Information dive: Dodgy information derails datacentre water debate
Growth of datacentre capability – currently pushed by demand for synthetic intelligence (AI) processing – has sparked rising concern concerning the {industry}’s influence on water provides. With datacentres set to multiply, some are warning that the shift in the direction of high-density computing may set off unsustainable calls for on water.
However the debate is characterised by an info hole. On one hand, we will discover narratives that depend on worst-case “do-nothing” eventualities and cooling fashions that will not precisely replicate the UK local weather or the evolution of contemporary {hardware}. Then again, the {industry} argues it has largely transitioned to cooling techniques that radically scale back water consumption, however discovering out from planning purposes what they really suggest to deploy is a tricky job.
This text makes an attempt to get below the pores and skin of those looming projections of a datacentre-driven water disaster. It appears to be like at water firm information concerning the provide scenario 20-plus years therefore, the nuances of present cooling design, and the obvious info gaps within the planning system that imply we lack clear industry-wide information and hamper the flexibility to precisely forecast the true influence of digital enlargement.
Push for mandated reporting
Final 12 months’s report from the Authorities Digital Sustainability Alliance (GDSA) seemed intimately at water consumption related to AI and datacentres.
The report makes some glorious suggestions, however among the supporting arguments quantity to over-egging the pudding. Right here we confer with its use of future water deficit predictions and the extent of datacentre water use, which we are going to take care of later.
The report, Water use in AI and information centres, makes its case round just a few key elements, which embody:
- Datacentres devour water via cooling techniques, electrical energy era, and the manufacturing of semiconductor chips wanted to equip them.
- England is projected to face a each day water provide deficit of almost 5 billion litres by 2050, an quantity equal to greater than one-third of the present public water provide.
- Many datacentres use “evaporative cooling”, which depends on steady replenishment with recent, potable water.
- The calls for of AI processing, notably throughout heatwaves, create the necessity for water-intensive cooling on the time provides are most constrained.
- There’s a geographical overlap between water-stressed areas and areas the place datacentres are clustered.
- Present nationwide water infrastructure plans don’t account for the water necessities of datacentres.
- There’s a “transparency deficit” within the {industry}, wherein a minority of datacentre operators monitor their water utilization.
To fight these, the report’s appendix recommends mandating all UK datacentres above 1MW to report location-specific, granular information on water, power and carbon utilization; integrating datacentre water demand into nationwide water useful resource planning; mandating high-efficiency cooling applied sciences in water-stressed areas and offering incentives for utilizing non-potable and recycled water sources; proscribing new datacentre developments in water careworn areas; and introducing necessary water effectivity targets and disclosure necessities for cloud and datacentre suppliers.
The report’s suggestions are clear and would make a distinction. Water use reporting by industrial services will not be necessary within the UK, so no unified set of knowledge exists. Additionally, anybody who has ever tried to divine projected water use in proposed datacentres by way of planning utility paperwork will know it’s, at greatest, typically obscured, and at worst, and typically, fully absent.
Getting an correct ongoing image of datacentre water use is important with regard to planning and consents for datacentre location, in addition to to construct future water provide infrastructure.
However there are causes to consider the report paints a barely grimmer image than obligatory.
Why belief one and never the opposite?
The GDSA report says England will face a water deficit of 5 billion litres a day in 2050. This quantity seems to come back from the Atmosphere Company’s Water Useful resource Administration Plan (WRMP).
The WRMP is the formal, legally mandated blueprint via which UK water corporations undertaking their capacity to fulfill buyer demand. It gives water supply-demand stability for at the very least the subsequent 25 years by water useful resource zone (the native community), water firm and geographical area, for England solely.
Information is cut up between “baseline” – what can be the case had been no enhancements made to infrastructure – and a “closing planning” dataset that displays water firm forecasts of provide and demand after they’ve carried out infrastructure enchancment throughout the planning interval.
The GDSA seems to get its “5 billion litres a day” determine by including up the WRMP baseline datasets supply-demand stability for 2049-2050. It equals a deficit of 4,947.7 million litres per day within the dataset up to date on 7 April 2026.
What the GDSA report doesn’t cite, nonetheless, is the ultimate planning dataset. In that, the April figures present a optimistic supply-demand stability for 2049-2050 of 895 million litres per day.
So, it seems the report cherry-picked figures to indicate a grimmer image than if it had additionally proven the WRMP closing planning information that predicts a water surplus.
Having stated that, to account for future datacentre demand over that interval and its impact on water provides isn’t any simple matter.
Many datacentres are in water-stressed areas
What we will see from the WRMP information within the above map (left) is that of the 101 whole water useful resource zones, 76% of them are projected to face some stage of water provide deficit by 2045-2046 if no infrastructure enhancements happen. We are able to categorise 67 of those as being areas of minor deficit (between zero and 100 million litres per day), eight as areas of reasonable deficit (-300 to -100 Ml/d) and two as extreme (< -300 Ml/d).
Then, if we plot the 119 deliberate datacentre tasks to those areas of water stress, 59% of them are inside zones at present projected to face a water deficit. These map to:
- Minor deficit: 29 datacentres.
- Average deficit: 17 datacentres.
- Extreme deficit: 24 datacentres.
Provisos listed here are that that is the “do nothing” situation, and that datacentres mapped to areas of water stress are solely the 119 at present in planning phases, with the furthest window into the long run reaching solely to 2037 (the 1GW Elsham Tech Park in North Lincolnshire), whereas WRMP projections go to 2050 and past.
If it appears counter-intuitive that high-rainfall areas like Yorkshire, Lancashire and the Lake District ought to present as water-stressed, that’s as a result of water provide infrastructure is impacted by environmental safety. Water corporations want authorized permission to take water from rivers and aquifers to take care of wholesome river flows and shield ecosystems. There are additionally local weather change impacts, the place water corporations now plan for one-in-500-year drought occasions, and inhabitants development.
Whereas that was the do nothing situation, the second model of the identical map (proper) – with water corporations’ closing planning for infrastructure upgrades included – exhibits no water area in deficit and so ought to have the ability to deal with datacentres positioned there.
However that doesn’t account for future datacentre growth.
Based mostly on the expansion pattern from present information (1.6GW in 2024 to 8GW in 2037), datacentre capability within the UK is increasing at an annual development fee of slightly below 14%. If we undertaking these figures, we get to about 7GW by 2036 (4.5 occasions what it’s now) and 24GW by 2046 (15 occasions the present capability).
The query is, does the WRMP information consider this quantity of development?
The opposite query is, how a lot does that matter, given what we find out about datacentre water utilization? It’s an space the place it’s essential to get terminology proper and a area of fast change proper now, so it’s value taking a look at.
Cooling vs warmth rejection
In line with the GDSA report, “evaporative cooling” is a “frequent methodology” in datacentres, and requires “steady replenishment with recent water”, “utilises potable (ingesting high quality) water” and may contribute to “considerably decrease water ranges and influence fish and different aquatic ecosystems within the surrounding areas”.
All these issues – bar one – are undoubtedly true.
Earlier than we take a look at that – and for the needs of readability – let’s distinguish between cooling and warmth rejection in datacentre design.
In datacentre design, “cooling” refers back to the strategy of transferring warmth away from IT tools to a secondary medium (normally air or liquid), whereas “warmth rejection” is the ultimate step of transferring that warmth out of the ability and into the exterior surroundings.
Whereas cooling typically makes use of closed-loop techniques to flow into water or refrigerant internally, the warmth rejection section is the purpose the place services determine between “dry” strategies, corresponding to mechanical refrigeration and warmth exchangers, or water-consumptive strategies, corresponding to evaporative or adiabatic cooling.
To return to the GDSA report’s factors, evaporative cooling – or extra strictly warmth rejection – is extremely wasteful of water. Its modus operandi is by way of water piped to an evaporative cooling tower and to a warmth exchanger, the place a relentless provide of water evaporates and carries away warmth.
It has been a preferred methodology of warmth rejection, notably within the hotter areas of the US. Evaporative warmth rejection locations much less demand on electrical energy and is well-suited to dry areas the place the air can soak up plenty of moisture.
The query is, nonetheless, to what extent evaporative cooling is used within the UK.
There isn’t any dataset to confer with right here.
Is evaporative cooling frequent?
What we will say is that many older and smaller datacentres – which make up a big proportion of the UK’s put in base – are much more prone to depend on air-conditioning/refrigerant strategies of warmth rejection. Of 190 present datacentres within the UK, 100 are of lower than 5MW capability and 40 extra are smaller than 10MW.
We additionally know that the present stage of evolution in chip design for AI graphics processing models (GPUs) implies that direct-to-chip cooling is now necessary. That’s as a result of the facility draw of the brand new generations of GPU processing is such that it operates at unprecedented wattages and ranges of warmth era and at larger temperatures on the chip than had been hitherto regular.
In direct liquid cooling (DLC), water is held in two closed loops just like a automotive radiator: one for cooling, one for warmth rejection. They’re stuffed as soon as. That may imply just a few hundred thousand gallons – at a fill quantity of roughly 10,000 to twenty,000 litres (10 to twenty cubic metres) per 1MW of compute capability – nevertheless it’s not ongoing consumption. As soon as it’s within the system, that’s it.
Within the newest datacentre designs, the warmth rejection loop goes out to dry coolers that blow ambient temperature air throughout warmth exchangers.
That additionally works as a result of the newest power-hungry chips can run at a lot larger temperatures, and so the warmth delta between cooling water and the surface air doesn’t should be as steep to convey the temperature down sufficiently.
That stated, techniques constructed with dry coolers typically use so-called adiabatic cooling for warmth rejection – the place water is used to spice up the warmth change course of – throughout very hot climate to take care of temperature spikes.
As for purely evaporative cooling, it’s not identified how prevalent that is within the UK. Most legacy datacentres depend on liquid cooling with air/refrigerant warmth rejection, whereas newer and bigger AI-capable datacentres will possible depend on DLC and dry warmth rejection, with some adiabatic throughout peak temperature intervals.
The prevalence of “evaporative cooling” within the UK is due to this fact speculative as a result of inadequate reporting information. Accordingly, evaluating the projected water footprint of recent UK services to the consumption profiles of desert-based US datacentres is tenuous.
Opaque planning purposes
It’s truly actually troublesome to find out what strategies of cooling can be utilized in datacentres which might be at present deliberate or below development within the UK as a result of the planning course of doesn’t mandate it, or at the very least such particulars should not compulsory till a undertaking is properly previous preliminary approvals.
We seemed on the high 10 largest datacentre tasks in information provided by development planning monitoring specialist Barbour ABI, after which checked out paperwork submitted with planning purposes.
Of those, solely the biggest – the aforementioned 1GW Elsham Tech Park – has any form of details about water use connected to its planning utility.
A lot of the relaxation lacked a lot in the way in which of paperwork submitted in any respect, not least when it got here to water use. That’s in all probability largely as a result of it’s not a requirement to set out something about water use at proposed datacentres till the detailed phases of native authority approval.
That’s not like power or visitors influence assessments, which have clear triggers. So, info on water is typically supplied throughout the detailed design section or in non-public discussions with utility suppliers somewhat than being a clear, publicly out there metric throughout preliminary public session and planning phases.
To search out out extra, we made contact with these behind two datacentres on the record – QTS’s 500MW web site at Cambois and Amazon’s 185MW (estimated) works on the former Didcot energy station.
Elsham, Cambois and Didcot
Elsham Tech Park supplied what seemed like a complete Water necessities report, wherein it informed us it’ll use “‘dry’ techniques, with water solely being required for the preliminary filling”.
Contained in the datacentre, it’s proposed {that a} chilled water system will hold temperatures down, with refrigerants taking up ought to temperatures rise past a sure stage. In the meantime, GPUs – that is an AI datacentre in an AI development zone, keep in mind – can be cooled by way of DLC with the secondary (warmth rejection) loop going to dry coolers outdoors the constructing.
It stated that in its “proposed design … it’s possible that for 99% of the 12 months, utilizing dry coolers alone can be enough for all the warmth rejection from the direct to chip cooling system… For the remaining 1% of the 12 months, chillers can be used to offer any extra cooling that’s required. There isn’t any extra water required for cooling, aside from the preliminary filling of the closed system.”
It stated annual water use can be 14,416m3, which incorporates places of work for as much as 900 folks, and is the equal of about 96 households’ yearly utilization within the area. It contrasts that to what it stated can be water use in an adiabatic system each year of 465,416m3 and in evaporative cooling of two.017 million m3 yearly.
Concerning QTS’s Cambois datacentre undertaking, the developer stated: “In 2025 alone, the QTS closed-loop cooling system saved an estimated 28 billion litres of water worldwide when in comparison with evaporative cooling.”
In the meantime, projected water use for the “preliminary section” – do they imply the fill? It’s not clear – at Cambois is roughly 2,268m3 and can use air-cooled chillers for warmth rejection with out adiabatic or evaporative cooling.
Amazon Information Service’s Didcot planning utility supporting paperwork said: “No water is used to chill the servers for 97% of the 12 months. It’s calculated that simply 96 hours of evaporative cooling are required yearly. Moreover, no potable water provide is used for industrial cooling functions, with all industrial water for cooling functions provided from an exterior water remedy facility on the broader Didcot energy station web site.”
The latter seems to have been obligatory as a result of Thames Water “recognized an incapability of the prevailing water community to accommodate the wants of this growth” in November 2025. In April 2026, it was introduced that water remedy agency Gradiant had been awarded a contract to design and ship a water remedy facility for the datacentre web site.
Arup evaluation: Trending away from evaporative and adiabatic
Within the absence of dependable details about the cooling and warmth rejection techniques in place and prone to be deployed in deliberate UK datacentres, we spoke to some {industry} consultants.
Gareth Williams, datacentre enterprise lead at design and structure agency Arup, stated most UK datacentres already use closed-loop techniques and reject warmth by way of refrigeration somewhat than evaporation, that the pattern is away from adiabatic and evaporative cooling, and that the UK’s local weather means water-assisted strategies are solely prone to devour water for a small fraction of the 12 months.
“Nearly all of UK datacentres already use closed-loop water techniques. These services use water internally however don’t devour it via evaporation. In that sense, ‘dry cooling’, the place there isn’t a internet water consumption for warmth rejection, is already properly established throughout the UK market,” stated Williams.
“The warmth from these closed-loop techniques is often solved via refrigeration somewhat than via water consumption. We might anticipate new UK services to proceed favouring these approaches. Broadly, there was a transfer away from adiabatic techniques in new UK datacentre designs, although they haven’t disappeared solely, and legacy techniques stay in operation,” he added.
“Lots of the very excessive consumption figures are based mostly on older legacy techniques within the US, the place the local weather is considerably completely different. Nicely-designed services working in a UK local weather have materially completely different efficiency.”
Patchy and fragmented planning info
The three datacentre tasks we reviewed illustrate a broader downside. Information concerning water utilization – and certainly, most particulars obligatory for large-scale infrastructure planning – is patchy and fragmented. Among the many 10 largest datacentres at present within the planning or development pipeline, substantial documentation was solely out there for Elsham, Cambois and Didcot within the planning portals of the related native authorities.
Even inside this small pattern, transparency is inconsistent. Solely Elsham disclosed its anticipated water utilization upfront. For the Cambois web site, we needed to contact the builders instantly, whereas the Didcot planning paperwork supplied utilization figures solely in percentages, somewhat than absolute volumes. Moreover, builders don’t universally disclose undertaking capability in megawatts – whereas Elsham supplied this determine, we needed to infer it for the opposite two websites based mostly on the proposed flooring space.
This opacity is compounded by the executive panorama. Planning purposes reside on separate, disparate council portals, with supporting info buried in unconnected PDF recordsdata – though that’s set to alter. As a result of the minimal info necessities for proposed datacentres are sparse and largely unstructured, collating this information for sector-wide reporting or nationwide infrastructure planning stays an arduous, if not unattainable, job.
Lack of transparency wants addressing
The controversy over datacentre water consumption is hampered by a big information hole. Whereas the GDSA report appropriately identifies the necessity for better {industry} accountability, its reliance on a “do nothing” baseline and its utility of high-consumption US cooling fashions to the UK market dangers skewing the image.
However the {industry}’s lack of transparency makes such debate inevitable. As a result of water utilization stays largely invisible till the ultimate phases of the planning course of, if in any respect, there isn’t a dependable technique to confirm the precise cooling methods of the UK’s deliberate infrastructure pipeline.
In the meantime, there’s a temporal mismatch. Whereas water corporations try and forecast demand many years into the long run, the datacentre {industry} evolves way more quickly.
The objective must be to make sure water infrastructure planning can hold tempo with digital development. Because the report recommends, this requires a compulsory, standardised framework to report operational water utilization on the level of utility. With out that, the true influence of the UK’s datacentre enlargement will stay a matter of hypothesis somewhat than rigorous planning.
