Andris Gailitis

Cloud, Colo & Data Center Pro

Not How Much. How Many Days.

A data centre beside a river at dusk, with rooftop cooling towers venting vapour, pump house and pipework along the bank, and cooling water discharging back into the watercourse.

Disclosure: I am the CEO of Delska, which operates a data centre in Riga – a cool-climate site, the category this article concludes in favour of. Read it with that in mind. And Latvia is the one Baltic state with no published water figure at all, mine included. That is set out below rather than left out.

Key takeaways

  • The useful question is not litres per kilowatt-hour. It is how many days a year a site uses any cooling water at all.
  • Microsoft reports 0% of the year in parts of Northern Europe and up to 40% in Phoenix. AWS uses no water for 95% of the year in Ireland and Sweden, and introduces it only above roughly 29 °C.
  • In the EU’s 2024 reporting table, Belgium’s 16 data centres took in more water than Germany’s 370.
  • Estonia reports 0.891 m³/MWh, above Germany. A cool climate permits near-zero water. It does not guarantee it.

The standard sentence about data centres and water is that they consume enormous amounts of it. The sentence is not wrong so much as incomplete, and the missing part changes the answer entirely. Water use is not a property of a data centre. It is a property of a site’s climate and its cooling design, and in some geographies the correct figure is zero – not low, not improving, zero.

The useful question is therefore not how many litres a facility uses per kilowatt-hour. It is how many days a year it uses any water at all. Two of the largest operators in the world publish that number, and it varies by a factor of eight across their own fleets.

The ladder

OperatorLocationShare of the year water is used
MicrosoftParts of Northern Europe0%
AWSIreland and Sweden~5% – no water for 95% of the year
MicrosoftDublin, Amsterdamunder 5%
AWSGlobal fleet~10% – free-air cooling about 90% of the time
MicrosoftVirginia10%
MicrosoftPhoenixup to 40%

Same company, same engineering standards, same reporting year. The only variable is latitude.

AWS adds the mechanism behind its own figure, which is more useful than the figure: water is introduced only when the ambient air exceeds roughly 29 °C. That is not an average, it is a threshold. In Riga, Stockholm or Dublin it is crossed on a handful of afternoons. In Phoenix it is the working condition for a third of the year.

Smaller European operators are more specific still. Green Mountain runs its Stavanger-area site on a closed loop with no water consumption, and reports around 330 days of free cooling a year at Rjukan. EcoDataCenter uses cooling water in Falun only from mid-June to mid-August, and none at all in Stockholm and Piteå. Hetzner cools with air alone and publishes a WUE of 0 litres per kilowatt-hour.

The finding nobody wanted

Here is the part that should temper any enthusiasm, including mine.

Of twelve major operators surveyed, nine publish no time-fraction at all. Google, Equinix, Digital Realty, Vantage, CyrusOne, STACK, QTS, NTT and Iron Mountain disclose an annual WUE – Equinix 0.91, Digital Realty 0.59, QTS 0.44 – or a count of waterless facilities. None of them publishes the fraction of the year during which water is used.

So this cannot be written as “the industry acknowledges that geography decides.” It does not. The one metric that answers always, or not comes from three companies out of twelve, one of them qualitatively. An annual litres-per-kilowatt-hour average cannot distinguish a site running evaporative cooling continuously from one that switches it on for six weeks – and that is the distinction the whole public argument turns on.

The regulator says it out loud

This is not only an operator talking point. The European Commission’s September 2026 report on data-centre energy efficiency – the first output of the EU reporting database created under the Energy Efficiency Directive – states the mechanism directly: water-based and evaporative systems consume substantial amounts of water, while air-side and liquid-direct designs operate with minimal to no water use, and cooler regions generally need less water cooling.

Behind that report sits a public dashboard, and it is not a chart. It is a table: reporting country, number of facilities, total water input, weighted WUE. For 2024 it covers 829 data centres across 22 countries.

Data-centre water intensity by EU member state, 2024Weighted water usage effectiveness in cubic metres per megawatt-hour from the EU Energy Efficiency Directive reporting database, reporting period 2024. Hungary is lowest at 0.001 and Belgium highest at 1.357. Estonia, at 0.891, is above Germany at 0.486. Latvia has no published value in any reporting period.A cool climate permits near-zero water.It does not deliver it.Data-centre water intensity (WUE, m³/MWh) by EU member state, 2024Reporting member stateBaltic stategrey = facilities reporting00.250.50.7511.25Hungary70.001Finland280.009Lithuania30.022Sweden100.044Poland390.128Czechia40.142Italy330.235Greece50.252France1520.263Bulgaria30.304Austria180.364Luxembourg100.395Spain380.470Germany3700.486Portugal80.552Denmark170.571Netherlands610.877Estonia30.891Belgium161.357Estonia — 3 facilities,above Germany’s 370.Latvia1no published value in any period — 1 facility, suppressedSource: European Commission, EU data centre reporting database (Energy Efficiency Directive),public dashboard, country straight-table, 2024. 829 facilities, 22 countries. Cyprus, Croatia, Malta suppressed.
Weighted WUE by member state, EU reporting database, 2024. Estonia sits above Germany; Latvia has no published figure in any period.
CountryWUE 2024 (m³/MWh)Facilities reporting
Hungary0.0017
Finland0.00928
Lithuania0.0223
Sweden0.04410
Germany0.486370
Denmark0.57117
Netherlands0.87761
Estonia0.8913
Belgium1.35716

Finland to Belgium is a factor of 150, inside one regulated market, under one reporting rule, in one year. Germany’s 370 facilities – by far the largest sample, and the hardest number to dismiss – land at 0.486.

Read the same table as volumes and it sharpens further. Belgium’s 16 data centres took in more water than Germany’s 370: 1.93 million cubic metres against 1.86 million. Sixteen facilities out-consumed three hundred and seventy, because of how they are cooled and where they are.

At the other end, Lithuania’s three reporting data centres consumed 285 cubic metres in the year. Not 285 thousand. Two hundred and eighty-five – a domestic swimming pool, for a national sector. In 2023 the same three reported a water input of zero, which is where that year’s WUE of 0.000 comes from.

One caution before anyone quotes a single EU average. The report’s headline of 0.58 m³/MWh counts only the 458 facilities judged reliable that use water at all. The dashboard, counting all 829 reporting sites including the dry-cooled ones, gives 0.45. Same dataset, two cuts, and they are not interchangeable.

Where the Baltic numbers are, and where one is not

Two of the three Baltic states are now in the regulator’s table. Lithuania is at 0.022. Estonia is at 0.891 – higher than Germany, just under the Netherlands, in a climate where free cooling is available for most of the year.

That number deserves to stay in rather than be quietly dropped, because it corrects the lazy version of my own argument. A cool climate permits near-zero water. It does not deliver it. Three Estonian facilities reported 7,000 cubic metres between them – roughly what Finland’s twenty-eight used. On a sample that small, one evaporative design sets a national figure. Geography removes the obligation to use water. Somebody still has to decide not to.

Latvia has no figure in any reporting period. It appears exactly once in the entire database – 2023, one facility – and every metric in that row is suppressed, because values are withheld for fewer than three reporting data centres. The Latvian blank is therefore a documented disclosure rule rather than a failed search, which is a different and better kind of gap. Cyprus, Croatia and Malta are blank in 2024 for the same reason.

Ireland and Norway are absent from the database altogether: Ireland reports data-centre electricity in detail but has no public water reporting of this kind, and Norway sits outside the directive. Germany goes further than the EU table and publishes down to facility level, 370 sites deep.

What actually deserves the worry

Even the litre is a slippery unit. Published WUE figures mix withdrawal with consumption – water taken and largely returned against water evaporated and gone – and a litre means nothing without the stress of the basin it came from. A litre evaporated in Arizona and a litre returned to a Norwegian fjord are not the same object.

Research published in October 2026 puts the real exposure at siting: roughly a quarter of the world’s data centres – 2,315 of 9,347 studied – sit in basins that run a water gap for at least one month a year, rising to a third under three degrees of warming. That is a map problem, not a technology problem, and it is being legislated as one: around a hundred bills across 39 US states now touch data-centre water.

The question to ask

For the first time there is a regulator’s table with country names on it, and it says what the operators’ own disclosures say: this is a geography and a design choice, not an industry constant. Lithuania’s entire reporting sector used 285 cubic metres in a year. Belgium’s weighted average is 150 times Finland’s.

But the table still reports litres, and litres were never the question. If you are assessing a facility, ask the two things its operator knows exactly and almost nobody prints: how many days last year did this site use cooling water, and above what outdoor temperature does it start?

Estonia is the reminder that the answer is not given to you by the map. At 0.891 it sits in a cold country near the top of the European table. The climate had made zero possible. Somebody chose otherwise.

Every figure in this article comes from a published source. The full evidence base – the operator time-share table, the EU country table, the signal grading and each source URL – is published alongside this article as a separate document, including the three figures for which no public URL exists and which are marked as such.

Evidence base: Data Centre Water – Evidence Base (PDF) – every figure in this article with its source URL, the operator time-share table, the full EU country table for all 22 reporting countries, and the grading of each signal.

Get the next research first

I publish one evidence-based research report most weeks – free, no paywall, full PDF attached. Subscribe to Cloud, Colo & Data Center Pro and the next one reaches you the day it is out.

#DataCenters #WaterUsage #LiquidCooling #AIInfrastructure #DigitalSovereignty #Baltics

Originally published on LinkedIn: https://www.linkedin.com/pulse/how-much-many-days-andris-gailitis-wk6yf/

Related articles:
Liquid Cooling vs Air Cooling in Data Centers: The Real Economics
EU Data Centre Regulation Tracker: Energy, Heat Reuse and PUE
Does AI Still Need Tier III? 59 Claimed It, Seven Could Prove It