Why Data Centers Are Moving from Air to Liquid Cooling?

Air cooling has a physical ceiling, and AI workloads are running straight into it. As compute density climbs — more GPUs, more accelerators, packed tighter per rack — the heat generated per square foot has outpaced what fans and chilled air can practically remove.

Liquid cooling is the industry's answer, and it's worth understanding as an engineering system, not just a buzzword.

Why Air Runs Out of Road ?

Air is a poor heat-transfer medium compared to liquid — it simply can't carry heat away fast enough once rack power density crosses a certain threshold. High-performance computing and AI training clusters are exactly the workloads pushing past that threshold, since they concentrate enormous, sustained power draw into small physical footprints.

Liquid, having a much higher heat capacity and thermal conductivity than air, can absorb and move far more heat through a much smaller physical interface — which is the entire engineering case for switching.

Two Main Approaches

Direct-to-Chip (D2C) cooling brings a liquid coolant into direct contact with, or very close proximity to, the chip itself — typically through a cold plate mounted on the processor. Heat transfers from the chip into the coolant, which then carries it away to be rejected elsewhere.

Immersion cooling goes further: entire servers are submerged in a non-conductive dielectric fluid, which absorbs heat directly from all the components at once rather than just the processor.

Both approaches can run in one of two modes:

• Single-phase — the coolant stays liquid throughout the cycle, absorbing and releasing heat purely through temperature change

• Two-phase — the coolant evaporates into vapor as it absorbs heat, then condenses back to liquid as it releases it, taking advantage of the much larger energy exchange involved in a phase change

The Component That Makes It Work: The Coolant Distribution Unit

Getting liquid to the chip is only half the system. Something has to manage that liquid's flow, temperature, and pressure across an entire facility's worth of racks — and that's the job of a Coolant Distribution Unit (CDU).

A CDU sits at the center of what's called a Technology Cooling System (TCS) — a closed loop that delivers cooled fluid out to the racks and takes the warmed fluid back in for re-cooling. Its core function is straightforward to state but non-trivial to execute at scale: deliver the right coolant flow, at the right temperature, to every rack, continuously, regardless of how the workload across the facility is fluctuating moment to moment.

Getting this regulation right is what actually determines whether a liquid cooling deployment delivers on its promised reliability and energy savings — an undersized or poorly controlled CDU can bottleneck an otherwise well-designed system just as easily as an undersized borefield can undermine a geothermal cooling installation.

Why This Represents a Real Engineering Shift, Not Just a Swap?

Moving to liquid cooling isn't a drop-in replacement for air cooling — it changes the facility's plumbing, its failure modes (leak detection and containment become genuinely important design considerations), and its maintenance procedures. Two-phase systems in particular introduce refrigerant-like phase-change engineering into a space that, until recently, was almost entirely electrical and airflow-focused.

That's part of why this shift is happening at the infrastructure level, not just the equipment level — new data center builds increasingly plan around liquid cooling from the ground up, rather than retrofitting air-cooled facilities after the fact.

Further Reading: www.coolingindia.in/advanced-desiccant-regeneration-methods/ , ACHR News