Summary

• Greenfield projects have risen from 8% of recent deliveries to 39% of Europe’s 2026–28 data-centre pipeline.
• Planned hyperscale sites average 175 kilometres from major hubs as operators pursue available power and cheaper land.
• London, Frankfurt, Paris, Amsterdam, and Dublin remain important for latency-sensitive workloads despite the outward shift.

Europe’s next generation of data centres is being built markedly further from its established technology hubs, as the electricity and land requirements of artificial intelligence make proximity to large cities less valuable than access to substantial blocks of usable power.

Research from JLL shows that hyperscale sites in the 2026–28 development pipeline are located an average of 175 kilometres from major hub cities, compared with 46 kilometres for projects delivered between 2022 and 2025. Greenfield developments have meanwhile increased from 8% of recent deliveries to 39% of the forthcoming pipeline.

That movement does not amount to an evacuation of Europe’s established data-centre markets. Frankfurt, London, Amsterdam, Paris, and Dublin — generally grouped as FLAP-D — reached about 3.8GW of live capacity during the first half of 2026, with another 1.4GW under construction and 2GW planned.

Instead, the market is separating according to what different computing workloads require. Applications that depend on very low latency still benefit from sitting near major concentrations of customers and networks, while large AI training clusters can tolerate greater distance and therefore follow electricity, development land, and grid connections into secondary locations.

Power is redrawing the development map

AI training facilities with requirements above 100MW can consume electricity on a scale that makes the availability of a suitable grid connection more important than being within a conventional technology district. JLL expects more than half of Europe’s AI growth to take place in Nordic and other Tier 2 markets, where operators can sometimes secure larger sites and power allocations than densely constrained metropolitan areas can offer.

Land economics reinforce that incentive. Prime powered land across FLAP-D has risen 82% in cost since 2021 to €2.26 million per megawatt, while primary markets command substantially higher prices than secondary and tertiary locations. Yet cheaper land is useful only when electricity can be delivered, leaving power rather than property cost as the underlying constraint.

The shift gives physical form to a trend already visible in European AI leasing. Recent capacity commitments have been spreading beyond Europe’s traditional data-centre centres, while JLL’s figures show that operators are not simply choosing another nearby city but considering sites considerably further from the customers ultimately using the computing capacity.

Even so, the large hubs continue adding supply. Paris delivered 72.5MW during the first half of the year, London added 49MW, Frankfurt 45MW, Amsterdam 16.3MW, and Dublin 11.4MW. Vacancy across the five markets stood at 6.4% during the second quarter, with Frankfurt tighter still at 3.1%.

AI separates computing from the customer

During much of the cloud era, data-centre location was strongly influenced by proximity to corporate demand, financial markets, network exchanges, and large populations. AI training changes part of that equation because a huge cluster can process data remotely without every computation taking place only a few milliseconds from its eventual user.

Distance does not remove the need for fibre connectivity, power resilience, planning approval, cooling, skills, or access to equipment. It does, however, allow operators to consider locations that would have been less attractive for conventional colocation, particularly where utilities can provide capacity that a major metropolitan grid cannot release quickly.

Ireland illustrates the trade-off. A revised connection framework is directing some development beyond Greater Dublin, while Frankfurt continues to contend with long grid connection lead times. Elsewhere, the same calculation is steering investment towards regions where generation, transmission capacity, and available land align more comfortably.

That movement can redistribute some of the economic effects of data-centre construction. Large projects bring construction activity, network investment, power demand, and tax revenues into regions outside established technology centres, although they also expose local electricity systems and planning authorities to facilities whose consumption resembles major industrial plants.

Core markets become one layer of the system

Europe’s principal data-centre markets remain valuable partly because some computing cannot simply move away. Scarce powered land and difficult grid connections preserve that premium, while larger remote campuses create an additional layer for workloads that are less sensitive to distance.

A facility described commercially as serving London, Frankfurt, or Paris may consequently sit far beyond the urban market associated with the service, with high-capacity fibre carrying data between power-rich computing locations and the organisations using them.

The geography of European AI will therefore be influenced as much by substations, transmission networks, planning decisions, and land availability as by processors. The demand can still originate in Europe’s commercial centres, but the infrastructure required to satisfy it is spreading considerably further across the map.