Eight years. That's how long I've been handling HVAC and process-cooling orders for industrial and data center facilities. It's also how long I've been keeping a list of my own mistakes—seven documented errors, totaling roughly $48,000 in wasted budget across two employers. I maintain our team's equipment selection checklist now, because I'd rather share my failures than watch someone repeat them.
The biggest lesson came from a comparison I initially got wrong: air-cooled vs. liquid-cooled data center design. Let me walk you through it the way I wish someone had walked me through it back in 2021—dimension by dimension.
What We're Actually Comparing
Air cooling moves heat the way a giant fan does: it pushes chilled air past hot servers and dumps heat into a room or outside. Liquid cooling captures that same heat in a water or coolant loop running through heat exchangers—often plate-and-frame units from Alfa Laval—and removes it far more efficiently.
Four dimensions matter here:
- How each system moves heat, and where heat pumps fit in
- Noise and working environment
- Maintenance and process reliability
- Total cost of ownership rather than sticker price
The last one is where I got burned. More on that later.
Dimension 1: How They Move Heat—and How a Heat Pump Works
Start with physics. Water's specific heat capacity is 4.18 kJ/(kg·K). Air's is about 1.005. That's already a 4:1 advantage, and water is roughly 800 times denser than air at sea level. Put it together, and a pipe the size of your forearm moves as much heat as an air duct you could almost park a car in. That's why air-cooled halls need enormous fan systems.
If you've ever used a Milwaukee leaf blower, you've felt the inefficiency firsthand. Moving heat with air means moving a colossal volume of it, continuously. A data center's air handlers are essentially industrial leaf blowers with cooling coils bolted on, running 24/7. Milwaukee's battery models are relatively quiet as leaf blowers go—mid-60s dB(A)—but imagine that sound multiplied by twenty, in an enclosed room, without ever stopping. That's the air-cooled operator experience.
The liquid-cooled alternative uses pumps and heat exchangers instead of giant fans. The coolant absorbs heat from the servers, carries it to a chiller or dry cooler, and returns cooler. The energy required to move that heat is dramatically lower.
Now the question I always get asked: how does a heat pump work? A heat pump moves heat instead of making it. Refrigerant absorbs heat as it evaporates, a compressor raises its pressure (raising its temperature), the hot gas then releases that heat through a condenser, and an expansion valve drops the pressure to repeat the cycle. For every kilowatt of electricity a heat pump uses, it can move three to four kilowatts of heat—a coefficient of performance (COP) of 3 to 4, per DOE and AHRI data.
Why does that matter in a data center? Because servers reject a lot of low-grade heat, around 30–35°C, which is otherwise just wasted. A heat pump can upgrade that heat to feed building hot water or hydronic heating loops. Alfa Laval data center cooling systems are often designed with this recovery path in mind. The 'waste' heat stops being a utility bill and starts being a resource.
ASHRAE TC 9.9's 2021 guidelines push allowed server inlet temperatures to 27°C (80.6°F), which makes warmer, more efficient chilled-water loops possible. Air cooling can operate in that range too, but liquid cooling does it with far less energy.
Verdict: Liquid cooling wins this dimension decisively. Uptime Institute's 2023 global data center survey puts the average PUE at 1.58; a well-designed liquid-cooled facility can run around 1.1. That gap is not a minor efficiency gain.
Dimension 2: Noise and the Working Environment
I keep coming back to the leaf blower analogy because it's accurate. CRAC units in air-cooled rooms typically run at 75–85 dB(A). OSHA's 8-hour action level for noise is 85 dB(A)—so the equipment itself can push your staff to the edge of mandatory hearing protection.
Liquid-cooled systems are quieter because they're not forcing massive air volumes through the space. The dominant sound is the pump, more of a low hum than a fan whine. The first time I visited a liquid-cooled facility, I waited for the familiar roar that never came. I could hold a conversation at normal volume. It felt wrong, honestly, for about ten minutes. Now it just feels like the design working as intended.
Verdict: Liquid cooling wins on noise, and that has real consequences for operator comfort and site compliance in urban settings.
Dimension 3: Maintenance, Water Chemistry, and My $14,000 Mistake
Now the flip side: liquid cooling demands water discipline. Plate heat exchangers foul if your water chemistry drifts. Strainers clog. Pumps lose efficiency. None of this means the technology is fragile—it means it asks for respect.
I learned this the hard way on the process side. We had a chemical processing line running on an Alfa Laval LKH-PF 50 Hz centrifugal pump—a pump designed specifically for chemical processing and hygienic transfer. I approved a third-party replacement impeller because it was 20% cheaper and looked 'practically the same.' It wasn't.
Three months later, the impeller failed and shed metal fragments into the product stream. The pump itself needed $3,200 in parts and labor. That was the small number. The big number was the batch we had to discard—$14,000 of product—plus a 3-day production delay while we waited for the genuine alfa-laval part. My $200 savings ended up costing roughly $17,000. In my experience managing HVAC and process projects for 8 years, the lowest quote has been the most expensive about 60% of the time. This was the case that made me believe that statistic.
Now I enforce a strict policy: for rotating and heat-transfer equipment in critical service, OEM or documented-equivalent parts only. If the supplier can't provide a material certificate and a performance curve, we don't buy. Simple checklist line item. Nearly fifteen grand to learn it.
Verdict: Maintenance is a draw. Air-cooled systems have belts, filters, and compressors; liquid-cooled systems have pumps, water treatment, and fouling. Both will bite you if you neglect them. But liquid cooling's failure modes are messier.
Dimension 4: Total Cost of Ownership (Where I Got It Wrong)
In 2021, I was asked to help choose a cooling approach for a 250 kW server hall. Air cooling had a lower first cost—roughly $150,000 lower, if I remember correctly; don't quote me on the exact estimate.
I went back and forth between the two options for two weeks. The air-cooled design offered simplicity and a smaller first-year budget. The liquid-cooled design offered lower utility bills and a quieter facility. Ultimately I chose air cooling because 'safe and familiar' felt like the lower-risk answer.
It wasn't.
The energy model said air cooling would use about 1.1 GWh more per year than liquid cooling—roughly $80,000 to $110,000 annually in extra operating costs, depending on tariff. By month 24, the air-cooled system had surrendered its $150,000 capex advantage. By year five, the net cost difference exceeded $250,000. The choice that looked cheap on the purchase order was the expensive one on the income statement.
The calculation I should have done at the start: the upside of air cooling was $150,000 in first-year savings. The downside was a decade of six-figure energy penalties and a permanently noisy plant. I kept asking myself whether $150,000 was worth potentially doubling the facility's lifetime energy spend. The answer, in hindsight, is obviously no—but I asked it in the wrong order, after the purchase order was signed.
When I later stood in a liquid-cooled facility and compared it, side by side in my memory, with our air-cooled hall, the lesson crystallized: energy models don't lie, but our biases do.
The cheapest option is only the cheapest if you count everything. Purchase price is a line item. Total cost of ownership is the whole story.
Verdict: Liquid cooling wins on total cost for any facility expected to run at meaningful utilization for more than three years. The exception is cheap electricity, low utilization, or a building you plan to gut anyway.
What I'd Recommend
If you're designing today, here's my rough rule of thumb.
Air cooling makes sense when: the load is under 50 kW, electricity is unusually cheap, you're renovating within five years, or local codes restrict water use. Simplicity is a legitimate feature, and air cooling has it.
Liquid cooling makes sense when: the load is above 100 kW, electricity is at typical commercial rates, noise matters (urban sites, adjacent offices), or you want waste-heat recovery. That's where the alfa-laval data center cooling product line, with its plate heat exchangers and integrated pump packages, builds a strong business case.
And a note for the home side of things: if you've been searching for 'hot water heater replacement near me'—the heat pump water heater in a modern home runs on the exact same principle I described above. It moves heat from your basement into your water tank, at three times the efficiency of electric resistance. Industrial data center heat recovery is the same cycle at industrial scale. The technology you trust to cool a 2 MW server hall is the same technology warming your shower.
My current checklist, for teams that ask:
- Water chemistry plan before commissioning, not after.
- OEM or certified-equivalent parts for all rotating equipment.
- Quarterly pressure-drop logs on every plate heat exchanger, to catch fouling early.
- Metered power compared against the energy model monthly for the first six months.
- Noise measurements at the site boundary before and after startup.
I'm not here to tell you liquid cooling is always right. It isn't. But I've watched too many engineers default to air cooling because it's familiar, and too many procurement teams chase the lowest quote without checking the lifetime math. Both mistakes cost real money.
Take it from someone who kept the receipts.
