Europe Data Center Power Demand 2026
By Axis Intelligence Research
Co-authors: James Porter (Networking & Infrastructure) | Last updated: September 10, 2026 | License: CC BY 4.0
Europe’s data centres drew an estimated 87–96 TWh of electricity in 2024 and are on a path to 199–254 TWh by 2035. Converted into the unit grid operators actually plan with, that is a continuous load rising from roughly 10 GW today to between 22.7 GW and 29.0 GW — while Italy alone is holding connection requests worth more than three times that entire future load.
Quick Answer
European data centre electricity demand is expected to rise more than 50% between 2025 and 2030, from 87 TWh in 2024 to 134 TWh by 2030 and up to 254 TWh by 2035 (ENTSO-E, May 2026). Axis Intelligence Research converts those annual figures into Continuous Grid Load (CGL) — the average generation capacity Europe’s power system must hold available all year — and finds CGL rising from 9.9–11.0 GW in 2024 to 22.7–29.0 GW by 2035. Installed data centre IT power in Europe currently stands at 12.7 GW across more than 10,500 facilities, of which 9.9 GW sits in the EU27.
Key Findings
- Axis Intelligence Research finds that Europe’s data centre Continuous Grid Load (CGL) — annual TWh converted to average continuous gigawatts — rises from 9.9–11.0 GW in 2024 to 22.7–29.0 GW in 2035, a net addition of roughly 16 GW of round-the-clock load.
- According to ENTSO-E’s May 2026 study, Europe hosts over 10,500 data centre facilities of at least 50 kW, totalling 12.7 GW of installed IT power supply, of which 6,842 MW is colocation, 3,179 MW enterprise and 2,660 MW hyperscale.
- Axis Intelligence Research finds that Italy’s data centre grid connection queue — 95.38 GW of requests recorded by Terna as of 21 August 2026 — is 3.3 times larger than the entire continuous grid load Europe’s data centres are projected to draw in 2035, and 7.5 times Europe’s total installed data centre IT power today.
- According to Ember’s June 2025 grid analysis, European data centres used only 44% of their annual nominal power rating in 2024, while Ireland’s EirGrid reports average draw at approximately 34% of contracted capacity.
- Axis Intelligence Research estimates Europe’s announced data centre pipeline will be delivered at a rate of about 54% by 2030, derived from the IEA’s finding that the pipeline represents 130% of installed capacity while modelled 2030 capacity growth reaches only 70%.
How Much Power Do European Data Centres Actually Draw in 2026?
Annual terawatt-hours are the wrong unit for this question, and that is why the answer keeps getting muddled.
Two independent European datasets now publish a demand trajectory. ENTSO-E, the association of Europe’s 40 transmission system operators, put regional data centre electricity demand at 87 TWh in 2024, rising to 134 TWh by 2030 and to between 199 TWh and 254 TWh by 2035. Ember, working from ICIS market data, put 2024 demand at 96 TWh — equal to 3% of the region’s total electricity — heading to 168 TWh in 2030 and 236 TWh in 2035.
The two paths differ mainly in scope and base year. They agree on direction and on order of magnitude, which is more than can be said for most numbers in this sector.
| Trajectory | 2024 | 2030 | 2035 | Source |
|---|---|---|---|---|
| ENTSO-E (TSO view) | 87 TWh | 134 TWh | 199–254 TWh | ENTSO-E, Data centres and the power system, May 2026 |
| Ember / ICIS | 96 TWh | 168 TWh | 236 TWh | Ember, Grids for data centres in Europe, June 2025 |
| Share of European electricity | 3.0% | 4.5% | 5.7% | ICIS, via Ember |
Source column: as listed. Scope note: ENTSO-E covers the wider ENTSO-E membership; Ember defines Europe as EU27 plus the UK, Norway and Switzerland.
James Porter (Networking & Grid Infrastructure): A terawatt-hour is an accounting unit. A gigawatt is a dispatch unit. When a TSO decides whether to build a 400 kV substation, nobody in the room asks how many terawatt-hours the load will consume over the year — they ask how many megawatts it will pull at 3 a.m. in January while the wind is down. Publishing data centre demand exclusively in TWh has let the sector be discussed for five years in a unit that cannot be planned against.
Source: Europe Data Center Power Demand 2026 — Axis Intelligence Research (CC BY 4.0)
The Axis Continuous Grid Load (CGL) Series
CGL — Continuous Grid Load. The average continuous electrical capacity, in gigawatts, that Europe’s power system must hold available across a full year to serve data centre annual electricity demand. It is the annual consumption figure restated as a permanent, round-the-clock draw.
Formula and inputs, stated in full:
CGL (GW) = annual data centre electricity demand (TWh) × 1,000 ÷ 8,760 hours
2024, ENTSO-E path: 87 TWh × 1,000 ÷ 8,760 = 9.93 GW
2030, ENTSO-E path: 134 TWh × 1,000 ÷ 8,760 = 15.30 GW
2035, ENTSO-E low: 199 TWh × 1,000 ÷ 8,760 = 22.72 GW
2035, ENTSO-E high: 254 TWh × 1,000 ÷ 8,760 = 28.99 GW
2024, Ember path: 96 TWh × 1,000 ÷ 8,760 = 10.96 GW
2030, Ember path: 168 TWh × 1,000 ÷ 8,760 = 19.18 GW
2035, Ember path: 236 TWh × 1,000 ÷ 8,760 = 26.94 GW
| Year | CGL, low estimate | CGL, high estimate | Net addition vs 2024 |
|---|---|---|---|
| 2024 | 9.93 GW | 10.96 GW | — |
| 2030 | 15.30 GW | 19.18 GW | +5.37 to +8.22 GW |
| 2035 | 22.72 GW | 28.99 GW | +12.79 to +18.03 GW |
Source: Axis Intelligence Research calculation from ENTSO-E (May 2026) and Ember/ICIS (June 2025) demand trajectories. Snapshot date: September 10, 2026.
Two things follow immediately. First, Europe has to add roughly 16 GW of permanent, non-seasonal, non-negotiable load over eleven years — a block that does not switch off at night, does not go on holiday in August, and does not respond to a price signal unless someone builds it a reason to.
Second, CGL is a floor, not a peak. It describes the average. Instantaneous demand runs above it, and the ENTSO-E study documents why that matters: AI training clusters swing 30–60% of their consumption within milliseconds, and UPS architectures can remove hundreds of megawatts from the grid instantaneously during a minor voltage dip. The average is what you generate. The swing is what you have to survive.
Why CGL is not a peak-demand forecast. CGL divides annual energy by annual hours. It deliberately does not model coincident peak, diversity factors or regional non-coincidence. It answers one question — how much continuous generation this sector represents — and it answers it reproducibly from published inputs, which is precisely what a peak-demand model cannot claim.
Why Is the Connection Queue So Much Bigger Than the Load?
Here is the number that should reframe every European data centre power demand debate.
Italy’s transmission operator, Terna, reported data centre connection requests of roughly 30 GW at the end of 2024, rising past 50 GW across more than 300 projects by 30 June 2025. Terna data reported by Il Sole 24 Ore put the figure at 95.38 GW across 531 applications as of 21 August 2026. Eight data centres are actually connected to the Italian transmission grid as dedicated consumption units.
One National Queue Against a Continent’s Load
Axis Intelligence Research sets the Italian queue against the CGL series:
Italy connection queue (Aug 2026) = 95.38 GW
Europe CGL 2035, high estimate = 28.99 GW
Ratio = 3.29×
Europe installed data centre IT power = 12.70 GW
Ratio = 7.51×
One country’s connection queue is 3.3 times the continuous load Axis Intelligence Research projects for every data centre in Europe in 2035. That is not a forecast of Italian demand. It is a measurement of how far request volume has decoupled from deliverable load.
What a Realistic Delivery Rate Looks Like
Apply the two sourced correction factors — a delivery rate and a utilisation rate — and the queue resolves into something a grid planner can use:
Queue 95.38 GW
× 10% built and energised 9.54 GW installed
× 44% utilisation (Ember) 4.20 GW continuous load
× 8,760 h 36.8 TWh per year
Axis Intelligence Research estimates that a 10% realisation of Italy’s queue would add 4.2 GW of continuous load and 36.8 TWh of annual consumption — more than a third of what all of Europe’s data centres consumed in 2024. The 44% utilisation figure comes from IEA and McKinsey data cited by Ember; the 10% realisation rate is an illustrative scenario, not a forecast, and the calculation is shown so anyone can substitute their own.
| Italy data centre connection requests | Capacity | Applications | As of |
|---|---|---|---|
| Terna register | ~1 GW | — | 2021 |
| Terna register | 5.40 GW | — | 2023 |
| Terna register | 31.28 GW | — | end-2024 |
| Terna register | >50 GW | >300 | 30 June 2025 |
| Terna register | 95.38 GW | 531 | 21 August 2026 |
| Connected as consumption units | — | 8 | August 2026 |
Source: Terna (2024–2025 figures, Terna Lightbox); 2021, 2023 and 2026 figures from Terna data reported by Il Sole 24 Ore, 21 August 2026.
James Porter: A connection queue is a free option. It costs a developer almost nothing to reserve a position at four sites and drop three, and it costs the TSO a great deal to plan around all four. Terna is now sizing substations against a request book seven times larger than every operating data centre in Europe combined, and only fifteen of 531 applications are anywhere near final approval. The queue is not measuring demand. It is measuring the absence of a price on optionality.
Which European Countries Carry the Heaviest Data Centre Load?
National shares tell a different story from the queue. Ember’s 2024 baseline:
| Country | Data centre share of national electricity, 2024 | Source |
|---|---|---|
| Ireland | 19% | Ember / ICIS |
| Netherlands | 7% | Ember / ICIS |
| Germany | 4% | Ember / ICIS |
| United Kingdom | 4% | Ember / ICIS |
| France | 2% | Ember / ICIS |
| Europe (average) | 3% | Ember / ICIS |
The metropolitan concentration is sharper than any national figure suggests. In 2023, data centres accounted for between 33% and 42% of all electricity consumed in Amsterdam, London and Frankfurt, and close to 80% in Dublin.
The Centre of Gravity Is Moving
The IEA notes that Europe held more than 25% of global data centre capacity in 2015 and 15% in 2024, having grown at roughly half the worldwide rate over that decade. Inside Europe, the FLAP-D cluster — Frankfurt, London, Amsterdam, Paris, Dublin — holds about 62% of capacity today, falling to 55% by 2030 and 51% by 2035 on Ember’s projection. Demand outside FLAP-D grows 110% to 2030 against 55% inside it.
Sweden, Norway and Denmark triple their data centre demand by 2030. Denmark ran 26% annual capacity growth over the past five years and is projected at 12% for the next five. Belgium adds 72% between 2024 and 2030, then a further 42% to 2035. Grid availability, not tax policy, is doing the sorting.
James Porter: Ireland at 19% of national electricity is the number everyone quotes, and it is the least interesting one on the table. Watch the Netherlands instead. Planned Dutch projects average more than three times the size of facilities already operating there; in Spain the multiple is seven. A market does not go from 7% to trouble gradually — it goes there the moment three 200 MW campuses energise inside the same 150 kV ring.
How Long Does It Take to Connect a Data Centre to a European Grid?
This is the binding constraint, and the sector says so itself. In the EUDCA survey cited by ENTSO-E, 76% of European data centre operators named access to power as their single biggest challenge over the next three years, against 36% for regulatory compliance and 34% for permitting.
| Market condition | Grid connection wait | Source |
|---|---|---|
| EU range, all markets | 2–10 years | IEA, November 2025 |
| FLAP-D hubs | 7–10 years, up to 13 | IEA; Ember |
| Italy | as little as 3 years | Ember |
| Average build time, new facility | ~4 years | Ember |
| Phased connection (US precedent) | 12–18 months, from 5–8 years | Ember |
Against a construction cycle of about four years, a seven-to-thirteen-year connection wait is not a delay — it is a veto. The IEA puts direct grid congestion costs across the EU at €4.3 billion in 2024, on ACER figures, before counting the economic cost of the projects that simply never happen.
Ember’s modelling finds that countries offering half the connection time of the incumbent hubs stand to capture around 20% more data centre growth by 2030. That is the entire competitive dynamic of European siting in one sentence.
Can Data Centre Flexibility Absorb Any of This Load?
Partly, and far less than the headline potential suggests.
ENTSO-E’s System Needs for the Energy Transition work projects Europe’s short-duration and ramping flexibility requirement roughly doubling by 2030, to an effective need of 15–30 GW. Analysis of five European markets — Germany, Ireland, the Netherlands, Norway and the UK — puts data centre technical flexibility potential at 16.9 GW by 2030, of which up to 3.8 GW is realistically available once participation willingness and operational constraints are accounted for.
Flexibility realisation = 3.8 GW ÷ 16.9 GW = 22.5%
Coverage of European need = 3.8 GW ÷ (15 to 30 GW) = 12.7% to 25.3%
Axis Intelligence Research finds that data centres would deliver roughly 22.5% of their own technical flexibility potential, covering between 12.7% and 25.3% of Europe’s projected 2030 short-duration flexibility need. Useful. Not a solution.
The structural reason sits in the business model, not the hardware. Colocation operators control the building but not the workloads, and commonly pass energy costs straight to tenants — which removes the incentive at exactly the point where the capability sits. Hyperscalers, who do control both, are structurally best placed, but frontier AI training runs at very high utilisation with little redundancy to give away.
Two mechanisms move faster than markets will. The IEA estimates that 30 hours of annual flexibility would more than double the grid capacity available to European data centres. And ENTSO-E cites US analysis showing that combining flexible connection terms with on-site generation brought full operation forward by three to five years, with curtailment held to 40–70 hours a year — better than 99% grid availability.
James Porter: Flexibility is being sold as a grid service and bought as a queue-jumping instrument, and only the second framing has closed a deal so far. An operator will accept 60 hours of curtailment a year to energise in 2028 instead of 2034. The same operator will not bid UPS batteries into FCR for a few euros per megawatt-hour when energy is roughly half of operating expenditure and downtime is the whole business. Price the calendar, not the kilowatt, and the flexibility appears.
Will Europe Hit Its Capacity Target?
The EU’s AI Continent Action Plan, published 9 April 2025, set a target of at least tripling EU data centre capacity within five to seven years. The IEA’s own modelling does not get there.
Announced pipeline = 130% of installed capacity today (IEA)
Modelled 2030 growth = 70% above 2024 installed capacity (IEA)
Pipeline realisation rate = 70 ÷ 130 = 53.8%
Axis Intelligence Research estimates a pipeline realisation rate of approximately 54% — roughly half of announced European data centre capacity reaching energisation by 2030, with grid congestion and local constraints named by the IEA as the primary drag. A doubling by 2030 is the realistic outcome; a tripling is the target.
The composition of what does get built is already settled. ENTSO-E projects total installed IT power supply rising 70% between 2025 and 2030, with colocation — increasingly large “scale colocation” leased to hyperscale tenants — driving more than 70% of the increase. Scale colocation grows at 30% a year against 8% for retail and wholesale, and by 2030 an estimated 60–70% of colocation capacity is leased to hyperscalers. Enterprise data centre capacity in Europe is stagnant.
| Segment | Installed IT power, Europe | Share |
|---|---|---|
| Colocation (incl. scale colocation) | 6,842 MW | 54% |
| Enterprise | 3,179 MW | 25% |
| Hyperscale | 2,660 MW | 21% |
| Total | 12,681 MW | 100% |
Source: ENTSO-E, Data centres and the power system, May 2026. Share column calculated by Axis Intelligence Research.
That shift is a grid story before it is a market story. Enterprise sites run 1–5 MW. Scale colocation and hyperscale campuses run tens to hundreds of megawatts at a single connection point — and Terna reports receiving requests above 500 MW for individual sites, against a hyperscale norm of 100–200 MW.
Methodology
Collection. Every figure on this page was retrieved from a source document opened during production on 10 September 2026 and is logged in the accompanying CSV with source organisation, document, URL and retrieval date. No figure is drawn from model memory.
Primary sources. ENTSO-E, Data centres and the power system: expected trends, challenges and opportunities (May 2026); Ember, Grids for data centres in Europe (19 June 2025); IEA, Overcoming energy constraints is key to delivering on Europe’s data centre goals (16 November 2025); Terna, Data centres and the future: the energy challenge facing Italy (8 August 2025); European Commission, The AI Continent Action Plan (9 April 2025). One secondary figure — Italy’s 95.38 GW queue as of 21 August 2026 — is Terna data reported by Il Sole 24 Ore and is flagged is_primary = no in the dataset.
Axis-calculated figures. Six figure families (19 CSV rows) are computed by Axis Intelligence Research from sourced inputs, each with its formula shown inline: the Continuous Grid Load series (seven readings); the Italian queue-to-CGL ratios; the 10% realisation scenario; the flexibility realisation rate and coverage band; the pipeline realisation rate; and the segment share column. Each is flagged axis_calculated = yes in the CSV with a method note.
Comparability. ENTSO-E and Ember use different definitions of “Europe” and different base data; both trajectories are published side by side rather than merged, and no average of the two appears anywhere on this page. Ember’s national and city shares derive from ICIS market data, not from metered national statistics. Where a mismatch affects a conclusion, it is stated at the point of use.
About This Dataset
Contents. 106 observations covering European data centre electricity demand trajectories, installed IT power by segment and country, national and metropolitan consumption shares, grid connection queue volumes and wait times, utilisation rates, flexibility potential, and Axis Intelligence Research calculated figures including the full Continuous Grid Load series.
Temporal coverage. 2015–2050 (historical observations and published projections).
Spatial coverage. Europe, European Union, Ireland, Netherlands, Germany, United Kingdom, France, Italy, Denmark, Belgium, Norway, Sweden, Spain.
License. CC BY 4.0. Free to use, redistribute and build on, including commercially, with attribution.
Citation:
- APA — Axis Intelligence Research. (2026). Europe data center power demand 2026: The gigawatts the grid must actually serve. Axis Intelligence.
- MLA — Axis Intelligence Research. “Europe Data Center Power Demand 2026: The Gigawatts the Grid Must Actually Serve.” Axis Intelligence, 2026.
- Chicago — Axis Intelligence Research. “Europe Data Center Power Demand 2026: The Gigawatts the Grid Must Actually Serve.” Axis Intelligence, 2026.
Frequently Asked Questions
What is Continuous Grid Load (CGL) and why use it instead of TWh?
CGL is the Axis Intelligence Research metric that restates annual data centre electricity demand as the average continuous capacity, in gigawatts, that the power system must hold available all year: annual TWh × 1,000 ÷ 8,760. Grid planning, generation adequacy assessments and substation sizing all work in megawatts, not terawatt-hours. For Europe, CGL runs 9.9–11.0 GW in 2024 and 22.7–29.0 GW in 2035.
Why is a data centre connection queue not a demand forecast?
Because queue entry is cheap and non-binding. Developers submit applications at several sites simultaneously, hold positions beyond any feasible development horizon, and withdraw without penalty in most jurisdictions. Terna holds 95.38 GW of requests and has eight data centres connected. ENTSO-E recommends financial guarantees, project maturity milestones, withdrawal penalties and capacity release mechanisms specifically to separate credible load from speculative reservation.
What load factor should a grid planner assume for a new European data centre?
Considerably below nameplate. European data centres used 44% of their annual nominal power rating in 2024 on IEA and McKinsey data; EirGrid measures average draw at about 34% of contracted capacity; Norway reported 150 MW drawn against 500 MW connected in 2023. A facility contracted for 50 MW may operate at 20–30 MW for years, which is why phased contractual capacity is displacing single-step firm allocation.
Does data centre load threaten generation adequacy in Europe?
Not yet in most regions, and already in some. ENTSO-E’s assessment is that data centres do not currently pose a generation adequacy challenge across most of Europe, but that projected growth makes it a near-term concern, with Ireland and several US regions already demonstrating it. The nearer-term risk is transmission capacity and dynamic stability in concentrated clusters, not continent-wide energy shortfall.
What is the difference between a firm, non-firm and phased grid connection?
A firm connection guarantees full contracted capacity and waits for reinforcement. A non-firm or flexible connection delivers reduced firm capacity plus conditional capacity that the operator may curtail under defined grid conditions. A phased connection ramps contracted capacity in predefined increments matched to lease-up. The trade is speed against certainty: US precedent cited by Ember shows phased approaches cutting queue time by around 80%, from five to eight years down to 12–18 months.
Why do UPS systems make data centres a stability risk rather than just a big load?
Because they can drop hundreds of megawatts instantaneously. In double-conversion mode a UPS behaves as a constant-power load — drawing more current as voltage falls, worsening a dip — before transferring to battery and disconnecting entirely at shallow thresholds. Reconnection is not standardised, creating disconnect-reconnect flapping. ENTSO-E identifies fault ride-through, RoCoF withstand, ramp-rate control and oscillation damping as the connection requirements needed to address this.
Which European markets have the shortest time to power in 2026?
Italy, the Nordics and Belgium. Italian connection waits run about three years against seven to ten in FLAP-D. Denmark, Norway and Sweden combine low congestion, low power prices and cold-climate cooling advantages, with Nordic data centre demand set to triple by 2030. Ember finds that markets offering half the connection time of incumbent hubs capture roughly 20% more growth by 2030.
How much of Europe’s electricity demand growth is data centres?
About a tenth. IEA modelling attributes 10% of EU electricity demand growth to 2030 to data centres under current policy settings. In absolute terms, Ember puts European data centre demand growth to 2030 at 72 TWh — larger than electric vehicles at 67 TWh and comparable with electrified industry at 80 TWh.
Can Europe still triple its data centre capacity by 2030?
Not on current modelling. The AI Continent Action Plan targets at least tripling EU capacity in five to seven years; the announced pipeline is worth 130% of installed capacity, but IEA modelling projects only 70% growth by 2030 — a pipeline realisation rate of about 54% by Axis Intelligence Research’s calculation. A doubling is the credible outcome, with grid congestion the named constraint.
What should I cite as the headline number for European data centre power demand?
For annual energy: 87 TWh in 2024 rising to 134 TWh by 2030, attributed to ENTSO-E; or 96 TWh to 168 TWh on Ember’s ICIS-based series. For installed capacity: 12.7 GW of IT power across more than 10,500 European facilities, attributed to ENTSO-E. For the load a grid must actually serve, cite the Axis Continuous Grid Load: 9.9–11.0 GW in 2024, 22.7–29.0 GW in 2035, attributed to Axis Intelligence Research.
Related research from Axis Intelligence Research: Europe Data Center Statistics 2026 · Data Center Electricity Demand Tracker · Data Center Moratorium Tracker · AI Data Center Tracker · Data Center Statistics · AI Capex Tracker · Tech Energy Research