Data Center Liquid Cooling Design Guide: ASHRAE Classes, Facility Water, Sizing and Materials

Every new generation of AI hardware pushes rack power higher, and 50 kW and 100 kW racks are no longer unusual. Most of the attention goes to cold plates and GPUs, but in liquid cooling data center design the hardest things to change later usually aren't the cold plates. They're the decisions made early in design: the facility water temperature, coolant distribution unit (CDU) capacity and redundancy, and what the secondary piping is made of. Once the facility is built, those choices have to serve several generations of IT hardware. You can swap cold plates. You can't easily swap pipe or water temperature.

This guide walks through the design process step by step. We've already covered what a data center liquid cooling system includes and when you need one, and how much water a liquid-cooled data center uses. Here we follow the design sequence and show facility engineers, design firms and system integrators what to decide at each step, and which numbers to base it on.

Two terms come up throughout. The technology cooling system (TCS) is the coolant loop that runs from the CDU through the manifolds to the cold plates. The facility water system (FWS) is the loop that carries heat from the CDU to chillers, cooling towers or dry coolers.

Key takeaways

• ASHRAE's liquid cooling classes (W17, W27, W32, W40, W45 and W+) are defined by facility supply water temperature, and the class you choose largely decides whether you need chillers.
• TCS supply temperature equals facility supply water temperature plus the CDU approach temperature. This temperature stack ties the facility plant to chip junction limits.
• Size CDUs against four separate limits: heat exchange capacity, flow, pump head and redundancy.
• Stainless steel (304 or 316L, passivated) is the most common TCS piping material. Polymer piping is in use but still debated.
• Commissioning (pressure test, clean, flush, fill) largely determines whether the TCS stays clean for its service life.

What Are the Steps in Data Center Liquid Cooling Design?

Data center liquid cooling design starts from the IT heat load and sets facility water temperature, CDU capacity and the secondary loop in seven steps:

1. Build a liquid heat schedule.
2. Choose an ASHRAE liquid cooling class and heat rejection plant.
3. Set the temperature stack.
4. Size CDUs and the loop.
5. Lay out distribution and the sealed TCS loop.
6. Select piping and manifold materials.
7. Plan controls, leak detection and commissioning.

Step 1: Build a Liquid Heat Schedule

The first design input isn't total rack power. It's the heat that goes into the liquid. In a liquid-cooled server, cold plates remove most of the CPU and GPU heat, but memory, power supplies and network cards still shed some heat to air. So tally the liquid share rack by rack, then roll it up by CDU or by row. For how to split heat between liquid and air, see the liquid-air split section of our data center liquid cooling overview. For sustained chip power, margins and flow conversion, see liquid cold plate design calculations.

At the system level, this step also has to answer two questions:

• How will the build-out be phased? How many liquid-cooled racks go in on day one, how many come in each later phase, and will CDUs and main headers be built out up front or expanded in stages?
• How much headroom should you leave for the next generation? Header sizes, CDU positions and facility water connections are hard to change once they're installed, so plan them around next-generation rack density.

Step 2: Choose an ASHRAE Liquid Cooling Class and Heat Rejection Plant

ASHRAE TC 9.9 classifies liquid-cooled IT equipment by facility supply water temperature. These are the W classes. The original 2011 guidelines defined five classes, W1 through W5, with upper limits of 17°C, 27°C, 32°C, 45°C and above 45°C. The fifth edition of the Thermal Guidelines (2021) renamed them by their upper limit as W17, W27, W32, W45 and W+, and added a new W40 class. W40 reflects the growing number of liquid-cooled systems designed to run around 40°C, warm enough to make heat reuse practical. The minimum water temperature for every class is 2°C.

The fifth edition also clarified what compliance with a W class means: the IT equipment must run at full capacity, without throttling, across the class's entire temperature range. Previously, some manufacturers labeled equipment W4 if it worked anywhere between 32°C and 45°C. Under the new definition, a facility designed for W32 can support any W32-compliant equipment, which makes things easier for owners and design engineers to verify.

The table below follows the liquid cooling class table in ASHRAE's fifth-edition reference card, which lists the typical primary and supplemental facilities for each class.

How to Pick a W Class

Three factors drive the choice:

• The W class your IT equipment is rated for. This is a hard limit. Facility water can't run warmer than the class the servers you plan to buy are rated for.
• Local climate. The warmer the facility water, the more hours each year you can reject heat through dry coolers or cooling towers alone, and the less the chillers run.
• Whether you plan to reuse heat. If you want to send return water to building heat or a district heating network, aim for W40 or warmer. ASHRAE lists district heating as a supplemental facility starting at W32.

The industry trend is clear. Many vendors now design direct liquid cooling for inlet water around 45°C, and some say chillers aren't needed at that temperature. The same report notes, though, that most hyperscalers and large operators still keep chillers and run water at 30–35°C. When you read figures like these, check whether they refer to the coolant reaching the server or to facility water. If it's the coolant at the server, facility water has to run cooler by the CDU approach temperature, as Step 3 explains. Set the class too high and you narrow your future IT options. Set it too low and you pay for chillers for years. A safer approach is to design facility water for the class your target equipment supports today, while making sure the piping and heat exchangers can handle warmer water later. That leaves room to raise the temperature for the next generation.

Matching the Plant to the Class

Once the class is set, the heat rejection plant largely follows. W17 and W27 rely on chillers, with cooling-tower economizers as a supplement. From W32 up, ASHRAE lists cooling towers as the primary facility, with chillers kept as a supplement at W32 and W40 and dropped from the list at W45. ASHRAE's table doesn't name dry coolers, but warmer classes are also what make dry coolers practical for more hours of the year. Water consumption varies widely across these options. We cover that in our article on liquid-cooled data center water use, so we won't repeat it here.

One detail is easy to miss: if facility supply water can drop below the dew point of the data hall or mechanical room, insulate the facility-side piping and the primary side of the CDU heat exchanger against condensation.

Step 3: Set the Temperature Stack

The W class only defines facility supply water temperature. What matters at the chip is the temperature of the coolant the TCS delivers to the cold plates. The CDU heat exchanger sits between the two, and this temperature stack is the most important calculation in liquid cooling design:

TCS supply temperature = facility supply water temperature + CDU approach temperature

TCS return temperature = TCS supply temperature + TCS temperature rise

The CDU approach temperature is the difference between the TCS supply temperature and the facility water entering the CDU. Schneider Electric published a worked example when explaining CDU rating conditions:

From there, work toward the chip. The 30°C coolant enters the cold plate, where cold plate thermal resistance and the coolant's own temperature rise use up part of the temperature margin. Junction temperature has to stay below the chip's limit. For that part of the calculation, see cold plate thermal resistance. If junction temperature comes out too high, you have three levers: lower the facility water temperature (and lean more on chillers), tighten the CDU approach (with a larger, more expensive heat exchanger), or use a cold plate with lower thermal resistance.

The same source makes another point worth remembering. When you compare CDUs from different vendors, check the approach temperature and load point used for the capacity rating, and whether a filter was installed during testing. Ratings measured without a filter look better, but in service the filter's pressure drop eats into the available external pump head.

Step 4: Size the CDU and the Loop

Check four limits separately when sizing a CDU. If any one falls short, the CDU doesn't fit:

1. Heat exchange capacity. At the approach temperature set in Step 3, can it remove the full liquid heat load it serves?
2. Flow. For a system-level sanity check, the industry commonly uses 1.2–1.5 L/min per kW. The exact figure comes from heat load, coolant properties and allowable temperature rise. See liquid cold plate design calculations for the formulas. As a quick example, a CDU serving 1 MW of liquid heat load would need roughly 1,200–1,500 L/min of TCS flow by this rule of thumb.
3. Pump head. Size it for the worst-case circuit: the farthest rack, hoses, row and rack manifolds, quick disconnects, valves, and a dirty filter, not a clean one.
4. Redundancy. Critical loads often call for N+1 CDUs, and pumps and power supplies inside the CDU are often redundant too. Decide on redundancy during design, since it affects floor space and budget.

As for in-rack vs. in-row CDUs: in-rack units are more flexible for single-rack deployments, retrofits and phased rollouts, while in-row units save space and simplify maintenance for high-density rows or entire halls. For a detailed comparison, see what a CDU is and how it works and choosing a direct-to-chip cooling architecture. For early estimates, The Green Grid's CDU Sizing Tool calculates the required fluid temperature, total flow and total pressure drop across the cold plates in a loop.

CDU Sizing Checklist

Step 5: Lay Out the Distribution Network and Sealed TCS Loop

Downstream of the CDU is the secondary distribution network. A typical layout runs from the CDU to row manifolds, then rack manifolds, then quick disconnects and finally the server cold plates. Settle these details early:

• Isolation valves for each rack and each section of header, so you can take part of the system offline for service instead of shutting down a whole row.
• Drains at low points and air vents at high points.
• Locations for fill points, sampling ports and temporary flushing connections.
• Capped connections for future expansion, and a plan to flush those dead legs during commissioning.

For manifold construction and flow distribution, see what a liquid cooling manifold is. For quick disconnect selection, see our guide to liquid cooling quick disconnects.

Sealing the TCS: Pressurization, Expansion and Makeup

The TCS is a sealed loop. Coolant circulates continuously between the cold plates, manifolds and CDU. This is the facility-level secondary loop, not the self-contained closed-loop liquid cooling module inside a server. To keep the loop stable for the long haul, the design has to handle three things:

• Pressurization. Set a static pressure high enough that the highest point never goes negative and draws in air, but low enough that the lowest point stays within the pressure rating of the cold plates, hoses and quick disconnects.
• Expansion. Coolant expands and contracts with temperature, so you need an expansion tank or the CDU's built-in reservoir to absorb the volume change.
• Makeup and air removal. Disconnecting quick disconnects for service loses a small amount of fluid, so the loop needs a makeup system. A newly filled loop also needs continuous air removal, because trapped air reduces flow and heat transfer.

Step 6: Choose the Best Materials for Your Liquid Cooling System

The TCS has much stricter water quality requirements than the facility water system. Cold plate channels are narrow, so particles or corrosion products shed by the piping turn directly into blockages and added thermal resistance. The first question for any secondary-side material is whether it will put anything into the coolant.

Stainless steel is currently the most common choice for TCS piping. OCP's pre-commissioning guidelines for TCS row manifolds assume pipework of at least grade 304 stainless steel, with internal surfaces cleaned, pickled and passivated per ASTM A380, and welds made under an inert atmosphere.

The debate over polymer piping is still open, and both sides' arguments come mainly from manufacturers. An article sponsored by Tate, which makes stainless steel manifolds, states that ASHRAE's liquid cooling guidelines don't list polypropylene (PP) as a recommended piping material for the TCS loop, citing concerns about polymer degradation and additive leaching over long service lives. Polymer piping makers counter that polymers don't corrode or release particles into the loop, and that their light weight makes prefabrication easier. On a real project, ask for each candidate material's maximum temperature rating, compatibility data for your coolant, and long-term leaching test results before you decide.

Beyond the piping, plan the metals in wetted parts such as cold plates and manifolds as a whole. Mixing copper, aluminum and other metals creates a risk of galvanic corrosion. For a wetted materials checklist and galvanic corrosion control, see our direct-to-chip coolant guide.

Step 7: Plan Controls, Leak Detection and Commissioning

Controls and Monitoring

Define the loop's control logic during design:

• Supply temperature control. The CDU modulates its primary-side valve to hold TCS supply temperature. It also calculates dew point from room temperature and humidity, and raises the supply temperature automatically when condensation is a risk.
• Flow or differential pressure control. Pump speed adjusts with load to hold a target flow or differential pressure.
• Sensor placement. Monitor supply and return temperature, pressure and flow at the CDU, and ideally at each row manifold as well.
• Leak detection and isolation. Tie leak sensors to valves so a leak automatically isolates the affected rack or branch.
• Monitoring interface. The CDU should connect to your DCIM or building management system. DMTF is defining CDU control points such as target flow and target supply temperature for the Redfish standard. It's still a work in progress, but it's worth asking whether equipment supports it.

Commissioning Sequence

Commissioning quality largely determines whether a TCS loop stays clean over its life. The first four steps below follow the sequence in OCP's pre-commissioning guidelines, which were written for TCS row manifolds but apply well to the wider loop:

1. Pressure test. Hydrostatically test the loop at the pressure set in the design documents, typically 1.3–1.5 times working pressure. Fix any leaks and retest.
2. Clean. Circulate procedure water that meets the specification, then circulate a chemical cleaner until filter differential pressure stops rising for the specified period. Drain completely from the low points.
3. Flush. Fill, circulate and drain repeatedly until outlet conductivity and turbidity are close to procedure water. Flush every terminal point, including future expansion connections.
4. Fill. Fill with coolant soon after flushing. After enough circulation, sample from several locations to confirm the concentration is consistent.
5. Sample early and often. Sample frequently right after startup, then step down once readings stabilize. For sampling frequency and water quality limits, see our coolant guide.

Liquid Cooling Data Center Design Checklist

Designing Around Your Loop with APALTEK

Steps 3 through 6 depend directly on the IT-side hardware. Cold plate thermal resistance sets how warm the TCS supply can be. CDU heat exchange capacity sets how warm the facility water can be. Manifold and cold plate pressure drop set how much pump head you need. When these parts are bought separately and optimized in isolation, the temperature stack and pressure drop budget are the first things to fall out of line.

APALTEK designs and builds cold plates, manifolds and CDUs in its own factory, so these numbers can be worked out together in one design:

• Cold plates. Custom channel designs based on chip package, power distribution and target supply temperature. See our cold plate products.
• Manifolds. Designed around rack layout and flow requirements. See our manifold products.
• CDUs. In-Rack CDUs (5 kW and 150 kW) suit single racks and phased rollouts, and In-Row CDUs (1.5 MW and 2.5 MW) suit high-density rows.
• Testing and validation. Pressure and leak testing before shipment, with flow and pressure drop verification to project requirements.

Learn more about our AI data center liquid cooling solutions.

Liquid Cooling Data Center Design FAQ

What are the ASHRAE liquid cooling classes?

ASHRAE classifies liquid-cooled IT equipment into six classes by facility supply water temperature: W17, W27, W32, W40, W45 and W+. The number is the class's maximum supply water temperature in °C, and W+ means above 45°C. They correspond to the 2011 classes W1 through W5, with W40 added in the fifth edition, and the minimum water temperature for every class is 2°C. Equipment complies with a class only if it runs at full capacity, without throttling, across the class's entire temperature range.

What is the difference between ASHRAE W classes and A classes?

A classes (A1–A4) define the inlet air environment, including temperature and humidity, for air-cooled IT equipment. The fifth edition also added an H1 class for high-density air-cooled equipment, with a narrower recommended inlet range of 18–22°C. W classes define the facility supply water temperature for liquid-cooled IT equipment. A liquid-cooled server usually has to meet both, because some of its heat still goes to air and the room environment still has to stay in range.

What facility water temperature does direct-to-chip cooling need?

It depends on the W class of the IT equipment you choose. Many direct liquid cooling platforms now accept water of around 40–45°C, which reduces or eliminates the need for chillers. If that figure refers to the coolant at the server, facility water has to run a few degrees cooler, by the CDU approach temperature. Many large operators still hold water at 30–35°C to keep more margin. Design to the W class your target equipment is rated for, then work out the TCS supply temperature as in Step 3.

What is a good CDU approach temperature?

There's no single standard. Published vendor examples range from about 2°C to 7°C, and Schneider Electric's rating example uses 4°C. A tighter approach lets facility water run warmer for the same TCS supply temperature, which adds free cooling hours, but it takes a larger, more expensive heat exchanger. When comparing CDUs, always compare rated capacity at the same approach temperature and load point.

Is stainless steel or plastic piping better for the TCS loop?

Stainless steel (304 or 316L, pickled and passivated) is currently the most common choice for TCS piping, and OCP's pre-commissioning guidelines assume grade 304 as a minimum. Some polymer piping is also used in liquid cooling loops because it doesn't corrode and is lightweight, but polymers vary widely, so check temperature ratings, coolant compatibility and long-term leaching data. Whichever you choose, plan it together with the wetted metals in the cold plates and manifolds.

Start with Water Temperature, Then Choose the Hardware

Order matters in liquid cooling data center design. First establish how much heat the liquid has to carry. Then choose a W class and heat rejection plant, work through the temperature stack to set TCS supply temperature, and only then select CDUs, distribution and materials. Do it in the wrong order and you end up with one GPU running hot at power-on, or pumps at full speed that still can't hit design flow.

Planning a liquid cooling project? Send APALTEK's engineering team your rack heat load schedule, facility supply water temperature (or target W class), project location and redundancy requirements. We'll check your temperature stack and recommend a cold plate, manifold and CDU configuration to match.