Product1 distinct publisher3 min readUpdated
Only 37% of the roughly 700MW Europe added in the second quarter went to Frankfurt, London, Amsterdam, Paris or Dublin, on CBRE's figures. Grid waits explain most of the rest.
The Product Desk · Product desk

Compiled by The Product DeskSomething wrong?How this is made
CBRE published its European data centre update on Tuesday showing that of the roughly 700 megawatts added in the second quarter, only 37% landed in the five established markets of Frankfurt, London, Amsterdam, Paris and Dublin [1]. That leaves 63%, or roughly 441MW, going somewhere else [1], and the reason is not land price or funding cost but how long it takes to get an electron. CBRE attributes the shift to supply constraint, saying "Demand continues to outpace supply, mostly due to a lack of available power causing delays in data center construction" [2], and reports more than half of quarterly take-up happening outside the five cities [3]. The binding number is time. Connecting a new site in the congested core markets can take "10 years or more", according to RaboResearch analysts Owen Thomson and Coen Hutters, against waits they describe as significantly shorter in Belgium, Portugal, Spain, the Nordics and Italy [4]. Amsterdam is the extreme, with Dutch grid operator TenneT saying there is no additional capacity across much of Noord-Holland for the next ten years [5]. The Netherlands also bans hyperscale projects above 70MW of IT capacity or ten hectares across most of the country [6]. For anything with a 2027 or 2028 revenue date, that is a closed market rather than an expensive one. The other two anchors are on similar timelines. Frankfurt's central grid will not see meaningful upgrades until the 2030s, and West London substation work is unlikely before the early part of that decade [7]; Nscale's flagship Essex site has been unable to switch on [8]. The constraint is not restricted to the old core either. Denmark's Energinet paused new connections earlier this year after its queue reached 60 gigawatts against national peak demand of about seven [9], a queue roughly 8.6 times the country's peak load [2]. What replaces the core is cheaper only in relative terms. Prime powered land in Europe now costs 2.26m euros per megawatt, up 82% since 2021 on JLL's figures, with primary markets carrying a 2.3 times premium over secondary sites and four times over tertiary [10] - which puts the land bill alone for a 100MW site at about 226m euros [3]. European industrial power ran at roughly twice US levels in 2025 on IEA figures [11]. Developers are still moving: greenfield rises from 8% to 39% of the 2026 to 2028 pipeline [12]. The demand side confirms the redirection. Signings by AI-focused neocloud providers reached 420MW in the second quarter against 89MW a year earlier, a 4.7-fold increase concentrated in the Nordics, which CBRE attributes to cheaper renewable power [13]. Spain's installed capacity hit 439MW at the end of 2025, with industry body Spain DC forecasting roughly 2,537MW by 2030 [14], close to a sixfold increase [4], and Amazon committing 33.7bn euros in Aragon [15]. Nebius is building a 310MW, 8.5bn euro site at Lappeenranta in Finland, and Stargate Norway near Narvik starts at 230MW on hydropower with 290MW more planned [16]. Uniper has committed 5bn euros to repurposing old power plant sites whose grid connections already exist [17]. Ireland's reopening shows what re-entry costs. Its connection moratorium ended in December 2025, conditional on behind-the-meter generation sized to the full grid connection, siting in unconstrained locations, and 80% of annual demand matched by renewable investment in the country [18], in a market where data centres already draw more than 22% of national electricity [19]. The installed base has not moved. Hyperscaler self-build reaches 4.3GW across Europe, up 22% in a seventeenth consecutive year of double-digit growth [20], with around 70% of operational self-build in Ireland, the Netherlands, Sweden and Belgium [21]. Core-market vacancy fell slightly to 6.4% last quarter while the rest of Europe is forecast to reach 19% by year end [22], a gap of 12.6 points [5].
Follow any of these and your For You feed starts watching them — no settings page required.
Ranked by verification strength, evidence, and original report placement.
Of the roughly 700 megawatts of European data centre capacity added in the second quarter, only 37% went to the five established markets of Frankfurt, London, Amsterdam, Paris and Dublin. Europe's total capacity will reach 13GW by the end of 2026, a fifth more than last year.
CBRE attributes the shift to supply constraint: "Demand continues to outpace supply, mostly due to a lack of available power causing delays in data center construction."
More than half of all take-up in the quarter happened outside the five established markets.
Connecting a new site in the congested core markets can take "10 years or more", according to RaboResearch analysts Owen Thomson and Coen Hutters, against waits they describe as significantly shorter in Belgium, Portugal, Spain, the Nordics and Italy.
Amsterdam is the extreme case at roughly a decade, and Dutch grid operator TenneT has said there is no additional capacity in much of Noord-Holland for the next ten years.
The Netherlands bans hyperscale projects above 70 megawatts of IT capacity or ten hectares across most of the country.
Evidence-backed comparisons of source perspectives and observed adoption signals. Read the methodology
Which Builder, Operator, and Investor concerns the observed source mix emphasized—not a truth score.
Evidence, demonstrated adoption, hype gap, incentives, and confidence are assessed independently, each on its own current evidence. How these are measured.
Well-sourced numbers, one publisher
Almost every quantitative claim is attributed to a named external data provider or operator — CBRE take-up and vacancy, JLL land pricing, IEA power prices, RaboResearch named analysts, TenneT, Energinet, Spain DC — which is unusually specific for a single article. But the cluster contains exactly one source, so none of it is independently corroborated here, the underlying CBRE and JLL reports are not directly inspectable, and a few assertions (Nscale Essex, the Netherlands 70MW cap) carry no attribution at all.
Relocation already visible in signed capacity
This is not a forward-looking thesis: the shift shows up in delivered and contracted megawatts. 63% of Q2 additions landed outside the big five, neocloud signings rose from 89MW to 420MW year over year, greenfield moves from 8% to 39% of the 2026-2028 pipeline, and named multi-billion projects are under construction in Finland, Norway and Spain while Uniper monetises connected brownfield sites. The counterweight is that 70% of operational hyperscaler self-build still sits in the legacy four countries, so the installed base has not moved — only the marginal build.
Mildly overstated framing on solid numbers
The article is unusually disciplined — it explicitly refuses the 'cheap land' framing, notes secondary land is only cheaper than Frankfurt, and flags that the old markets retain the installed base. The overstatement is confined to the causal absolutism of the framing: queues are shown to correlate with siting, but power price (roughly 2x US), moratoria, land cost and renewable availability are all in the same evidence base, so 'not land or capital' is stronger than the cited data proves. The forecast 19% non-core vacancy is also reported without asking what it implies if AI signings normalise.
Broker and association data with growth interests
The quantitative backbone comes from parties with commercial exposure to the trend: CBRE and JLL are transaction intermediaries whose fee base grows with take-up and land pricing, Spain DC is an industry association forecasting a 5.8x expansion of its own members' market, and the project figures are company commitments (Amazon, Nebius, Uniper) disclosed by the sponsors. RaboResearch, TenneT, Energinet and IEA inputs are comparatively disinterested, and the publisher shows no direct stake beyond audience interest in AI infrastructure.
Directionally solid, single-source specifics
The direction of travel is well supported by contracted megawatts and multiple independent institutional data points reported in the piece, so the core finding is credible. Confidence is capped by having one publisher, no direct access to the CBRE or JLL underlying reports, unattributed specifics such as the Nscale Essex situation and the Dutch 70MW cap, and a mix of measured quarterly data with 2030 forecasts that cannot be evaluated here.
invest
Nvidia is brokering the Nordic build-out, not just supplying it2 distinct publishers
invest
The chips never move: Washington's fix for the Southeast Asia compute loophole1 distinct publisher
invest
Behind-the-meter gas is the data center buildout's real cost: 318 Mt a year1 distinct publisher
leadership
Sponsorship Is Not Retention: What An Amazon H-1B Win Still Does Not Buy1 distinct publisher
Distinct publishers with included, body-backed reporting in this cluster.
1 article · August 19, 2026