IN Brief:
- European hyperscale developments planned for 2026–2028 are being located substantially farther from major urban data-centre hubs.
- Access to electrical capacity and faster grid connections are becoming central site-selection criteria as AI drives larger power requirements.
- The shift could redirect major digital investment towards secondary locations where transmission capacity and developable powered land remain available.
European AI data-centre development is moving farther from established metropolitan hubs as access to electricity, connection times, and the cost of powered land increasingly determine where new hyperscale computing infrastructure can be built.
Analysis supplied by JLL indicates that hyperscale centres planned to enter service between 2026 and 2028 will sit an average of around 175km from major urban hubs, compared with approximately 46km for projects developed between 2022 and 2025.
The change is being driven partly by the electrical scale of artificial-intelligence infrastructure. Large training and inference campuses require substantial blocks of continuous power, leaving an otherwise attractive development site with little value if the local network cannot deliver the necessary megawatts within a commercially useful timetable.
Powered-land prices illustrate how valuable that connection capability has become. Current market analysis puts costs at around €2.7 million per megawatt in Amsterdam, compared with roughly €200,000 per megawatt in areas around Bordeaux, creating a substantial incentive to look beyond the established European data-centre markets.
Grid capacity has therefore become part of the development asset itself. Fibre routes, planning, land cost, construction capability, and proximity to customers remain important, but a hyperscale site increasingly has to answer a more basic engineering question first: how much electrical capacity can be secured, at what voltage, and by what connection date?
AI is making that calculation harder because rack density is rising sharply. High-performance accelerator clusters concentrate far more computing equipment into each hall than conventional enterprise IT, increasing demand on incoming transformers, switchgear, uninterruptible power systems, busways, distribution equipment, batteries, and cooling systems.
The load is also difficult to treat as routinely interruptible. Computing operators can shift some workloads or participate selectively in demand management, but facilities sold on availability cannot assume that hundreds of megawatts of load will disappear every time the transmission system becomes constrained.
Developers are consequently beginning to move computing capacity towards electricity rather than assuming the grid can always be expanded around the preferred computing location. Secondary cities, former industrial regions, and greenfield sites may offer both cheaper land and transmission infrastructure that would be expensive or slow to reproduce in London, Frankfurt, Amsterdam, or other established clusters.
That relocation changes the geography of power-system planning. A hyperscale campus drawing 100MW or several hundred megawatts resembles a heavy industrial load, and several developments in one region can alter reinforcement requirements for substations, circuits, reactive compensation, and the wider transmission boundary.
Data centres also introduce a different demand pattern from many traditional industrial sites. Computing can run continuously through nights, weekends, and seasonal demand troughs, meaning a large campus can add a comparatively persistent block of electrical demand rather than a load concentrated mainly during working hours.
The transmission system must be able to meet that requirement not only under normal operating conditions but through credible circuit and transformer outages. Network planners therefore have to consider redundancy, voltage stability, thermal limits, fault levels, and the ability to maintain supplies when part of the surrounding infrastructure is unavailable.
The data centre then adds another layer of resilience behind the grid connection. Multiple utility feeds, transformers, switchboards, UPS systems, batteries, and backup generation are commonly used to prevent a single electrical fault from interrupting computing operations, increasing both capital cost and the complexity of protection coordination.
Power quality also becomes more significant as computing density rises. Large quantities of switched-mode power electronics and high-capacity converters can affect harmonics and reactive power, while cooling equipment introduces further motor and drive loads. Electrical design consequently has to consider the behaviour of the entire campus rather than treating the facility as a simple resistive load at the transmission boundary.
Connection timing remains one of the more difficult constraints because digital projects and electricity networks operate on different development cycles. A data centre can move from site acquisition to construction far faster than a new transmission line can pass through routing, consent, procurement, civil works, and commissioning.
Capital behind AI infrastructure gives developers little incentive to leave expensive computing equipment waiting for a future grid reinforcement. That encourages projects towards locations where electrical capacity already exists, even where those areas sit well outside the traditional clusters favoured for earlier generations of cloud infrastructure.
The strategy is not without limits. A secondary location that appears to have spare capacity can quickly become constrained if several data centres, industrial electrification projects, EV infrastructure, heat electrification, and new residential demand converge on the same network.
Operators are also exploring on-site batteries, generation, and flexible demand to improve resilience and manage peak requirements. Those technologies can influence the load profile seen by the grid, but they do not remove the requirement for a credible connection where high-density computing operates continuously.
The resulting competition for power is beginning to make electricity infrastructure a determinant of digital geography. The most valuable data-centre sites may increasingly be those where transmission capacity, land, and development timetables align — even when the nearest major technology hub is more than a hundred kilometres away.