Why Data Center Cooling Isn't What You Think It Is (and Why Your Efficiency Targets Are at Risk)

When I first started working with data center cooling, my entire mental model was completely wrong. I thought that buying a more efficient chiller or simply lowering the supply air temperature setpoint was the main path to lower PUE. Two years and several expensive misdiagnoses later, I realized the real battle isn't about raw cooling capacity—it's about moisture control and load granularity.

My initial approach to selecting an air handling unit for a mid-tier colocation facility was embarrassingly simple: calculate the heat load, pick an AHU with enough tonnage, and ensure redundancy. I assumed the biggest cost driver was the compressor or the fan energy. I was wrong.

The Surface Problem: Energy Bills Are Too High

The facility managers I talk to all have the same complaint. Their PUE is hovering around 1.6 or 1.7, and they're being told by leadership to get it down to 1.3 or better. The immediate assumption is that the cooling system is underperforming. So they look at the chillers, check the condenser coils, and maybe upgrade to a VFD for the fans. But these surface-level fixes rarely deliver the promised savings.

In March 2024, I was called in to consult on a facility that had just spent $400,000 on a 'high-efficiency' chiller upgrade. Their PUE dropped by only 0.05. The engineering team was baffled. The chiller had a better IPLV rating, but the actual power draw hadn't changed much. The problem wasn't the chiller—it was the airside.

The Deep Cause: Humidity Is the Hidden Energy Vampire

Here's the part that everyone forgets. The human body cools itself by sweating. Data centers don't sweat. But the air inside them is constantly fighting a battle against humidity. If the air is too humid, you risk condensation on cold surfaces and corrosion. If it's too dry, you risk catastrophic electrostatic discharge that can take down a server rack.

The deep cause of high PUE isn't poor chiller efficiency—it's the energy wasted on reheat and humidification to maintain a stable dew point. A traditional chilled water system with a constant air volume setup will cool the air to remove humidity (overcooling it significantly), and then use a reheat coil to bring the temperature back up to the supply setpoint. This is criminally inefficient. You're literally heating air you just spent energy cooling.

I didn't fully understand the magnitude of this until I saw a site's data historian. The reheat coil was consuming nearly 15% of the total cooling plant energy on a moderate spring day. The facility was struggling to maintain 68°F supply air with a dew point setpoint of 45°F. The mismanagement of latent heat was the real cost.

"I've never fully understood why so many designs still specify fixed dew point control without considering the actual outdoor air conditions or the server load profile. My best guess is it's a carryover from legacy 'swing' strategies that assume constant load."

The Cost of Not Fixing This

The financial impact of ignoring this goes far beyond just a higher electricity bill. It's about capacity. If your cooling system is wasting 20% of its energy on reheat and humidification, that's 20% of your cooling capacity you cannot sell. For a colocation provider, that means leaving money on the table.

Last quarter, I worked with a client whose power density per rack was increasing. They had no more capacity on their chilled water loop. The initial plan was to build a new mechanical floor—a $2 million capital expense. After a deep dive into the airside control logic, we found that by optimizing the supply air temperature reset schedule and implementing a dedicated heat pump for dehumidification (instead of overcooling), we freed up 35% of the existing coil capacity. The $2 million expansion was deferred indefinitely.

The opportunity cost of ignoring humidity control is that you are forced to overbuild your mechanical infrastructure. You add more chillers, more cooling towers, more pumps—all of which increase your operational complexity and maintenance burden. Per FTC guidelines on advertising, I can't claim this solution eliminates the need for new infrastructure in every case, but the data strongly supports the hypothesis that moisture management is a primary lever for capacity optimization.

The Solution (Short and Punchy)

So what's the actual fix? It's not magic. It's a shift in design philosophy.

• Decouple sensible and latent cooling. Use a dedicated outside air system (DOAS) or a heat pump loop for dehumidification, allowing your main AHU to focus solely on sensible cooling. This avoids the wasteful 'cool & reheat' cycle.
• Implement supply air temperature reset. Instead of a fixed 55°F or 65°F supply, let the temperature float higher (up to 75°F or even 80°F) when the IT load is low. This reduces fan energy and minimizes dehumidification load. The industry standard of 300 DPI doesn't apply here, but the concept of operating at the higher end of the ASHRAE allowable envelope is sound.
• Stop chasing chiller efficiency alone. A chiller with a 20% better EER rating is useless if your airside is wasting 30% of its energy. Look at the total system COP, not just the component nameplate data.

Granted, this requires a more sophisticated control system than a simple PID loop on the chilled water valve. But in my experience, the control upgrade costs about 10% of what a new chiller costs and delivers 3x the ROI. Best practices from 2020 suggested lower supply air temperatures; in 2025, the data supports a higher 'floating' setpoint to minimize total energy.

I get why people stick with the old way of doing things—it's simpler and less risky. But the cost of that simplicity is a persistent, invisible 20% waste on your utility bill. Dodged a bullet when I finally convinced that client to look at the psychrometric chart before buying a new chiller. Almost went with the standard 'bigger is better' approach, which would have meant spending $400k to solve a $50k problem.