In Europe, electrical connection arrives ten years too late for AI

A data center is built in eighteen months. Obtaining the electrical connection to power it takes between seven and thirteen years in Europe’s major metropolitan areas. It is this gap, not money or legislation, that transforms European digital sovereignty into a paper ambition.

In June 2026, the European Commission adopted the Cloud and AI Development Act. The text is ambitious: triple the continent’s data center capacity by 2035, deploy AI gigafactories as part of the InvestAI fund—endowed with 20 billion euros to finance all 4 to 5 planned gigafactories—, reduce dependence on American hyperscalers. The intention is serious. The problem is physical.

The Essentials

  • Connection delays to the electrical grid reach 7 to 13 years in major European data center hubs (FLAP-D: Frankfurt, London, Amsterdam, Paris, Dublin), compared to 12 to 24 months to build the infrastructure itself, according to the International Energy Agency.
  • The Cloud and AI Development Act, adopted in June 2026, aims to triple Europe’s data center capacity by 2035 and fund AI gigafactories via the InvestAI fund, endowed with 20 billion euros for the entire program.
  • The gap between the construction schedule and the connection schedule makes it physically impossible to meet targets set for 2030.
  • If delays are not structurally reduced, the benefits (computing power, skilled jobs, strategic security) will concentrate with operators capable of securing already-connected land, while network infrastructure costs are socialized at the national level.

In Ireland, network operator EirGrid suspended new connection requests in the Dublin region between 2022 and 2024. In Amsterdam, the municipality imposed a one-year moratorium on new data centers in 2019, before lifting it in 2020—distinct restrictions were subsequently imposed nationally starting in 2022. In Frankfurt, waiting times for a connection to the high-voltage network now exceed a decade according to data compiled by the IEA. These situations are not local anomalies. They are the symptoms of electrical infrastructure designed for a world without artificial intelligence.

Seven to Thirteen Years to Plug In a Building

The figure deserves to be faced squarely. Seven to thirteen years. That is the average time required to connect a large data center in Europe’s most sought-after hubs. Meanwhile, an operator can order, design, build and commission its computing infrastructure. It can train its teams. It can sign contracts with its customers. Then it waits.

The IEA details the reasons for this wait in a note published in May 2026. First cause: queues for connections have exploded. In Germany, transmission network managers had received approximately 270 gigawatts of connection requests by the end of the third quarter of 2025, of which 211 gigawatts for batteries alone—estimates including distribution and all technologies rising to more than 720 gigawatts. In France, RTE processes similar volumes over several years. Second cause: transport and distribution networks were built in the last century to deliver electricity from power plants to homes, not to absorb concentrated consumption spikes in a few industrial areas. A 100-megawatt data center is equivalent to the consumption of a mid-sized city. Bringing ten of them within a fifty-kilometer radius requires investments in high-voltage lines, transformation stations and distribution networks that network managers have not foreseen, budgeted or planned.

Third cause, less visible: authorization procedures. Building an extra-high-voltage line in Europe takes between 4 and 8 years according to the IEA, or up to 15 years according to the European Parliament, when combining permitting phases, environmental impact studies and construction. Legal appeals, public inquiries, and impact assessments accumulate. No one is wrong in this system. Residents contesting a line have legitimate reasons. Regulators demanding impact studies are doing their job. But the sum of these procedures produces a result that nobody intended: Europe finds itself incapable of physically cabling its digital ambitions within the allotted time.

The CADA Creates Demand Without Solving Infrastructure

The Cloud and AI Development Act is the most ambitious text Europe has ever adopted regarding digital infrastructure. It creates a financing framework for AI gigafactories, facilitates access to industrial land for operators, simplifies building permit procedures and commits European public funds to co-finance infrastructure. It is a serious response to a documented backlog. The Draghi report had stated it unambiguously: Europe suffers from chronic underinvestment in strategic technologies, and this deficit is paid for through dependence.

But the CADA addresses the building, not the cable. Building permit procedures for data centers are accelerated. Electrical connection procedures are not. Industrial land is mobilized. The question of how to power it remains open. It is like building a highway without planning the access ramps.

Analysts agree in predicting rapid and sustained growth in global demand for computing power for AI through the end of the decade. In this context, every year lost in connection queues is a year during which European operators lose ground to their American and Asian competitors, some of whom benefit from connection times of two to three years in their priority areas.

The Commission has become aware of the problem. A technical annex to the CADA explicitly mentions the need to align connection procedures with deployment timelines. But mentioning a problem is not solving it. Network managers remain under the supervision of national regulators. Authorization procedures for extra-high-voltage lines fall under urban planning law in each Member State. The European Union can finance, it can incentivize, it can legislate on building permits. It cannot short-circuit twelve national legal systems with a single stroke of the pen.

Who Captures Benefits When Infrastructure Becomes Scarce

The scarcity of connected sites produces a capture economy. When the number of locations where one can actually connect a large data center counts on one’s fingers, these locations become strategic assets. Their holders, whether landowners, industrial park operators or companies that anticipated the problem five years ago, find themselves in a position of strength in all negotiations.

This concentration has several consequences. From an economic perspective first: large technology companies, which have teams capable of navigating connection procedures and balance sheets to secure capacity reservations years in advance, appropriate most available sites. Microsoft, Google, Amazon and Meta invested tens of billions of euros in Europe over the past three years, in part precisely because they had the resources to secure connections where smaller European operators could not.

From a fiscal perspective next. Data centers consume considerable quantities of electricity. Ireland, which hosts a disproportionate share of European digital infrastructure, sees data centers represent more than 20% of its national electricity consumption according to CSO Ireland data, and this figure is rising. Yet this consumption is partly indirectly subsidized: investments in networks enabling these connections are shared across all consumers via transmission charges. Infrastructure costs are socialized. Benefits, computing power and competitive advantage, are privatized.

This is not an argument against data centers. It is an observation about their regulation. AI mobilizes growing energy resources that existing infrastructure was not designed to absorb, and the question of who finances this adaptation deserves an explicit political answer rather than a silent rate adjustment.

Paths That Already Exist

The picture is not without solutions. Several countries and operators have begun working on the right levers.

The first path is geographic dispersion. FLAP-D hubs (Frankfurt, London, Amsterdam, Paris, Dublin) are saturating because everyone wants to locate there. But other European regions have underutilized electrical capacity, available industrial space and properly dimensioned networks. Finland, Sweden, Poland and certain reconverting German industrial regions offer significantly shorter connection times. A few operators have already migrated parts of their projects to these areas. But regulation and fiscal incentives must support this movement rather than leaving it to each operator’s ad-hoc initiative.

The second path is reforming connection queues. In the United Kingdom, Ofgem undertook a radical reform of the connection queue system in 2024, which had accumulated hundreds of gigawatts of requests, some of which concerned speculative or abandoned projects. By distinguishing serious projects, with land, financing and defined timelines, from preventive reservation requests, the regulatory authority began to unclog waiting lists. The model is imperfect but instructive. Several EU Member States are watching this precedent with interest.

The third path is accelerating authorization procedures for electrical lines, by analogy with what Europe has already done for renewables. The RED III directive reduced authorization delays for wind and solar farms by establishing regulatory frameworks and creating priority zones. A similar mechanism for strategic electricity transmission lines, coupled with accelerated consultation procedures, would significantly reduce authorization times. This is not negligible when the horizon is 2035.

Fourth, some operators are exploring onsite energy storage solutions, allowing connection at lower contracted power while absorbing peaks via industrial batteries. This is a partial solution, more expensive per megawatt-hour, but allows operations to start before the network is fully dimensioned. Pilot projects of this type are underway in Ireland and the Netherlands.

What the CADA Must Correct To Not Be Just Another Text

The current trajectory produces a predictable result. Operators capable of navigating administrative complexity, with the legal teams, institutional relationships and balance sheets necessary to secure connections over a ten-year horizon, will deploy their infrastructure. Others will wait or relocate. Europe will have built a favorable regulatory framework, funded gigafactories, simplified building permits, and nonetheless watch computing centers establish themselves primarily in third countries where electricity arrives faster.

The CADA has until the end of 2026 to be supplemented by implementation texts. The essential will be decided in these texts. Three points deserve particular attention: explicit inclusion of electrical connection procedures within the scope of administrative simplifications, creation of a coordination mechanism between national network managers to treat strategic digital infrastructure projects as projects of common European interest, and definition of transparent rules on network cost socialization.

AI agents entering European companies en masse need servers. These servers need electricity. This electricity needs cables. The rest of European AI industrial policy depends on this physical link that no one sees because it is buried beneath the cities.

The real question, leading up to the CADA review planned for 2028, is the following: will Europe reform its electrical grid law with the same determination it brought to reforming its digital competition law?

Sources

  1. International Energy Agency, Overcoming energy constraints is key to delivering on Europe’s data centre goals, May 2026 — iea.org
  2. European Commission, Cloud and AI Development Act, June 2026 — commission.europa.eu
  3. Ember Energy, European Electricity Review, data on data center consumption in Ireland — ember-climate.org
  4. IEA – Connection times FLAP-D (7-10 years) — iea.org
  5. CSO Ireland – Data Centres Metered Electricity Consumption 2024 — cso.ie
  6. EirGrid / CRU – Dublin Moratorium 2022-2028 — energyconnects.com
  7. Baker McKenzie – Connection requests in Germany (Q3 2025) — bakermckenzie.com
  8. Data Center Dynamics – Amsterdam moratorium — datacenterdynamics.com
  9. Ember Energy – Grids for data centres in Europe — ember-energy.org
  10. Knowledge Centre Data & Society – InvestAI and gigafactories — data-en-maatschappij.ai