Data Centre Desalination

Dublin has a water problem. It also has a data centre problem. What if the waste heat from one could solve the other?

I put together a brief on the idea of using waste heat from data centres to provide desalinated water to Dublin. The concept is straightforward: data centres generate enormous amounts of heat that currently gets vented into the atmosphere or dissipated through cooling systems. Desalination — the process of turning seawater into fresh water — requires heat energy. The two problems are complementary.

Ireland is uniquely positioned for this. We have a coastline, a growing demand for fresh water in the capital, and an expanding fleet of data centres that the grid is already straining to support. If we're going to tolerate the energy and water consumption of these facilities, we should at least be capturing the waste heat and putting it to productive use. Otherwise, we're just heating the sky and wondering why the taps are running dry.

Dublin's Water Problem Is Not Future Tense

Dublin's water supply is not "near capacity." It is past it. The Greater Dublin Area can sustainably produce about 600 million litres a day. Demand hit 636 million litres in January 2024. It hit 659 million in January 2025.¹ Eighty-five per cent of what comes out of the taps starts in the River Liffey. The recommended engineering buffer — 10–15% spare capacity for maintenance, demand spikes, and the unexpected — does not exist. There is no margin.

Dublin has been growing steadily for twenty years and is not about to stop. The CSO's 2024 projections put population growth at 8–25% by 2042, depending on migration.² Ireland's success in attracting tech, pharma, and finance FDI has stacked commercial water demand on top of residential. Martin Shanahan, the former IDA CEO, told the Business Post that water concerns make it "much more difficult" to win overseas investment.³ And much of the infrastructure — Poulaphouca, Vartry, the Bohernabreena Reservoirs on the Dodder — was designed for a smaller city with fewer demands. If one source gets contaminated or a drought hits, there is very little to fall back on.

The fix has been in the works for two decades. Uisce Éireann's Eastern and Midlands Region Water Supply Project — the Shannon pipeline, basically — reached the planning application stage in December 2025, with construction proposed for 2028. It could cost up to €10 billion.⁴ In the meantime, housing developments have been slowed or blocked because local authorities know the water isn't there. That feeds directly into rising housing costs and the affordability crisis that defines Dublin politics. In the summer of 2018, demand hit 615 million litres a day against the system's limit, and the hosepipe ban went national.⁵ That was seven years ago. The supply has not materially improved.

What Data Centres Throw Away

Ireland is one of the largest data centre markets in Europe relative to its population — AWS, Google, Meta, Microsoft, and others all operate here.⁶ Data centres accounted for 22% of Ireland's electricity demand in 2024. That is projected to hit 31% by 2034.⁷ The IEA projects the country's total electricity demand will nearly double by 2035.⁸

That growth creates problems. But it also creates heat — enormous quantities of it, thrown away. The newer cooling systems — direct liquid cooling, where coolant runs right across the processors instead of blowing air at them — produce waste heat hot enough to actually be useful. We are talking 50–60°C, depending on the workload, though some of the latest AI chips are pushing that back down again because they run so hot that the coolant has to start cooler to keep up.⁹ The point is: the heat is there, it is significant, and right now it goes nowhere.

What Membrane Distillation Actually Is

Right, so MD — membrane distillation — is a way to separate salt from water using heat instead of pressure. You have a membrane made of a material that lets water vapour through but blocks liquid water — PTFE or PVDF, typically. You heat the seawater on one side. The other side stays cool. The temperature difference creates a vapour pressure difference, and water vapour crosses the membrane and condenses as fresh water on the other side. Salt stays behind.¹⁰

There are different configurations — Direct Contact, Air Gap, Vacuum, Sweeping Gas — but they all use the same principle: moderate heat does the work, not the massive mechanical pressure that reverse osmosis needs. MD can technically run with feed water as low as 40°C, but it gets noticeably better the hotter you go. Going from 50°C to 70°C roughly halves the energy cost per unit of water.¹¹ The 50–60°C that DLC waste heat provides is inside the operating range, but at the inefficient end.

A pilot in Australia ran MD on waste heat below 40°C from a gas-fired power station for three months and hit 92.8% water recovery.¹³ That matters because it proved the concept works with real industrial waste heat, not just under lab conditions.

The Money

MD water is expensive.

Large-scale RO produces water at $0.50–$1.50 per cubic metre. Recent Middle Eastern plants have got that below $0.50. MD with free waste heat — the best-case scenario — comes in at $1.60–$4.10/m³, and can hit $9.60/m³ in less favourable setups.¹⁴ Uisce Éireann charges businesses €2.19/m³. Even with the heat for free, the membranes need replacing, the pumps consume electricity, and maintenance is not optional. Free heat is not free water.

The only study that has looked specifically at data centre waste heat for MD desalination — a 2025 Lund University thesis on a 1.5 MW DLC facility in Barcelona — concluded it was "technically feasible but faces economic challenges, largely due to high costs associated with pump investment and operation in relation to the low water production achieved by the low waste heat temperatures."¹⁵ Selling the water to the grid was not economical.

There is also an obvious opportunity cost. Data centre waste heat already has a proven, profitable use: heating homes. Meta heats buildings in Odense, Denmark. Microsoft and Fortum target 40% of Espoo's district heating in Finland. AWS in Tallaght already supplies heat to civic buildings.¹⁶ Choosing desalination over heating means choosing the unproven option over the one that works.

But that assumes the choice exists equally everywhere, and it doesn't. Dublin has almost no district heating infrastructure. The Dublin District Heating project is in early development, covering Poolbeg and the Docklands. That's it. Denmark and Finland have decades of heating networks built out. Ireland does not. You cannot sell waste heat to a network that isn't there. And building a heating network from scratch takes as long as building a desalination plant — probably longer.

There is a more fundamental problem with the RO comparison. Building a reverse osmosis plant adds new electricity demand to a grid that is already at breaking point — data centres already take 22% and rising.⁷ The IEA says demand will nearly double by 2035.⁸ MD driven by waste heat adds almost no grid load, because the thermal energy is a byproduct of computing that is already happening. In a country where the grid is the constraint on everything — housing, industry, the data centres themselves — that distinction matters more than the price per cubic metre.

And then: who pays? An RO plant is a new public infrastructure project — another call on the state, another decade of planning, another strain on the exchequer. MD driven by data centre waste heat can be funded by the companies generating the heat. Microsoft, AWS, and Meta have all committed to being "water positive" by 2030.²⁴ They are already spending money on water replenishment programmes. This gives them something concrete and local to spend it on — and something that buys them the social licence they desperately need in a country that imposed a moratorium on their growth.

Nobody Has Done This

No. Nobody has combined data centre waste heat with membrane distillation for desalination anywhere in the world.

Data centre waste heat recovery for district heating is real and growing. Cloud&Heat Technologies, a Dresden company, built distributed data centre systems that capture waste heat for building heating. Their showcase project — the Eurotheum in Frankfurt — uses servers on two floors to heat offices and a hotel, with water coming out at 60°C.¹⁷ Using waste heat as a resource rather than a disposal problem is an established idea.

What has not been proven is directing that heat toward desalination. The Australian pilot proved MD works with industrial waste heat.¹³ The Lund thesis proved it is technically feasible with data centre heat specifically.¹⁵ But the combination has never been deployed — not even as a pilot. The district heating lessons apply in principle, but the desalination application remains untested.

Why hasn't anyone tried? Three reasons, as far as I can tell. First, district heating is a better-proven and more profitable use of the same heat. Second, RO is cheaper and more mature — if a city needs desalinated water, RO powered by renewables is the established path. Third, MD has not reached commercial scale for anything: the largest MD pilot plant in the world produces about 400 m³/day.¹⁸

The pure efficiency argument is simple: RO makes cheaper water, district heating makes better use of waste heat. That is true in the abstract. It is not true in Ireland, where the grid cannot support a new RO plant, the district heating network does not exist, and data centres face a political environment that treats them like parasites. The most efficient solution is not always the one that can actually get built. Sometimes the imperfect thing that the right people are willing to pay for is the one worth pursuing.

And there is a timing argument that the efficiency comparison misses entirely. Nobody has combined these things because nobody has planned a new data centre build with desalination integrated from the start. The existing fleet was not designed for this. But new capacity is coming — the moratorium is lifting, the CRU has published its new connection policy, and operators are actively looking for sites. There is a window, right now, to say: if you are going to build here, build near the coast, and build with MD from day one. The lead time on a data centre is 2–5 years. The lead time on the desalination infrastructure is closer to 10. If you want them to arrive together, you start both now.

The Geography Is Awkward

Dublin's existing data centre fleet is inland. Microsoft at Grange Castle, 15 km from the coast. AWS in Mulhuddart, 12 km. Meta in Clonee, 25 km. The four major clusters — Grange Castle, Mulhuddart/Blanchardstown, Clonee, and Clonshaugh/Tallaght — are all 7–25 km from the sea. Piping waste heat from any of them to a coastal desalination plant means 30 km of pipeline with thermal losses that could gut the whole premise.

But the argument here is not about the existing fleet. Data centres have a 2–5 year lead time from planning to operation. Infrastructure like this — pipelines, treatment, marine intake — has a lead time closer to 10 years. The proposal is to plan them together. Site new data centre capacity at the coast, with desalination integrated from the start. Not retrofitting something that was never designed for it. Building something that is.

Poolbeg West is designated for 3,500 homes. Dublin Port is mid-expansion. Those are spoken for. But there is coastline between Clonshaugh and Portmarnock, stretches around Howth, and sites further north into Fingal that are not. The point is not to squeeze a data centre into Ringsend. It is to find a site — probably in Fingal, probably not glamorous — where the sea, the grid connection, and the planning conditions can all be made to work. And to start looking now, because the lead time is the entire argument. If you wait until the crisis is worse, you are another twenty years out.

Regulation — The Tailwind and the Wall

Ireland's Climate Action and Low Carbon Development (Amendment) Act 2021 commits the country to net-zero by 2050 and encourages circular economy practices — waste heat reuse fits.¹⁹ The National Planning Framework recognises water infrastructure as critical.²⁰ Uisce Éireann's Water Services Strategic Plan supports innovative supply sources.²¹ Grand.

But there is a wall too. The CRU imposed a de facto moratorium on new data centre grid connections in Dublin from 2021 to 2025. The replacement policy, published in December 2025, requires new data centres to install on-site generation covering their full electricity demand and source 80% of annual consumption from additional Irish renewables within six years.²² Any proposal that depends on expanding data centre infrastructure in Dublin needs to account for that.

Here is the tailwind. From July 2026, the EU Energy Efficiency Directive requires new data centres above 1 MW to reuse waste heat. From July 2028, at least 20% must be reused — or operators have to prove through a cost-benefit analysis that it is technically or economically unfeasible.²³ And the directive does not limit qualifying uses to heating. The EU guidance explicitly lists seawater desalination alongside district heating, swimming pools, and industrial processes. This is important because data centres run every hour of every day — but heating demand is seasonal. In summer, when Dublin's water demand peaks and drought risk is highest, nobody needs the heat for warmth. The cooling does not stop. The mandate does not pause. Desalination gives the heat somewhere useful to go all year round.

Coastal infrastructure adds further regulatory layers: marine licences under MARA, Environmental Impact Assessment, Appropriate Assessment under the Habitats Directive, Foreshore Licence applications, marine spatial planning compliance. The Shannon pipeline spent more than twenty years going from concept to planning application. Calling these "procedural hurdles" would be dismissive. They are multi-year, multi-agency processes that can kill projects entirely.

Who Would Pay, and Who Would Build It

The structure would be a public-private partnership between three parties: Uisce Éireann, a data centre operator, and the state. The operator builds and runs the facility — data centre and desalination together. Uisce Éireann takes the water into the mains. The state provides planning support, absorbs the regulatory risk it created with the moratorium, and coordinates through the local authority. Microsoft, AWS, and Meta have all committed to being "water positive" by 2030.²⁴ They are already spending money on replenishment programmes. This gives them something concrete and local to spend it on. The European Investment Bank has €15 billion earmarked for water infrastructure through 2027.²⁵ Horizon Europe could fund the research end.

Think of what Bord na Móna did in the midlands — a state company pivoting industrial assets toward renewable energy, with private partners and EU funding. This would be similar in shape. Not retrofitting. Building something new with a different purpose baked in from the start. Ireland's PPP track record is mixed — the Western Building Systems scandal, the motorway tolls — and that should inform how this one is structured. But the alternative to a PPP is not a better model. It is no model. The state will not fund this alone. The companies will not build it alone.

What a Realistic Timeline Looks Like

The data centre build and the desalination infrastructure need to be planned in parallel, not sequentially. A data centre takes 2–5 years from site selection to operation. The marine and environmental assessments for a coastal desalination intake take 18–36 months alone — a full annual cycle of seasonal ecology data, coastal process studies, EIA scoping, marine licensing through MARA. If you start both at the same time, they arrive at roughly the same time. If you wait for one to finish before starting the other, you are looking at a decade before water flows.

Phase one — 18 to 36 months: a joint feasibility study between Uisce Éireann, a data centre operator, and the relevant local authority. Site analysis, marine assessment, process design, cost-benefit modelling. The data centre operator does site selection in parallel — coastal Fingal is the obvious area. EU funding through Horizon Europe or the Innovation Fund could support the research end. This is when the PPP terms get negotiated.

Phase two — the data centre builds while the pilot runs. A credible first step for MD is 100–500 m³/day — ambitious, but within an order of magnitude of what has actually been done. The largest MD pilot ever built produces about 400 m³/day.¹⁸ For perspective, Dublin consumes about 600,000 m³/day. Nobody is claiming this replaces the mains. The pilot proves whether waste heat at the temperatures this specific facility produces can drive MD at a cost that makes the full build worthwhile.

Phase three: scaling, if the pilot works. Modular MD expansion, integration into Uisce Éireann's distribution network, and — critically — a template that can be replicated at the next coastal data centre, and the one after that. This is not one project. It is the first project.

Why Bother?

After everything above, the honest question is: why pursue this at all? The economics are marginal. The technology is immature. Nobody has tried it. The geography is awkward. Why not just build an RO plant?

Because this is not only about the most efficient way to make water. It is about whether Ireland is willing to do something imperfect now that provides a net benefit over fifty years. Efficiency — the relentless optimisation of every decision for cost, scale, the theoretically best outcome — is itself part of the problem.

Ireland has spent twenty years pursuing the efficient solution to its water crisis. The Shannon pipeline is the efficient solution. It has been in planning since the early 2000s. The planning application was lodged in December 2025. Construction might begin in 2028. Dublin's taps will not see a drop of Shannon water for years after that — if it survives the opposition, the judicial reviews, and the cost overruns that accompany every major Irish infrastructure project. Twenty years of searching for the optimal answer, and the city is further past capacity than when the search began.

This is the pattern. Housing cannot be built until the water supply is secured. The water supply cannot be secured until the pipeline is approved. The pipeline cannot be approved until every objection is addressed. Meanwhile the population grows, demand rises, and the crisis deepens. The pursuit of the efficient, optimal, comprehensive solution becomes the reason nothing gets done. Perfect crowds out good enough.

Data centre desalination is not the efficient answer. It is an imperfect, partial, expensive, unproven answer. But new data centres are coming — the moratorium is lifting, operators are looking for sites, and the CRU has published its new connection policy. There is a window to say: if you are going to build, build near the coast, build with desalination from day one, and fund it through the PPP that Uisce Éireann, the state, and the operator negotiate together. It does not add load to a broken grid. It satisfies a legal mandate the EU is imposing regardless. And it could produce water within a decade rather than within a generation. Over fifty years, a net benefit — even a small one, even an expensive one — compounds. The question is not whether this is the best way to make water. It is whether doing something now is better than waiting for the best way.

The moratorium on data centre connections did not happen because of electricity alone. It happened because the public sees data centres as parasites — consuming power, consuming water, consuming land, contributing nothing visible to the communities around them. Whether that perception is entirely fair is beside the point. It is politically real, and it constrains every planning application, every grid connection, every expansion. A data centre that can say "we make fresh water for Dublin" changes that conversation. The value is not just the water. It is the permission to exist.

Dublin's water crisis is not going away. The Shannon pipeline, if approved, is years from delivering and may cost €10 billion. Climate change is making droughts more frequent. The population is growing. Every new housing development and every new data centre adds demand to a system that is already past it. The 2018 drought forced a nationwide hosepipe ban. January 2025 saw 659 million litres demanded against a 600 million litre supply. This is not a future problem.

The EU is now mandating data centre waste heat recovery — and desalination is explicitly listed as a qualifying application.²³ That mandate does not pause in summer. Data centres do not pause in summer. But heating demand does. In a country with almost no district heating network, desalination may be the most practical year-round use for waste heat that the directive requires to be used. The companies have to do something with it. This gives them something worth doing.

Membrane distillation is pre-commercial today. So was solar energy twenty years ago. The cost curve is early. A country that invested in a pilot now — even a small, imperfect one — would be positioned to scale when the economics shift. And the economics will shift, because the EU mandate creates guaranteed demand for waste heat applications, because membrane materials are advancing, and because the cost of doing nothing with the heat will rise as regulation tightens.

The alternative is not RO. Ireland does not have the grid capacity for a new large-scale RO plant. It does not have the district heating network to absorb waste heat. It does not have twenty years to wait for the Shannon pipeline. What it has is a coastline, a growing demand for data centre capacity, and a water system that is past breaking point. The next data centre is going to be built somewhere. The question is whether it gets built with desalination integrated from the start — funded by the company building it, producing water for the city that hosts it, satisfying a legal mandate that is coming regardless — or whether it gets built the same way as the last one, and we heat the sky again.

This is not a blueprint. It is an argument for a feasibility study — properly funded, honestly scoped, run with the expectation that it might conclude the idea doesn't work. But it is also an argument against waiting for the perfect answer while the taps run dry. The only way to find out is to try to prove it wrong.