Waterless Data Center Cooling: Why Dry Coolers Beat Compressor Only Designs
Going waterless is a real win. Going waterless with only a compressor cooling your racks is a different decision entirely and in most U.S. climates it is the expensive one.
We have been pitched high-lift chillers hard this year. The narrative is clean and it is attractive: eliminate the evaporative cooling tower, eliminate the water footprint, reject heat straight into the atmosphere, and do it on a 110 °F afternoon without blinking.
The physics behind the pitch is sound and the engineering is impressive, but when our team sat down to actually design around it, we found that the conversation in the market has quietly skipped a step. Almost every discussion we have had compares high-lift chillers to evaporative cooling towers. Almost none of them compares a high-lift chiller to the alternative that beats it on both water and power.
So here is the walk-through, in the order our own engineers had to resolve it.
How High Lift Chillers Work in Waterless Data Center Cooling
First: does “high-lift” mean you get rid of the compressor? No. It means the opposite.
When a manufacturer offers a high-lift waterless machine rated for a 110 °F design day, what they are selling is a refrigeration system built to survive an unusually large pressure ratio typically on heavy-duty, multi-stage centrifugal compressors. The compressor is not removed. It is reinforced.
The reason is straightforward. Heat only moves from hot to cold. If the air outside is 110 °F, a plain closed-loop radiator has nothing to give: your 105 °F return water is colder than the air you are trying to dump it into. A high-lift machine gets around this by compressing the refrigerant to a much higher pressure, driving condensing temperature up to roughly 110–132 °F depending on ambient. Now the coil is meaningfully hotter than the outdoor air, the temperature difference across it is large, and heat flows again as pure sensible transfer, with no evaporative water consumption at all.
That part works: water use goes to zero, but the question is what it costs you the rest of the year.
Why Data Center Dry Coolers Still Matter in Waterless Cooling Systems
Second: if a high-lift chiller can do it alone, are dry coolers redundant? This is the question that actually matters, and the answer is no. . . emphatically.
A standalone high-lift chiller has exactly one path from your racks to the outdoors, and the compressor is standing in the middle of it. There is no bypass. There is no second route. Which means the compressor runs 8,760 hours a year, whether or not the weather requires it.
Now picture a 48 °F night in February. Your return water is 105 °F. The air outside is nearly 60 degrees colder than the fluid you are trying to cool. Every watt of compressor power you spend in that moment is buying you something the weather was prepared to give away.
Add a dry cooler bank and a changeover valve and that night costs you fan power and pump power, nothing else. The compressor sits idle. This is not an exotic control scheme; it is the same economizer logic the commercial HVAC industry has used for forty years. What is new is how much of the year it now covers, and that is entirely because of the loop temperature.
Figure 1 — The same heat, two very different routes out of the building. The combo architecture adds a path the standalone design does not have.
How Warm Water Loops Unlock Data Center Free Cooling
Third: why does the 95/105 °F loop change the answer so completely? Because free waterless cooling is a temperature race, and a warm loop starts the race almost at the finish line.
A dry cooler can produce leaving water at roughly the outdoor dry bulb plus its approach. At an 8 °F approach, a dry cooler makes 95 °F water any time the ambient is at or below 87 °F. That is the entire changeover criterion, while everything else follows from it.
In Nashville, the ambient is at or below 87 °F for about 8,300 hours a year. Ninety-five percent of the time, a properly sized dry cooler bank can carry the full IT load by itself, and the chiller has nothing to do.
Figure 2 — The same building on a legacy 45/55 °F loop gets free cooling 23 % of the year. On a 95/105 °F CDU loop it gets 95 %. The loop temperature is doing the heavy lifting, not the chiller.
The industry keeps crediting the chiller for what the loop temperature is actually doing. On a 95/105 °F CDU loop, the chiller is a trim device 95 % of the year, if you give it a way to get out of the path.
It also cuts the other way. On the 460 hours a year when the chiller does run, it is running at a very small lift, which is exactly where a magnetic-bearing centrifugal is at its best. The warm loop makes free cooling viable and makes mechanical cooling cheap. A design that only captures the second half of that is leaving most of the benefit on the table.
Choosing the Best Chiller for High Temperature Data Center Cooling
Which compressor belongs on a loop like this?
Lift is the enemy of compressor efficiency, and a 95/105 °F loop has very little of it. That changes which machine is the right answer. Here is how the three candidates compare at these parameters.
Standalone high-lift centrifugal | Dry coolers + magnetic-bearing centrifugal | Air-cooled scroll bank | |
Power at 95 °F ambient | 0.72 kW/ton | 0.55 kW/ton | 0.89 kW/ton |
Power at 110 °F ambient | 1.06 kW/ton | 0.89 kW/ton | 1.32 kW/ton |
Power at 65 °F ambient | 0.46 kW/ton | 0.05 kW/ton (free cooling) | 0.69 kW/ton |
Blended annual, Nashville | 0.48 kW/ton | 0.075 kW/ton | 0.65 kW/ton |
Blended annual system COP | 7.3 | 47 | 5.4 |
Compression | Multi-stage centrifugal, dynamic | High-speed direct-drive centrifugal, dynamic | Orbiting scroll, positive displacement |
Lubrication | Oil-lubricated, hydrodynamic bearings | Oil-free, magnetic levitation | Oil-lubricated, sumps and separators |
Surge at low lift | Low risk; heavy aerodynamic stability | Managed — requires VFD, IGV and hot-gas bypass | None — inherent to positive displacement |
Hours the compressor runs | 8,760 | 460 | 8,760 |
Table 1 — All power figures are whole-system heat rejection (compressor + fans + heat-rejection pumps) at a 95 °F supply / 105 °F return facility loop, derived from the bin model in the appendix. They are not catalog full-load ratings and not AHRI IPLV, which is measured at chilled-water conditions that do not exist in this application.
Comparing Chiller Technologies for Data Center Cooling: The three machines, briefly
Scroll compressors
Scroll compressors trap and compress refrigerant between interlocking orbiting spirals. They cannot surge, which makes them forgiving, but mechanical friction and coarse capacity steps cap their efficiency well below what this loop could deliver.
High-lift multi-stage centrifugals
High-lift multi-stage centrifugals are built for heavy pressure ratios and will hold a facility online through weather that shuts other machines down. That resilience is real and worth paying for, as insurance. It is not an efficiency strategy.
Magnetic-bearing centrifugals
Magnetic-bearing centrifugals levitate the rotating shaft in an electromagnetic field. No metal contact, no friction, no oil, no oil management loop at all. Danfoss builds the compressors under the Turbocor brand; the chillers themselves come from Smardt, Multistack, Daikin and others. At the small lift this loop presents, these machines post efficiencies that are difficult to believe until you see the selection.
How Magnetic Bearing Chillers Prevent Surge at Low Lift
If lift is that small, what keeps a centrifugal from surging? A fair objection, and one worth answering directly, because it is the legitimate engineering risk in this design.
A centrifugal is a dynamic machine, not a positive-displacement one. It works by accelerating gas, not by trapping it. When the pressure it is asked to develop falls too far relative to the gas it is moving, flow detaches from the impeller blades and momentarily reverses -- surge.
On a data center loop the trigger is unusual: not low load, since the IT load is constant, but low lift. On a 45 °F winter night the condenser pressure collapses, the pressure split across the wheel becomes very narrow, and an impeller still spinning fast enough for design conditions is now generating far more flow than that narrow split can support.
Modern magnetic-bearing platforms handle this with three layers working together:
Variable-speed control.
As lift shrinks, the drive slows the impeller so it stops generating pressure the system no longer needs. Magnetic bearing sensors track shaft position thousands of times per second and correct axial displacement before gas reversal can develop.
Inlet guide vanes
At low lift the guide vanes pre-swirl the gas before it reaches the impeller, changing its entry angle so the stage stays aerodynamically stable at reduced mass flow.
Hot-gas bypass
If the pressure ratio still compresses past the stable envelope, an anti-surge valve routes a small fraction of discharge gas back to suction. It maintains minimum mass flow across the wheel and keeps the machine online instead of tripping it.
Worth noting: the standalone high-lift machine faces the same winter problem, and its answer is worse. With no bypass path and a compressor that must run regardless, it typically holds an artificially elevated minimum condensing temperature all winter simply to stay out of surge, staging condenser fans off to keep the head pressure up. It spends energy manufacturing lift it does not need, in order to protect a compressor that should not have been running in the first place.
Data Center Cooling Efficiency Analysis for a 1,000 Ton Facility
Running the numbers: 1,000 tons in Nashville
Nashville is a useful test case: a genuinely hot, humid mid-summer, a real winter, and an emerging data center market. We modeled 1,000 tons of continuous IT load on a 95/105 °F facility loop, with dry coolers selected for an 8 °F approach.
Data Center Cooling Performance at Peak Design Conditions: The design day
At Nashville’s 0.4 % design condition of about 94.6 °F dry bulb, the dry coolers produce roughly 103 °F condenser water. Against a 95 °F leaving evaporator temperature, that is an 8 °F water-to-water split — about 18 °F of actual refrigerant lift once evaporator and condenser approaches are counted. A magnetic-bearing machine at 18 °F lift runs near COP 8.5:
Chiller = 3.517 ÷ 8.5 ≈ 0.41 kW/ton
+ dry cooler fans 0.095 + heat-rejection pumps 0.040 = 0.55 kW/ton
The standalone high-lift machine at the same hour draws about 0.72 kW/ton. On the single worst day of the year, the combo is roughly 24 % better. That alone would not be worth writing about.
How Data Center Free Cooling Cuts Annual Energy Use: The other 8,300 hours
This is where the decision is actually made.
Figure 3 — Nashville’s year, sorted by temperature. Only 460 hours sit above the 87 °F changeover. The design-day argument governs 5 % of the calendar.
Below 87 °F, the compressor is off and heat rejection costs fan power and pump power only. And fan power falls away faster than most engineers expect, because fan power scales with the cube of speed. At 65 °F ambient the temperature difference driving the coil has tripled, the fans need roughly a third of design airflow, and the fan law puts that at about five percent of design power. Free cooling in the shoulder seasons is not merely cheap — it is very nearly free.
Figure 4 — The gap on the design day is modest. The gap across the rest of the year is not.
Annual Energy Savings From Dry Cooler Data Center Cooling: The annual total
Outdoor bin | Hours | Standalone kW/ton | Combo kW/ton | Standalone kWh | Combo kWh |
Above 90 °F | 175 | 0.72 | 0.55 | 125,400 | 95,600 |
87 – 90 °F | 285 | 0.57 | 0.43 | 161,900 | 122,100 |
80 – 87 °F | 1,000 | 0.54 | 0.08 | 539,600 | 81,600 |
70 – 80 °F | 1,400 | 0.51 | 0.05 | 710,900 | 72,700 |
60 – 70 °F | 1,200 | 0.46 | 0.05 | 557,400 | 57,600 |
50 – 60 °F | 1,300 | 0.45 | 0.05 | 586,700 | 62,400 |
40 – 50 °F | 1,400 | 0.45 | 0.05 | 626,700 | 67,200 |
Below 40 °F | 2,000 | 0.45 | 0.05 | 895,200 | 96,000 |
Total | 8,760 | 0.48 | 0.075 | 4,203,823 | 655,163 |
Table 2 — Bin analysis, 1,000 tons continuous. Hour distribution is an approximate typical meteorological year for Nashville; confirm against site TMY data before using these figures in a design submittal.
Figure 5 — Same load, same climate, same water consumption — zero. The difference is whether the compressor has a way to step aside.
The combo architecture avoids roughly 3.5 million kWh per year. At an industrial rate of $0.08/kWh that is about $284,000 per year of avoided operating cost — an 84 % reduction in heat-rejection energy, on a facility that consumes no cooling water either way. Over a ten-year hold, at that rate, it is roughly $2.8 million.
And note where the combo’s remaining energy goes: at 655,163 kWh/yr, more than half of it is heat-rejection pumping, which runs continuously regardless of mode. The compressor accounts for less than a quarter. That is a fundamentally different machine to operate.
Data Center Cooling Costs: Dry Coolers Versus High Lift Chillers
What about first cost? Here the honest answer is: it depends, and it depends on one number.
The standalone approach forces you to buy a premium industrial chiller plant sized for 100 % of peak load at design ambient, every ton of it, rated for the worst hour of the year. In the combo architecture the chiller is a trim device. It bridges the gap only when the dry coolers can no longer close it on their own, so it can be selected substantially smaller.
Against that, a dry cooler bank is not a cheap piece of equipment, and the tighter you specify its approach, the larger and more expensive it gets. That single selection sets your changeover temperature, which sets your free-cooling hours, which sets the entire economic case:
Dry cooler approach | Changeover ambient | Free-cooling hours | Blended kW/ton | Annual savings @ $0.08 |
6 °F | 89 °F | 8,585 (98 %) | 0.061 | $293,000 |
8 °F | 87 °F | 8,300 (95 %) | 0.075 | $284,000 |
10 °F | 85 °F | 8,300 (95 %) | 0.080 | $280,000 |
12 °F | 83 °F | 7,300 (83 %) | 0.121 | $252,000 |
15 °F | 80 °F | 7,300 (83 %) | 0.130 | $245,000 |
20 °F | 75 °F | 7,300 (83 %) | 0.158 | $225,000 |
Table 3 — Dry cooler approach temperature is the single highest-leverage decision in this architecture. Hour counts step rather than curve because the underlying bins are coarse; the trend is what matters.
The useful observation is how flat the right-hand column is. Even a loose 20 °F approach (a much smaller, cheaper coil) still captures 83 % of the year in free cooling and still saves roughly $225,000 annually. The architecture wins across the whole range. Tightening the approach is an optimization, not a prerequisite.
The Best Waterless Cooling Architecture for Data Centers
The verdict: High-lift chillers are a real achievement and they solve a real problem.
If you are building in a climate where the ambient genuinely sits above your changeover temperature for a meaningful share of the year (Phoenix, Las Vegas, parts of the Gulf) you need that high-ambient capability, and you should buy it.
But buying it as your only path to the outdoors is the mistake. The pitch is framed as high-lift chiller versus cooling tower, and against that comparison it wins on water. Framed correctly, high-lift chiller versus dry coolers with a trim chiller, it consumes the same water, which is none, and roughly six times the energy.
Specify the high-lift capability for the 460 hours that need it. Then give the compressor a way to step out of the path for the 8,300 hours that do not.
Pair a downsized magnetic-bearing chiller bank with a properly selected dry cooler network, and you protect the water table and the power bill at the same time. That is what a waterless data center should actually look like.
Data Center Cooling Design Assumptions and Engineering Basis
Appendix: Design basis and assumptions
Every figure in this article is derived from the following basis. We publish it so the numbers can be checked, and so they can be re-run against a different climate or a different selection.
Parameter | Value | Note |
IT load | 1,000 tons continuous | 8,760 h/yr, no diversity |
Facility water loop | 95 °F supply / 105 °F return | CDU primary |
Dry cooler approach | 8 °F | Sets 87 °F changeover |
Dry cooler fan power at design | 0.085 kW/ton | Fan law: power ∝ speed³ |
Heat-rejection loop pumps | 0.040 kW/ton | Continuous, combo only |
Mag-bearing chiller efficiency | 28 % of Carnot | COP capped at 12 at minimum lift |
High-lift chiller efficiency | 28 % of Carnot | COP capped at 8 at minimum lift |
Air-cooled condenser approach | 22 °F | Standalone architecture |
Minimum condensing, standalone | 100 °F | Held for surge avoidance |
Evaporating refrigerant temp | 90 °F | 5 °F below leaving chilled water |
Nashville 0.4 % design DB | ≈ 94.6 °F | 1 % value is ≈ 92 °F |
Electricity rate | $0.08/kWh | Industrial, flat |
Table 4 — Design basis. The CDU-side distribution pumps are common to both architectures and are excluded from both. The heat-rejection loop pumps are charged to the combo architecture only, which is the conservative direction for the comparison presented here.
Data Center Cooling Analysis Limitations
What this analysis does not do
– It does not use site-specific TMY data. The Nashville hour distribution is an approximate typical year, adequate for architecture selection and not adequate for a design submittal.
– It does not model part-load IT loads, redundancy (N+1 or 2N), or the energy penalty of running redundant equipment at reduced load.
– It does not price equipment. The CapEx discussion is directional; the dry cooler bank is a significant capital item and net first cost depends heavily on the approach temperature selected.
– It does not address glycol concentration, freeze protection, or the capacity derate that comes with both — all of which matter in a real dry cooler selection in a climate with a winter.
Maktinta Energy designs and builds mechanical and cooling systems for data centers, commercial and industrial facilities. If you are weighing heat-rejection architecture on a project, we would be happy to run this analysis against your climate and your loop temperatures. Maktinta@gmail.com · Tel: (408) 432-9900 · www.maktinta.com
Comments