The Inflexible Load
What electricity actually costs a European data centre per megawatt hour reaching the IT equipment — the energy a contract can replace, the network charge it cannot, and the cost of the hours when the system is short and the facility is still drawing.
A data centre cannot reduce load when power is scarce. That is the point of the asset: it runs. So its exposure is driven by the tail rather than the average price. That tail sits in the half of the European market published under an open licence. This page is a working draft circulated for correction, not a product and not a quotation for power.
This page prices one year. The same measurement across eleven years is at the foot of this page, including the year the scarcity tail was more than half the bill.
Cost per IT‑megawatt‑hour
Delivered price and cooling, plus the expected cost of hours above the threshold. Move the threshold and watch the order hold or break.
“We don’t pay those rates.”
Correct — and it touches only one of the three bars. A power purchase agreement fixes the energy component. It does not touch the network charge, which is set by the transmission and distribution operators and is the same for every consumer in your band. That charge runs from €4.10/MWh in Belgium to €30.00 in Germany — a sevenfold spread no contract negotiates away.
Nor does a PPA remove the tail. A pay-as-produced contract delivers the generator’s profile, not yours; whenever output undershoots a flat load, the shortfall settles at the imbalance price. A hedged buyer should read the amber bar more closely than an unhedged one, not less.
Switch the basis above to My contract and type your strike: the published energy price is replaced by yours, and the network charge and the tail stay.
Price duration curve
Imbalance price against the share of the year it was exceeded. Both axes logarithmic, because the question lives in the last one per cent. Hover to identify a zone.
Same cost, different risk
The closest pair on delivered cost with the widest gap in tail. This is what an average price cannot show.
Five zones refused
Each is reported with its reason rather than estimated. An imputed row would make the table longer without making it more reliable.
Every figure behind the charts
Method
What is measured, what is assumed, and what is missing.
Why imbalance price, not day-ahead
ENTSO-E opened an enumerated list of 35 article references under CC BY 4.0. The day‑ahead price is not on it, and neither is actual generation, installed capacity or actual load. Imbalance price (Art. 17.1.g) is, and for a load that cannot switch off it is the better measure anyway: it prices being short when the system is short. Category A05 throughout.
The tail is a bound, not a forecast
It assumes full exposure to imbalance in the scarcity hours — the conservative case. A facility on a PPA or a hedge bears less. That is why the tail segment is hatched and the delivered segment is not.
Level
Eurostat nrg_pc_205, non‑household electricity, 2025‑S2, band IG
(≥150,000 MWh/yr), excluding taxes and levies. A 50 MW IT load draws 438 GWh a year before
cooling, so it belongs in the top band; quoting it at a smaller band's price misprices the
comparison. Cooling is modelled from ERA5 cooling degree days, 2015–2025.
The split, and its two vintages
Components come from Eurostat nrg_pc_205_c, band IG, annual
2025. The headline price is semi-annual 2025‑S2. They are different
periods, so the two are reconciled per country and the residual is
published in the table — it runs between −6% and +6%. On the
published basis the split is an apportionment; on a contract basis the
network charge is used directly.
Not in this index
Grid connection queues, land, construction, fibre, carbon intensity, tax treatment and redundancy. Several of those decide a siting question on their own. Ireland's binding constraint today is connection availability, not price.
Zones, not countries
Norway settles in five bidding zones, Sweden four, Italy six. Oslo is NO1, Stockholm SE3, Milan IT‑North. Germany publishes imbalance under a different code from the DE‑LU zone every price query uses. Each resolution is recorded with a confidence, not assumed.
Reproducibility
Every series is fetched through a cache that records when it was retrieved, parsed by position rather than by order, and weighted by settlement duration — several zones moved from hourly to quarter‑hourly settlement during 2025, one of them mid‑month. Corrections welcome and will be published.
The eleven-year panel2015–2025 · the year the tail was 55% of the bill
The tool above prices 2025 and finds that scarcity risk barely reorders Europe. That is what 2025 shows. Run the same measurement over eleven years and it does not hold.
of the delivered cost of a megawatt in France, 2022 — the scarcity tail alone
between Austria’s smallest and largest annual tail ($1.01 in 2017, $163.45 in 2022), counting only years where the smaller figure clears a dollar
year-on-year transitions changed a zone’s rank — every adjacent pair of years, every zone publishing in both
zone-years survived the coverage and stated-position floors; the rest are listed below
Eleven years, ten zones
Delivered price and cooling plus the expected cost of hours above 200 EUR/MWh, in US dollars per megawatt hour reaching the IT equipment. Lines break where a zone-year was refused. One FX rate across all years, so the panel measures power rather than currency.
The lines move together. A siting comparison measures the gaps between them, and those gaps are small next to the distance every zone travelled. A buyer choosing between two markets in 2020 was weighing ten or twenty dollars. Within two years the whole surface moved by three hundred.
The tail is not a small number that stays small
In calm years the scarcity tail is worth a few dollars against a delivered cost of one to two hundred: 0.18 dollars in France in 2019, against a delivered 68. That is the number our own tool reports for 2025, and it is why the tool concludes that scarcity barely matters.
In 2022 the same measurement for France returned $150.20 — 55 per cent of the total cost of a delivered megawatt, and 839 times the 2019 figure.
This applies to our own tool as much as to anyone else’s. A one-year comparison puts a number on the tail, and in a calm year that number says the risk is immaterial. Over eleven years the same measurement ranges across three orders of magnitude. One observation cannot support an underwriting case for it.
The ranking did not hold
Across every adjacent pair of years, counting only zones that published in both, 29 of 57 transitions changed a zone’s rank — 51 per cent. The reordering happens in most years, not only between the two endpoints.
Austria travelled the furthest: cheapest zone on the board in 2015 and the most expensive in 2020. Netherlands travelled the full width too, between 2016 and 2023. A siting decision taken in either zone’s good year, and held, would have been taken on the wrong side of the board within 5 years.
The two endpoints tell the same story more narrowly: of the 3 zones observed in both 2015 and 2025, 2 finished in a different position.
| Zone | 2015 | 2016 | 2017 | 2018 | 2019 | 2020 | 2021 | 2022 | 2023 | 2024 | 2025 |
|---|---|---|---|---|---|---|---|---|---|---|---|
| AustriaVienna | 661.6 | — | 601.0 | 700.3 | 792.6 | 792.1 | 13824.3 | 421163.5 | 28825.8 | 21620.7 | 19718.8 |
| BelgiumBrussels | — | 802.3 | 623.8 | 765.4 | 601.5 | 571.7 | 13726.6 | 405140.2 | 18218.7 | 13121.3 | 13211.2 |
| GermanyFrankfurt | — | — | — | — | — | — | — | — | 22122.9 | 19723.2 | 17911.0 |
| Denmark DK2Copenhagen | — | — | — | — | — | — | — | 38091.7 | 1547.2 | 1499.1 | — |
| SpainMadrid | 1050.0 | 870.0 | 1010.0 | 1070.0 | 870.2 | 690.0 | 14415.7 | — | — | — | 1370.5 |
| FranceParis | — | — | 630.2 | 620.3 | 680.2 | 720.5 | 11626.7 | 274150.2 | 1309.9 | 1189.4 | 999.4 |
| NetherlandsAmsterdam | 870.0 | 770.0 | 550.0 | 590.0 | 640.0 | 736.0 | 12629.4 | 343141.1 | 29235.1 | 18843.5 | 14917.8 |
| Norway NO1Oslo | — | — | — | — | — | — | — | 12646.2 | 651.4 | 571.4 | 551.0 |
| PortugalLisbon | — | — | 860.0 | 800.0 | — | 770.9 | 15617.1 | 31836.4 | 1392.4 | 1207.4 | 1311.9 |
| Sweden SE3Stockholm | — | — | — | — | — | — | — | 16239.7 | 992.3 | 821.6 | 9716.1 |
Exposed US dollars per IT-megawatt-hour, with the scarcity tail beneath each figure. Em dash marks a zone-year that did not survive the guards.
What could not be measured, and what that says
51 of 121 zone-years were refused rather than estimated. Which ones failed, and why, follows a pattern worth reading on its own.
Deep history exists only for western Europe. Germany publishes no usable imbalance series before 2023. Norway, Sweden and Denmark have nothing usable before 2022. An eleven-year European power comparison is therefore not available to anyone, at any price, for those markets — and a vendor quoting one has estimated it.
Italy is unmeasurable throughout. In every year from 2022 the Italian system-short series carries a published value on between zero and one per cent of its settlement positions; the rest would be filled by carrying the previous value forward. Spain fails the same test in 2022, 2023 and 2024 and passes in 2025.
What this is for
An investment committee underwrites a fifteen-year hold on a power case. The record here says that case is built on the least stable input available, that the input is not stable even in rank order, and that for several markets the history required to test it does not exist.
Building anyway can be the right decision. What the record argues for is quoting a range instead of a point, and keeping the vintage — so that when the committee revisits the case in three years it can still answer what was knowable at approval.
Sources and method. Imbalance prices: ENTSO‑E Transparency Platform,
Article 17.1.g, category A05 (system short), monthly requests aggregated per calendar year —
Source: ENTSO-E Transparency Platform, licensed under CC BY 4.0. Indicate any changes made. Delivered price: Eurostat nrg_pc_205, non-household consumers,
consumption band IG (150,000 MWh a year and above), excluding taxes and levies, both semi-annual
halves averaged for each year. Cooling: ERA5 reanalysis via Open-Meteo.com (Copernicus Climate Change Service).
FX EUR/USD 1.1567 held constant across all years, so the panel measures power and not
currency.
Guards. A zone-year is reported only if at least 90 per cent of the year is published and at least 50 per cent of its settlement positions carry a stated value rather than a value carried forward. Statistics are weighted by settlement duration, because several zones moved from hourly to quarter-hourly settlement inside the period.
What this is not. These are realised outcomes, not vintages. ENTSO‑E
resettles imbalance prices and the platform serves the current revision; no snapshot of what it
said in 2021 was held, because the crawl did not exist then. From 18 August 2026 the retrieval
record fixes that going forward. Reproduce with
python analyses/exposure_history.py. Corrections welcome.
Tell us where this is wrong
This is a working draft circulated for review. If you site, finance or supply power to European data centres: which figure does not match what you see, and what would you need added before it were useful — connection queues, firm-power pricing, a specific zone, your own load profile?
Or write directly to tnk@newwaycapital.com.
Sources. Delivered electricity price: Eurostat nrg_pc_205,
non-household consumers, period 2025‑S2, consumption band IG (150,000 MWh a year and above),
basis excluding taxes and levies. Price components: Eurostat nrg_pc_205_c, band IG,
annual 2025. Imbalance prices: ENTSO‑E Transparency Platform, Article 17.1.g, category A05,
2025 calendar year — Climate: weather data by
Open‑Meteo.com (ERA5 reanalysis, Copernicus Climate
Change Service). FX
Figures are United States dollars per megawatt hour delivered to IT equipment, at the stated load, price basis and scarcity threshold. New Way Capital Advisory, Geneva. This page carries no investment advice and is not an offer to supply power.
Practices
Contact
- tnk@newwaycapital.com
- +41 76 381 47 48
- Geneva, Switzerland
Prices and imbalance figures are records drawn from published statistical and market sources. They are not quotations, forecasts or investment advice, and no representation is made that any figure reflects the terms available to any particular buyer.