power or suitable development land is readily available everywhere in France. Grid capacity, planning, local acceptance and connection timing remain material considerations. However, where those issues can be overcome, France offers the possibility of serving a major demand market through a system with substantially lower carbon intensity than many European alternatives.
The future European data center market is therefore unlikely to be defined by a simple choice between metropolitan hubs and less established locations. It will become significantly more distributed. Some capacity will remain close to Europe’s largest populations and businesses. The most power-intensive workloads will increasingly be directed towards locations where electricity is available, infrastructure can be delivered and operational lower carbon intensity can be addressed credibly.
Europe’s ability to capture the economic value of AI will depend in part on how quickly its grids, planning systems and infrastructure policies adapt to that reality. The Council of European Energy Regulators reported in June 2026 that shortages of grid connection capacity are affecting generation, storage and new demand across many member states. It recommended stronger tests of project maturity, financial guarantees, “use-it-or-lose-it” provisions and more flexible connection agreements to ensure scarce capacity is given to credible, deliverable projects.
That is a sensible direction. The objective should not be to give data centers unconditional priority over housing, transport, industrial electrification or other users. It should be to create a transparent system that distinguishes between speculative applications and projects capable of supporting real customer demand.
The first phase of the cloud growth was concentrated in a handful of large cities. AI will require a more diverse infrastructure model. The countries and regions best placed to benefit will be those that can bring together low-carbon power, connectivity, and grid access.
For Verne, this is the approach we have followed for more than a decade: starting with the efficiency attributes of a location, then designing and engineering infrastructure to optimise for a given location that can support demanding compute over the long term.