AI Data Centres Have a Water Problem. Liquid Cooling Is the Answer. Here Is Why FRP Is the Secondary Infrastructure Specification That Makes It Work.
Training GPT-3 consumed an estimated 700,000 litres of on-site cooling water. Equinix's Slough campus holds a groundwater abstraction licence for up to 72,000 litres per hour — the equivalent of 3,000 average UK households. Global AI-related water withdrawals could reach 6.6 billion cubic metres annually by 2027. The AI data centre boom has a water problem. The industry's response is liquid cooling. And liquid cooling has a secondary infrastructure problem that FRP solves.
Published by Reinforce Technology | 14 July 2026
The International Energy Agency forecasts that global data centre electricity consumption will more than double by 2030, reaching approximately 945 terawatt-hours, surpassing the annual electricity consumption of Japan (Lombard Odier, 2026). AI will be the principal driver, with AI-optimised facilities expected to see power usage quadruple within five years. In the UK, almost 100 new data centres are in the pipeline representing a 20% increase in capacity by 2030, with £47 billion of investment announced since 2023 (Data Centre Insight, 2025). The UK has already crossed the threshold at which data centres account for 5.9% of national electricity consumption, and the AI-driven expansion of the coming four years will push that figure substantially higher.
Electricity is not the only resource that AI data centres consume at an extraordinary rate. Water is the less-reported but equally significant dimension of the AI infrastructure challenge. Cooling alone can account for 30 to 40% of a facility's total energy use, and the thermal management of AI GPU clusters, which generate substantially more heat per unit of rack space than conventional server hardware, requires cooling systems that operate at intensities conventional data centre air cooling cannot match (Expert Assessment, 2025). The UK's Equinix Slough campus, one of the country's most important AI-ready infrastructure facilities, has obtained a groundwater abstraction licence for up to 72,000 litres of water per hour — equivalent to the hourly water footprint of approximately 3,000 average UK households (Aquaswitch, 2026). Global AI-related water withdrawals are projected to reach 4.2 to 6.6 billion cubic metres annually by 2027 (Expert Assessment, 2025).
The industry's response to both the thermal and water intensity of AI infrastructure is liquid cooling. Direct-to-chip cooling circulates non-conductive coolant through cold plates attached to processors, removing heat at source rather than relying on ambient air circulation. Immersion cooling submerges server hardware in dielectric fluid, providing even more efficient heat transfer. Rear-door heat exchangers capture heat from conventional rack air cooling before it enters the facility's air management system. Each of these liquid cooling technologies creates a new category of secondary infrastructure inside the data centre, and each of that infrastructure operates in a coolant-wet, high-voltage, and continuously operational environment where the material specification of cable management, structural supports, and access systems determines whether the facility performs for 25 years without secondary infrastructure maintenance events or generates them repeatedly from the first year of operation.
The Liquid Cooling Transition and What It Creates
Conventional data centre air cooling works by passing cold air through server racks and exhausting hot air into a raised floor or overhead plenum, from which it is returned to computer room air conditioning units. This system was adequate for server hardware generating 5 to 10 kilowatts per rack. AI GPU clusters generating 50 to 100 kilowatts per rack, and next-generation configurations targeting 200 kilowatts and beyond, generate heat at densities that air cooling systems cannot manage within the physical constraints of a standard data centre floor.
Liquid cooling addresses this by moving thermal management closer to the heat source. Direct-to-chip cooling attaches cold plates to individual processors, circulating coolant through a closed loop that carries heat to a facility heat exchanger. The coolant used is typically a glycol-water mixture or a specialist dielectric fluid. Immersion cooling submerges entire server chassis in tanks of dielectric fluid, with the fluid carrying heat to external heat exchangers by natural or forced convection. Both approaches require plumbing infrastructure that runs coolant from a central mechanical plant to individual rack-level heat exchangers or immersion tanks, creating an entirely new category of pipework, fittings, manifolds, and secondary structural support that does not exist in a conventionally air-cooled data centre.
That new infrastructure operates in a persistently wet environment. Coolant leaks, condensation from cold surface contact with humid data centre air, and the routine maintenance of coolant connections all mean that the secondary structural and cable management infrastructure inside a liquid-cooled AI data centre is in a persistently moisture-exposed environment, in close proximity to the high-voltage power distribution systems that supply GPU clusters drawing 50 to 100 kilowatts per rack. This combination — persistent moisture, high-voltage electrical proximity, and a requirement for maintenance-free performance across 25 years in a facility that operates continuously — is precisely the combination of secondary infrastructure demands where FRP's properties are most directly relevant.
Where FRP Is the Correct Specification in AI Liquid Cooling Infrastructure
1. Cable Management in High-Density GPU Environments
AI data centre cable management carries a combination of power and data cables that is more complex and more densely packed than conventional data centres. High-voltage AC distribution to power distribution units, DC bus systems connecting uninterruptible power supplies to the AI compute load, and the data networking cables connecting GPU clusters to storage and compute fabric all run through overhead cable management systems above the high-density server rows. In a liquid-cooled facility, that overhead cable management sits in an environment where coolant pipework also runs overhead, where condensation from cold coolant pipes in warm data centre air creates moisture exposure on nearby surfaces, and where the heat density of the compute rows below means the overhead environment has elevated temperature cycling.
FRP cable trays in AI data centre overhead environments are non-conductive in the high-voltage AC and DC distribution environment, eliminating the earthing and bonding requirements that steel cable management demands in proximity to the GPU power distribution systems below. They are corrosion-immune in the moisture-exposed environment created by overhead coolant pipework and condensation. And they are lightweight — approximately 75% lighter than steel equivalents — reducing the dead load on overhead cable management support structures in facilities where every kilogram of non-compute infrastructure loading is a kilogram that could otherwise support higher rack density (IntechOpen, 2022).
2. Structural Supports for Coolant Distribution Manifolds
The coolant distribution manifolds and pipework of a direct-to-chip or rear-door cooling system require secondary structural support running from the facility's mechanical plant to individual rack rows. These structural supports operate in the same moisture-exposed, high-voltage-adjacent environment as the cable management, carrying the additional dead load of coolant-filled pipework and manifolds that can be substantially heavier than equivalent data cabling when fully charged with liquid.
FRP pultruded structural profiles for coolant distribution support systems provide corrosion-immune, non-conductive structural sections that perform without degradation in the moisture and coolant exposure of the liquid cooling environment, do not require the earthing and bonding programme that metal support structures demand in proximity to GPU power distribution, and are compatible with the glycol and dielectric fluid coolants used in direct-to-chip and immersion cooling systems without the material compatibility concerns that some metal alloys raise in certain coolant chemistries (IntechOpen, 2022).
3. Immersion Cooling Tank Surrounds and Access Infrastructure
Immersion cooling tanks, in which server hardware is submerged in dielectric fluid, require safe access for maintenance personnel to retrieve and replace server chassis from the tank. The access infrastructure around immersion tanks, including walkways, steps, and handrails, operates in an environment where dielectric fluid spillage and vapour exposure is a routine maintenance condition. Metal access infrastructure in immersion cooling tank environments faces both the corrosion challenge of persistent fluid exposure and the slip hazard of dielectric fluid on metal grating surfaces.
FRP moulded grating around immersion tank surrounds provides anti-slip access whose surface performance does not degrade under dielectric fluid contamination. The integral grit surface maintains consistent traction in fluid-wet conditions. FRP handrail systems provide non-conductive, corrosion-resistant access barriers that do not require earthing and bonding in the high-voltage environment of an immersion cooling facility. And FRP's non-conductive property is specifically relevant in immersion cooling environments where the dielectric fluid itself is non-conductive, and where introducing conductive structural materials into the tank surround environment could compromise the electrical isolation properties of the cooling system.
4. Data Centre Perimeter and Cooling Plant Infrastructure
The external cooling plant of an AI data centre — the dry coolers, cooling towers, heat exchangers, and mechanical equipment that reject the heat absorbed by the internal liquid cooling system — operates in the same outdoor, weather-exposed environment as any other industrial mechanical plant in the UK. The structural and access infrastructure around outdoor cooling plant is exposed to UV, rain, and in many cases salt air, as significant numbers of UK hyperscale data centre sites are in or near coastal locations. FRP structural profiles and grating for outdoor cooling plant access provide corrosion-immune, maintenance-free secondary infrastructure that matches the 25-year operational horizon of the facility's cooling plant without the periodic recoating that galvanised steel structures in outdoor industrial environments require.
The Water Scarcity Connection
The water intensity of AI data centre cooling has become a genuinely significant policy concern in the UK. The Environment Agency's National Framework for Water Resources, published June 2025, projected a 5 billion litre daily water shortfall for England by 2055, and its modelling did not include data centre growth at all, despite ballooning estimates of the AI infrastructure capacity that will be needed (The Ecologist, 2026). With 28 planned UK data centre sites located in areas served by Thames Water, the largest single water supplier in the South East, the interaction between AI data centre water demand and regional water scarcity is a planning and regulatory concern that will become more significant as the current data centre pipeline is built out.
The transition to liquid cooling is partly a response to this concern. Closed-loop liquid cooling systems that recirculate coolant between the rack-level cold plates and an external heat exchanger consume substantially less water than open-loop evaporative cooling systems, because the coolant is not consumed in the heat rejection process. A survey by TechUK in August 2025 found that 51% of UK data centre sites already use waterless cooling systems (Aquaswitch, 2026). The transition to liquid cooling in AI data centres is accelerating for both thermal performance and water consumption reasons simultaneously.
The secondary infrastructure of that transition — the coolant distribution manifolds, the cable management in coolant-exposed overhead environments, the access infrastructure around immersion tanks, the structural supports for high-density liquid cooling plant — is the infrastructure where FRP's combination of non-conductivity, corrosion immunity, and coolant chemical compatibility is most directly relevant. As the UK's AI data centre pipeline builds out against a backdrop of regional water scarcity concerns and a regulatory environment that is scrutinising data centre water abstraction with increasing attention, the secondary infrastructure of the liquid cooling systems that address that water concern should be specified in materials that perform reliably for 25 years in the coolant-wet, high-voltage environments they occupy.
Reinforce Technology FRP Products for AI Data Centre Liquid Cooling Infrastructure
Reinforce Technology supplies FRP cable trays, structural profiles, grating, and handrail systems for AI data centre infrastructure across the UK, including liquid cooling environments, high-density GPU facilities, and outdoor cooling plant secondary infrastructure. Available in polyester and vinyl ester resin systems, non-conductive, lightweight, and corrosion-immune in the coolant-exposed environments of direct-to-chip, rear-door, and immersion cooling installations.
Contact us to discuss your AI data centre project and the correct FRP specification for your specific cooling configuration, rack density, and secondary infrastructure requirements.
Final confirmation of suitability for any specific data centre or liquid cooling application, including coolant chemical compatibility for specific coolant formulations, remains the responsibility of the appointed project engineer. Reinforce Technology provides technical guidance and material recommendations based on information supplied to us, but specification sign-off should always sit with the qualified professional responsible for the design.
References
Aquaswitch (2026) AI Data Centre Water Usage in the UK and Impact on Resources. Available at: https://www.aquaswitch.co.uk/blog/ai-data-centre-water-consumption/ [Accessed: 14 July 2026]. [Equinix Slough groundwater licence 72,000 litres per hour, equivalent to 3,000 UK households; 51% of UK data centres surveyed use waterless cooling systems].
Data Centre Insight (2025) 2026: UK's AI Data Centre Infrastructure Is in Full Acceleration. Available at: https://datacentreinsight.co.uk [Accessed: 14 July 2026]. [28 planned sites in Thames Water catchment; almost 100 new data centres in pipeline; 20% capacity increase by 2030; water usage identified as defining challenge alongside electricity].
Expert Assessment (2025) The Systemic Environmental Risks of Artificial Intelligence. Available at: https://arxiv.org/pdf/2512.11863 [Accessed: 14 July 2026]. [Cooling accounts for 30 to 40% of facility energy use; GPT-3 training consumed 700,000 litres on-site cooling water; global AI water withdrawals projected 4.2 to 6.6 billion cubic metres annually by 2027].
IntechOpen (2022) 'Fibre-Reinforced Polymer (FRP) in Civil Engineering', in IntechOpen Engineering Series. Available at: https://www.intechopen.com/chapters/84203 [Accessed: 14 July 2026]. [Non-conductive and corrosion-immune properties; 75% lighter than steel; no maintenance requirements in corrosive environments].
Lombard Odier (2026) Why Liquid Cooling Will Dominate AI Data Centres in 2026. Available at: https://www.lombardodier.com [Accessed: 14 July 2026]. [IEA forecast: global data centre electricity consumption to reach 945 TWh by 2030; AI facilities power usage to quadruple within five years; direct-to-chip cooling dominant approach].
NACE International (2016) International Measures of Prevention, Application and Economics of Corrosion Technology (IMPACT). Houston, TX: NACE International. Available at: http://impact.nace.org/economic-impact.aspx [Accessed: 14 July 2026].
The Ecologist (2026) Data Centres Cooling Drives Heatwave Demand. Available at: https://theecologist.org [Accessed: 14 July 2026]. [Environment Agency National Framework for Water Resources did not include data centre water demand despite ballooning AI infrastructure estimates; 5 billion litre daily water shortfall projected by 2055].
Younis, A., Ebead, U. and Judd, S. (2018) 'Life cycle cost analysis of structural concrete using seawater, recycled concrete aggregate, and GFRP reinforcement', Construction and Building Materials, 175, pp. 135-144. doi: 10.1016/j.conbuildmat.2018.04.183.
Comments