In brief
Choose cooling around the IT load, rack density, allowable temperatures, availability target and heat-recovery opportunity. In direct-to-chip systems a CDU typically transfers heat between the technology cooling system (TCS) and the facility water system (FWS). Residual heat into room air must also be accounted for.
A data center is a process plant. It moves heat, fluids and energy through closed loops, and it has to keep doing so without interruption year after year. Seen from the compute side it is servers. Seen from the plant side it is pumps, heat exchangers, valves, instrumentation and automation. Uptime rests on the second list, not the first.
Cooling is the most critical support system in a data center and, after the IT equipment itself, its largest single energy consumer. Rising compute density has changed the picture fast. Where a conventional rack drew a few kilowatts, a rack built for AI workloads can draw tens of kilowatts or more than a hundred. Densities at that level are no longer cooled with air. The heat is moved into liquid as close to the chip as possible.
This guide brings the whole cooling picture together: how the cooling methods differ, how a liquid-cooled plant is built, how performance is measured and verified, how waste heat becomes a product, and what the system demands from maintenance and condition monitoring. Deeper articles on each area are linked throughout.
What a buyer should provide
For an early scope review, share the expected IT load, expansion phases, temperatures, FWS/TCS schematic, resilience target and responsibility matrix. Finkova can then discuss the installation and commissioning interfaces relevant to its delivery.
Why cooling decides data center economics
Cooling affects the return on a data center through four separate routes.
Availability. A loss of cooling shows up in the hall in minutes, not hours. A liquid-cooled hall heats up faster than an air-cooled one because there is less air mass to buffer the rise. Protection, redundancy and condition monitoring are part of cooling design, not an accessory to it.
Energy consumption. Cooling is typically the largest non-IT cost item. Its efficiency drives the site’s PUE directly, and with it the electricity bill.
The value of waste heat. Every kilowatt-hour fed into a data center leaves it as heat. Liquid cooling delivers that heat at a higher and steadier temperature than air cooling, which is what makes district heating recovery viable. In Finland this became more significant when data center electricity moved from the reduced tax class II to the general class I on 1 July 2026. Heat sold to a district heating network is a revenue stream that offsets part of the operating cost.
Regulation. The EU Energy Efficiency Directive (2023/1791) requires data centers with at least 500 kW of installed IT power to report their energy performance annually to the European database. The reported indicators include PUE, WUE, ERF and REF, with calculation methods set out in Commission Delegated Regulation (EU) 2024/1364. Sites above 1 MW are additionally required to make use of their waste heat where this is technically and economically feasible. In practice this means the metering has to be designed to be reportable before the first kilowatt-hour is consumed.
How the cooling methods differ
Air cooling: suits IT loads for which airflow, space and temperature can be controlled with the chosen containment strategy.
Direct-to-chip: removes heat at cold plates. Size residual room air cooling as well.
Immersion: requires compatible IT hardware, fluid and service procedures.
Free cooling and dry coolers: assess climate, allowable supply temperatures and peak-load backup.
Heat recovery: evaluate temperature, annual hours, buyer requirements and alternative heat rejection.
Liquid cooling changes how the plant is designed
Moving to liquid is not a component swap. It brings piping, pressure, flow and leak risk into the white space, which is to say it brings in exactly the things process industry has managed for decades. The design questions are loop separation, material selection, jointing, leak detection, and how the hall is serviced without stopping the compute.
The differences between direct-to-chip and immersion cooling, hybrid arrangements, and what the shift means for the structure of the hall are covered in the article liquid cooling in the data center.
The CDU is the heart of a liquid-cooled plant
The Coolant Distribution Unit separates the server loop from the facility water loop. It contains the heat exchanger, pumps, filtration, expansion and control, and it holds the server loop steady regardless of what happens on the facility side.
A CDU has to manage three things at once: supply temperature, flow and pressure. TCS and FWS pressures are specified against OEM limits and leak scenarios. A heat-exchanger leak flows from the higher-pressure side to the lower-pressure side. Lower TCS pressure does not prevent FWS fluid from entering TCS. The temperature is held above the hall’s dew point so that moisture does not condense on cold pipework. Neither of these is fine tuning. Both are conditions whose failure damages equipment.
CDU types, sizing criteria, installation and commissioning are covered in the article CDUs in the data center.
TCS and FWS: two loops and the boundary between them
A liquid-cooled site splits into two loops. The Technology Cooling System (TCS) is the closed loop from the CDU to the equipment. The Facility Water System (FWS) runs from the CDU to heat rejection, whether that is dry coolers, chillers or heat recovery.
ASHRAE’s TC9.9 committee classifies the conditions in these loops. The facility side has the W classes W17, W27, W32, W40, W45 and W+, where the number is the upper limit of the supply water temperature in degrees Celsius, with corresponding S classes for the equipment-side TCS loop. The higher the class, the larger the share of the year the site runs on free cooling, and the more valuable the recovered heat. The upper limit comes from the equipment manufacturer’s requirements, so the class is agreed with the hardware supplier rather than chosen at the design desk. Loop architecture, redundancy and scope splits are covered in the article on TCS and FWS: cooling loop architecture and scope splits.
Water chemistry decides how long the system lasts
A liquid cooling loop is small, tight and mixed in its materials. Cold plate channels are narrow, and the loop may contain copper, stainless steel, aluminium and a range of seal and hose materials. That creates three risks: particles block the channels, dissimilar metals drive corrosion, and stagnant warm water allows microbial growth.
All three are manageable, but only if fluid chemistry, filtration and flushing practice are specified at design stage and monitored in operation. In practice that means a defined fluid specification, monitoring of inhibitors, correctly sized filtration and a documented fill and flush procedure. The subject is covered in the article liquid cooling water chemistry, filtration and material compatibility.
Measurement is what makes performance provable
Cooling performance cannot be assessed without flow and energy metering. Flow tells you whether each loop is getting what it was designed for. Flow combined with temperature difference gives the transferred power, and from that comes both the real efficiency of the cooling and the quantity of heat recovered.
In data centers the measurement is often ultrasonic, taken from outside the pipe. Clamp-on metering causes no pressure drop, has no moving parts and does not require the loop to be opened, which makes it usable in an existing hall as well as a new one. A portable meter handles surveys, loop balancing and verification of fixed meters, while fixed metering serves continuous energy monitoring and reporting.
The differences between measurement technologies are covered in more depth in the complete guide to industrial flow meters, and data center specific measurement points in the article flow and energy metering in data center cooling loops.
Product options are in our flow meter range.
Waste heat becomes a product once the temperature is high enough
An air-cooled hall returns heat at a low temperature and spread thin. Liquid cooling concentrates it into one loop at a higher temperature, which is the precondition for any recovery. A district heating network still requires a considerably higher temperature than the server loop produces, so an industrial heat pump sits in between.
Whether recovery pays rarely comes down to heat pump efficiency alone. What decides it is the distance to the network, the network’s temperature level and seasonal profile, the price paid for the heat, and how recovery affects cooling availability through the summer months. These questions are worked through in the article waste heat recovery and sales to district heating. The control and safety valves and pumps around the recovery loop are part of the same package.
The indicators: PUE, WUE and ERF
PUE expresses total facility electricity against IT consumption, WUE expresses water consumption against IT consumption, and ERF the share of site energy that is reused. The formulas are simple, but comparability depends entirely on measurement boundaries and the period measured. The same site can show different numbers depending on where the meter sits and which twelve months you take.
Now that these numbers carry a reporting obligation, the measurement boundaries are worth fixing at design stage rather than in the week before the first reporting deadline.
Condition monitoring keeps capacity online
The rotating machines in a cooling system, meaning chiller compressors, pumps and fans, are the part of the plant that breaks. The most critical machines get machinery protection that trips them before damage occurs. The wider fleet is served by continuous or wireless condition monitoring, which sees a developing bearing fault weeks before it causes an outage.
A vibration baseline taken at commissioning is central to this. Without a baseline, later measurements give you a value but not a trend. Condition monitoring in a data center context is covered in the article data center cooling: failure modes and condition monitoring, and the available solutions on our safety automation and condition monitoring page.
How a cooling solution is selected
Cooling is a plant design decision rather than a product choice. The factors that matter are:
- compute density and how it is expected to develop over the life of the site
- the supply temperature and flow required by the hardware, meaning the ASHRAE class
- the redundancy requirement, and what a single fault is allowed to stop
- the share of the year free cooling is available in the local climate
- the potential for heat recovery and the distance to a district heating network
- water consumption and environmental permit terms, noise included
- serviceability in operation, meaning what can be maintained without stopping the hall
- measurement and reporting requirements
Of these, density growth and serviceability are the two most often underweighted. Both come due later, and expensively.
Commissioning decides whether the design works
Cooling turns from a drawing into a plant during commissioning. Flushing, filling, loop balancing, flow verification, protection testing and alarm walkdowns are the stages where design errors and installation deviations surface. If they do not surface there, they surface in production.
The handover baseline is worth measuring and documenting: flows loop by loop, temperature differences, machine vibration levels and energy meter readings. Every later measurement is compared against these. Without them, a performance discussion later on is an exchange of opinions. The commissioning stages and handover documentation are covered in the article on data center cooling: commissioning and handover.
Finkova’s scope and next step
Finkova installs and commissions data center cooling in the Nordics, with instrumentation and lifecycle support. The work scope, OEM requirements and acceptance criteria are agreed for each project.
Contact Finkova with the project inputs above to discuss the relevant installation, testing and handover interfaces.
Frequently asked questions
How is a data center cooled?
A data center is cooled with air, with liquid, or with a combination of the two. Air cooling suits moderate densities up to rack loads whose airflow and temperature can be controlled. Denser AI and HPC racks are liquid cooled, with heat moved straight from the processor into liquid. Most projects today are hybrids.
How much energy does data center cooling use?
Cooling is typically the largest energy consumer in a data center after the IT equipment itself. Its share depends on the cooling method, the climate and the temperature level. In Finland free cooling cuts consumption considerably, because outdoor air can reject the heat for much of the year without mechanical cooling.
What is liquid cooling?
Liquid cooling moves heat into a liquid as close as possible to where it is generated. The most common approach is a cold plate fitted on the processor, with coolant flowing through it. Liquid carries heat far more effectively than air, which makes it suitable for dense compute racks.
Why are data centers built in Finland?
From a cooling perspective there are two reasons. Finland’s cool climate allows free cooling for much of the year, which lowers cooling energy use. And waste heat can be put to use in extensive district heating networks, turning it into saleable energy.
What does PUE mean?
PUE, Power Usage Effectiveness, expresses a data center’s total electricity consumption relative to the consumption of its IT equipment. The closer it is to one, the smaller the share of electricity going to cooling and other support systems. Comparability depends on where and over what period it is measured.
Do data centers have to report their energy use?
Yes, if the site has at least 500 kW of installed IT power. The EU Energy Efficiency Directive requires annual reporting of energy performance to a European database. The reported indicators include PUE, WUE, ERF and REF, with calculation methods set out in Delegated Regulation (EU) 2024/1364.
Can data center waste heat be reused?
Yes, and liquid cooling makes it considerably easier, because heat is recovered at a higher temperature. A district heating network still needs a higher temperature than a data center produces, so an industrial heat pump sits in between. For sites above 1 MW, reuse is an EU obligation where technically and economically feasible.
What is a CDU?
A CDU, Coolant Distribution Unit, is the central piece of equipment in a liquid-cooled data center. It separates the loop serving the servers from the facility water loop and holds the server loop’s temperature, flow and pressure at setpoint, protecting the cold plates from facility loop pressure and contamination.
What are the ASHRAE W classes?
The ASHRAE W classes W17, W27, W32, W40, W45 and W+ describe the upper limit of facility loop supply water temperature in degrees Celsius. The higher the class, the more of the year a site runs on free cooling and the more valuable the recovered heat. The limit comes from the hardware manufacturer’s requirements.
How is a cooling solution chosen?
A cooling solution is a plant design decision, not a product choice. The key factors are compute density and its growth, the supply temperature the hardware requires, the redundancy requirement, the share of free cooling, the potential for heat recovery and serviceability in operation. Density growth and serviceability are most often underweighted.
