A dry cooler is a closed-loop heat exchanger that rejects heat to outdoor air by pushing air across finned coils with fans, without evaporating water. In a data center, it sits on the facility side of the cooling system, pulling heat out of the water or water-glycol loop after that loop has already picked up heat from the IT equipment. Liquid cooling changes how heat gets into that loop. It does not remove the need for the loop itself, or for something outdoors to reject the heat it carries.
This guide covers how a data center dry cooler fits into that cooling path, why large units are built with multiple fans instead of one, and what that means when you are specifying fans for a dry cooler project. It does not repeat the full adiabatic cooler versus dry cooler versus cooling tower comparison or the EC versus VFD control breakdown. Both are covered in dedicated guides linked below.
Key Takeaway
Liquid cooling moves heat off the chip. It does not remove the need for facility-level heat rejection, and a dry cooler is one of the most common ways that heat still reaches outdoor air. Large data center dry coolers almost always run multiple fans instead of one big one, for reasons that go well beyond “the manufacturer had extra fans lying around.”
Where the Heat Actually Goes: From Chip to Outdoor Air
A common misconception is that liquid cooling removes the need for air-side heat rejection. It doesn’t work that way. Liquid cooling is good at pulling heat off a chip efficiently, using a cold plate in direct contact with the processor, but that heat has to leave the building somewhere, and it almost always leaves through air moving across a coil.
The path usually runs in stages. A cold plate picks up heat at the chip. A coolant distribution unit, or CDU, moves that heat from the server-level loop into the facility water loop. The facility loop then carries that heat to the site’s heat-rejection system, which might be a dry cooler, a cooling tower, or a chiller. A dry cooler or cooling tower can reject heat straight to outdoor air, but a chiller works differently: it uses a refrigeration cycle to move heat from its evaporator side to its condenser side, and that condenser side still needs its own heat-rejection equipment, such as an air-cooled condenser, cooling tower, or dry cooler, to send the heat outdoors. The final step usually rejects the heat to ambient air, though some designs recover it for another use instead.
So the honest way to describe it: liquid cooling does not eliminate facility-level heat rejection. It just moves the first leg of the job, chip to loop, off of air and onto liquid. Everything downstream of that loop, including the fans on a dry cooler, still has real work to do. This isn’t a legacy holdover, either: current ASHRAE guidance for AI data centers documents dry cooler-based heat rejection as a standard architecture for high-temperature liquid-cooled facilities. If you’re evaluating a liquid-cooled build, that’s the part of the system this guide focuses on.
What Is a Dry Cooler in a Data Center Cooling System?
A dry cooler is a coil-and-fan heat exchanger that rejects heat straight to outdoor air, with no evaporative water use at the coil. Air is pulled or pushed across a finned coil carrying the facility loop’s hot water or water-glycol mixture, and the coil gives up heat to the cooler outside air. Because there is no evaporative spray or wetted media involved, a dry cooler behaves more like a large outdoor radiator than a cooling tower and avoids the water consumption tied to evaporative heat rejection. In freezing climates, though, the fluid loop still needs its own freeze protection, such as a correctly specified water-glycol mixture, drain-down, or heat trace. Fans built for this kind of outdoor duty are usually rated well below freezing too. Proventofan’s large-frame EC and AC external rotor axial fans, for example, are rated down to -25°C and -30°C respectively, but that fan rating does not by itself establish freeze protection for the dry cooler’s fluid circuit.
That’s the short version. A cooling tower rejects heat by evaporating water, which lets it reach lower approach temperatures but adds water consumption and water treatment to the picture. A chiller adds a refrigeration cycle on top of any of this, trading more energy use for tighter, more stable output temperatures. If you need the full breakdown of water use, freeze risk, and typical use cases across all three, we’ve already written one in our adiabatic cooling guide, which covers dry coolers, cooling towers, and the adiabatic middle ground between them. If your site already runs a cooling tower and you’re weighing an EC retrofit there instead, see our EC fans for cooling towers guide.
For this guide, the part that matters is simpler: a dry cooler earns its keep with airflow, not water. And once you scale a dry cooler up to data center size, that airflow has to come from somewhere, which is where fan count starts to matter.
Why Large Data Center Dry Coolers Use Multiple Fans, Not One Big One
Walk up to a data center dry cooler and the first thing you’ll usually notice is a row of fans, not one giant one. Honestly, the first time someone sees six or eight fans lined up on top of a single unit, it can look like overkill. It isnt. Each fan is doing a specific job, and none of the six reasons below hinge on a single point of failure.
Redundancy. A multi-fan array can retain partial airflow after one fan is unavailable, provided the cooler’s controls and system topology allow the remaining fans to keep operating. That’s a meaningfully different failure mode than a single large fan, where losing the one fan means losing the unit’s only active airflow source.
Part-load control flexibility. Multiple fans can be staged on and off, or ramped at different speeds, to track the actual heat load instead of running at one fixed output. That gives an operator more ways to match airflow to load than a single large fan offers, though whether that translates into lower energy use in a given installation depends on the staging strategy and the actual operating profile, not just the fan count.
Coil-face airflow coverage. A large dry cooler coil has a lot of surface area to move air across, and several fans placed across that face give you multiple independent airflow points instead of one. That helps distribute airflow across the coil, but how evenly it distributes still depends on fan spacing, plenum design, and coil geometry. More fans isnt an automatic guarantee of even coverage.
A quantified consequence when a fan is lost. Because the array is made of identical, known-capacity units, you can put a number on what one failed fan costs you instead of just saying performance drops. We’ll work through that math below.
Serviceability. Smaller fan cells are generally easier to handle and replace than one large fan, and some multi-fan cooler designs can continue operating at reduced capacity while one fan cell is isolated for service. Whether a given unit supports that kind of in-service maintenance depends on its isolation, guarding, and structural design, so it’s worth confirming with the cooler manufacturer rather than assuming it as a default. The same logic shows up in a closely related equipment category: NIH’s facility engineering guidance on fan arrays in air handling units notes that multiple small fans reduce the need to shut down the entire fan system for service, and that fan array sizing sees diminishing returns in cost, redundancy, and maintenance beyond a certain fan count.
Acoustic control flexibility. Several mid-sized fans running at staged or reduced speeds give you more levers to manage sound pressure at a given location than one large fan running flat out. That said, multiple fan sources add their sound power together, so a multi-fan array isn’t automatically quieter than a single large fan. It’s a different set of controls, not a guaranteed win.
Thats the engineering case for the array. It’s also a real-world constraint. There’s a practical ceiling on how big one fan gets. Within Proventofan’s current catalog, the highest single-fan free-air rating in this comparison is about 25,000 m³/h, from an 800mm eight-pole AC external rotor unit, with the 900mm six-pole version close behind at about 22,000 m³/h, both rated for a -30°C to +60°C ambient range typical of outdoor installations. If a dry cooler’s required duty point, meaning airflow at the actual static pressure the coil presents, exceeds what one available fan can deliver, the designer has to use multiple fan cells or move to a different fan platform. That decision has to be made at the required airflow and static-pressure operating point, not from a free-air maximum alone.
The same logic plays out on the EC axial side. A 630mm EC axial fan is rated at up to 14,500 m³/h, with 240 Pa of static pressure, 69 dB(A) noise, 800 W input, and 1,150 RPM, at IP54 protection. Selected models from 250mm upward also include an RS-485 control interface. That’s worth flagging honestly: having an RS-485 interface confirms the fan can be wired into a communication bus, but it doesn’t by itself confirm the protocol, addressing scheme, or read/write functions available on a given motor. Confirm the protocol, addressing, and read/write functions for the selected motor before designing a networked fan array around it. See the full 630mm EC axial fan specifications and 500mm EC axial fan specifications for the complete duty range.
| Fan | Platform | Max free-air rating | Noise | Ambient range | Control interface |
|---|---|---|---|---|---|
| 630mm | EC axial | 14,500 m³/h at 240 Pa | 69 dB(A) | -25°C to +60°C | RS-485 interface, protocol to confirm |
| 800mm, 8-pole | AC external rotor | 25,000 m³/h | 80 dB(A) | -30°C to +60°C | Grouped or individually wired externally |
| 900mm, 6-pole | AC external rotor | 22,000 m³/h | 77 dB(A) | -30°C to +60°C | Grouped or individually wired externally |
Data source: Proventofan catalog specifications, factory test conditions. Airflow figures are free-air ratings, not installed performance at your actual static pressure.
Here’s where the “quantified consequence” point becomes concrete. In a simplified nameplate example, picture a data center dry cooler built around six identical 630mm EC axial fans, each rated at 14,500 m³/h free air, for a nameplate total near 87,000 m³/h. If one fan is taken out of service, five of six identical fans leave about 83% of the original free-air rating available. Actual installed airflow won’t necessarily fall in exactly that proportion, because the operating point moves with the system resistance curve, not just the fan count. If the remaining fans were operating below their rated maximum speed, the controller may be able to increase their speed within the approved operating range to partially offset the loss.
By contrast, if that same cooler relied on one fan only, failure of that fan would remove 100% of the fan-driven airflow from that cooler unit. This is a representative design example built from real catalog numbers, not a documented field project, meant to illustrate the math rather than promise a specific recovery percentage on your installation. The number worth taking away for your own project: run this same fraction, fans remaining divided by total fans, against your actual fan count and duty point before you commit to an array size, since that ratio is what tells you how much redundancy margin you are really buying, not a guess.
Redundancy at the equipment level is also a familiar idea one level up, in facility design. The Uptime Institute Tier Classification System rates data centers on redundant components and concurrently maintainable infrastructure, which is the same underlying idea as an N+1 fan array, even though Uptime’s own framework doesn’t use that specific notation. Whether a specific dry-cooler fan array must provide N+1 capacity, in the sense HVAC engineers commonly use the term, depends on the facility’s topology and cooling-system design, not on the Tier rating alone. Treat the Tier framework as useful background for why redundancy matters at the facility level, not as a rulebook for how many spare fans a specific dry cooler needs.
When EC Axial Fans Add Value in Multi-Fan Arrays
A fan array’s control architecture becomes more important as fan count grows, which is where the choice between EC and AC external rotor motors can start to matter. Proventofan’s standard AC external rotor product line doesn’t include the built-in RS-485 interface that EC models from 250mm upward carry. External control architecture can still group or individually wire AC external rotor fans, but that’s a different, more hands-on approach than a fan that ships with a communication interface already built into the motor housing.
Important Note
An RS-485 interface confirms the fan can be wired into a communication bus. It does not by itself confirm the protocol, addressing scheme, or read/write functions available on a given motor. Verify those specifics against the motor’s own protocol documentation before you design a networked array around them.
The potential value grows as fan count increases, but only if the selected motor’s communication protocol actually supports the functions your control architecture needs. If addressable control, speed feedback, or fault-status readback matters to the project, verify those functions against the motor’s own protocol documentation before specifying the array. This isn’t a hypothetical concern: on a different large-frame industrial fan project, a customer’s engineering team asked us this exact list of questions before committing to a control architecture, including whether the PLC could write speed directly, whether maximum RPM could be limited in firmware, and what happened on communication loss. That was a different piece of equipment than a dry cooler, but the underlying discipline carries over directly: get the control behavior confirmed engineer to engineer before the array gets specified, not after. On a smaller installation with two or three fans, that communication layer may not justify the added integration complexity, and a straightforward AC external rotor fan can be the simpler, lower-cost answer.
So the honest framing is this: AC external rotor fans arent the wrong choice for a data center dry cooler. They’re a proven, widely used option, and Proventofan’s line covers sizes up to 900mm for exactly this kind of large outdoor duty. EC’s RS-485 interface is a potential edge for larger arrays with more demanding control or communication requirements, provided the protocol and addressing support what you need, which is a question for the motor’s own documentation, not a blanket promise. For the deeper mechanics of how EC control differs from an AC motor paired with a VFD, see our EC fan vs VFD comparison.
What to Check Before Specifying Fans for a Data Center Dry Cooler
Once you understand the array logic and where EC adds a real edge, the next step is turning that into a specification. A handful of assumptions trip up dry cooler fan specs more often than the rest, so here’s the list as a check against them:
| Do | Don’t |
|---|---|
| Specify airflow at the coil’s actual required static pressure. | Rely on a fan’s free-air maximum as if it were installed performance. |
| Confirm fan arrangement, spacing, and the induced-draft or forced-draft layout against the coil design. | Assume any fan diameter will fit without checking the mounting envelope. |
| Verify RS-485 protocol, addressing, and read/write functions directly with the motor documentation if networked monitoring matters to the project. | Assume an RS-485 interface by itself means independent status readback. |
| Get target noise levels in writing before selecting fan count and speed. | Wait until after installation to find out the site has a noise limit. |
| Confirm the site’s ambient temperature range and the fluid loop’s own freeze-protection plan. | Assume a fan’s low-temperature rating covers freeze protection for the dry cooler’s fluid circuit. |
| Treat a fan-quantity estimate (total airflow divided by per-fan airflow at your static pressure) as a starting point. | Finalize fan count before checking it against coil geometry, spacing, and the combined system curve. |
Confirm supply voltage and phase available at the installation site rather than assuming a default, and have a rough fan-quantity estimate ready. Fan count is one of the biggest drivers of total project cost, so even a preliminary number helps a supplier respond faster.
A list like this is what turns a vague airflow requirement into a quote a supplier can actually work from, instead of a back-and-forth email thread.
Frequently Asked Questions
What is a dry cooler in a data center?
A dry cooler in a data center is an outdoor heat exchanger that uses fans to move air across finned coils, removing heat from the facility water loop before that loop recirculates back into the cooling system. It rejects heat without evaporating water, unlike a cooling tower.
Is there a way to cool data centers without water?
Dry coolers can reject heat without evaporative water consumption, although the closed facility loop may still circulate water or a water-glycol mixture. What changes is where the water stays. It stays inside the loop instead of evaporating into the outdoor air.
Why don’t data centers reuse cooling water?
Many data-center cooling loops do reuse the same circulating fluid. Water consumption mainly becomes a separate issue where evaporation, blowdown, or make-up water is involved, which is more common with cooling towers than with dry coolers.
What’s the difference between a dry cooler and a chiller?
A dry cooler rejects heat directly to outdoor air through a coil and fans, with no refrigeration cycle involved. A chiller uses a refrigeration cycle to produce chilled water at a controlled temperature, which costs more energy to run but keeps output temperature stable across a wider range of outdoor conditions than a dry cooler can, within its designed operating envelope.
Do dry coolers use a lot of electricity?
Fan power is the main electricity draw on a dry cooler, since there’s no compressor involved. It scales with the number and size of fans and how hard they’re running, so a well-controlled multi-fan array that only ramps up the fans it needs usually draws less power than running every fan at full speed all the time.
Why do large dry coolers use multiple fans instead of one big one?
Multiple smaller fans can support redundancy, staged part-load control, broader coil-face airflow coverage, and modular serviceability in ways a single-fan layout cannot. Whether the array provides true redundant capacity depends on the cooler’s design duty, control strategy, and available capacity with one fan unavailable. If one fan goes down, the remaining fans can often continue moving air at reduced capacity, and individual fans can sometimes be swapped without shutting down the entire unit.
What happens if one fan fails in a multi-fan array?
The array loses one active fan cell, but the resulting change in installed airflow is not necessarily equal to that fan’s nameplate share. In a simplified example, if one fan out of six identical units fails, the nameplate free-air rating drops by about one sixth, though actual installed airflow depends on the system resistance curve and how much speed headroom the remaining fans have within their approved operating range.
Can dry cooler fans run at independent speeds?
Some Proventofan EC axial fans from 250mm upward include an RS-485 control interface, which is a step toward per-fan control. Whether a specific installation can run independent speeds and read back status from each fan depends on the protocol and addressing support of the selected motor, so confirm this with the manufacturer before designing a networked array.
Do liquid-cooled data centers still need dry coolers?
Liquid-cooled data centers still need a facility-level path for rejecting or reusing that heat, but that path does not have to be a dry cooler. A facility may use a dry cooler or cooling tower, a chiller paired with its own condenser-side heat-rejection equipment, or in some cases a heat-recovery system.
References
- Uptime Institute. Tier Classification System. https://uptimeinstitute.com/tiers
- ASHRAE. Integrated Design Principles, AI Data Center Energy Performance Framework. https://www.ashrae.org/technical-resources/ai-data-center-framework/integrated-design-principles
- National Institutes of Health, Office of Research Facilities. Fan Arrays in Air Handling Systems (Technical News Bulletin, June 2023). https://orf.od.nih.gov/TechnicalResources/Documents/Technical%20Bulletins/23TB/Fan%20Arrays%20-%20June%202023%20Technical%20Bulletin%20_508.pdf
Disclaimer
This article is general technical reference material. Airflow, static pressure, and noise figures cited above come from Proventofan’s own catalog data and reflect factory test conditions, not the performance you will necessarily see once a fan is installed against a specific coil and system resistance curve. RS-485 availability on a given motor does not by itself confirm protocol, addressing, or monitoring support. Confirm these details, along with duty point, voltage, and environmental requirements, directly with Proventofan or your dry cooler manufacturer before finalizing a fan specification.
What to Send Us for a Fan Quote
If you’re specifying fans for a data center dry cooler project, send us:
- Total airflow requirement (m³/h or CFM)
- Available mounting footprint and fan arrangement (induced-draft or forced-draft)
- Working point (airflow and static pressure from the same duty curve)
- Site ambient temperature range and any dust, salt spray, or coastal exposure
- Noise limits, if the site is near a noise-sensitive area
- Supply voltage and phase
- Whether you need an RS-485 control interface
- Estimated fan quantity
Reach out through our contact page with as much of that as you have, and we’ll help you narrow it down. Browse our EC axial fan range for EC models, our AC external rotor axial fan range for conventional AC options, or our EC Fan Solutions hub for the full EC lineup.
