“Liquid cooling” gets talked about like it’s one thing. It’s not. There are at least five distinct approaches being deployed or pitched for data center racks right now, and their performance ceilings range from 60 kW to 250 kW per rack. Choosing the wrong one for your density target is an expensive mistake.

“There’s a wide spectrum of liquid cooling technologies and they’re not all equal,” said Elvis Leka, New Product Development Engineer at Parker’s Sporlan Division, in a recent interview with The Data Center Engineer.

Here’s how they stack up. Air cooling tops out around 60 kW per rack and rear-door heat exchangers push that capacity toward 80 kW per rack. With a chip thermal design power (TDP) of roughly 600 W, according to the Open Compute Project’s OAI System Liquid Cooling Guidelines. TDP is the maximum sustained heat a chip generates under load, measured in watts. It’s the number the cooling system has to be designed around. The OCP document notes that air cooling was originally characterized at 450 W per accelerator module, with recent applications pushing that ceiling to approximately 600 W through advanced heatsinks and package improvements. 

In the examples Leka cited, oil immersion landed in roughly the same 60 to 80 kW-per-rack band. His explanation was that full submersion increases contact area, but the fluid properties still matter: oil generally moves heat less aggressively than water-based cold-plate loops or phase-change approaches. “The oil side of things is a little weaker. It’s kind of comparable to air cooling,” Leka said.

Refrigerant immersion does better, reaching about 150 kW per rack. The difference is phase change: the refrigerant boils at the chip surface, and the energy absorbed during that liquid-to-gas transition is far greater than what oil can pick up through temperature rise alone.

Single-phase direct-to-chip cooling, using a water-glycol loop to a cold plate, is also being discussed around the 150 kW-per-rack range and climbing. Two-phase direct-to-chip systems, which use a refrigerant and phase change at the cold plate, can push higher still; Leka cited examples in roughly the 175 to 250 kW-per-rack range from systems shown at OCP and SuperCompute.

The hybrid transition is already happening

Most data centers are still air-cooled. Before 2023, Leka said, about 95% were. But the transition isn’t a clean swap from one method to another. What’s actually being deployed is a mix.

“Most of these transition systems are hybrid systems where it’s air and liquid cooling together,” Leka said. “Usually, what we’ve seen is a 70 to 30% split. So 70% liquid cooled and 30% air cooled.”

In a typical hybrid rack, the highest thermal loads (GPUs, CPUs) get direct-to-chip liquid cooling while everything else on the server stays air-cooled or forced-air-cooled. That split works for current hardware. But Leka noted that new AI server blades are being designed with full liquid cooling from the start, specifically direct-to-chip, because it reaches the highest-flux areas most effectively.

Single-phase direct-to-chip with water-glycol is where most new builds and retrofits are focused. “That seems like the main path,” Leka said.

Same method, different risk

For engineers who’ve narrowed the choice to direct-to-chip liquid cooling, the next decision is which fluid. Water-glycol and refrigerant don’t just perform differently. They fail differently.

Water-glycol loops introduce an electrically consequential leak risk once real-world contamination, ionic content, and operating conditions are considered. A leak near live electronics in a multimillion-dollar rack is a major reliability and damage risk. The industry response has been aggressive detection: rope-style leak sensors around racks that trigger immediate shutdown, plus ongoing filtration and water quality monitoring. “You have a lot of bacteria growth, you have a lot of contamination, so you have to service and make sure you’re staying up to date with servicing your filters and checking water quality,” Leka said.

Refrigerant is dielectric. A leak won’t short your electronics. “Even if it comes in contact with live electricity, there’s not gonna be a shortage, so you don’t have to worry about leaks as much,” Leka said. But refrigerant systems operate at higher pressure, so a leak releases fluid with more force. Building codes and environmental regulations also apply differently, and a large refrigerant release in an enclosed space has safety implications even if the fluid itself is non-toxic.

The choice isn’t about which fluid leaks less. Both can be engineered for reliability. It’s about which failure mode your facility is better equipped to manage, and which maintenance regime fits your operations.