Liquid cooling for data centers uses fluid to transfer heat directly from server components, achieving power usage effectiveness (PUE) as low as 1.03 compared to 1.3 or higher with traditional air cooling. You need this technology when rack densities exceed 40 kW, which is now standard for AI and high-performance computing workloads.
How Liquid Cooling Works in a Data Center
Liquid cooling is a thermal management method that circulates a coolant, typically water or engineered dielectric fluid, to absorb heat from processors, GPUs, and memory modules. The liquid carries thermal energy to a heat exchanger or cooling distribution unit (CDU), where it rejects heat to an external loop connected to a dry cooler or cooling tower. The cycle repeats continuously, maintaining chip junction temperatures within safe operating ranges.
Air cooling moves heat by blowing cold air across heatsinks attached to components. This approach works up to roughly 15 to 25 kW per rack. Beyond that threshold, the volume of air required becomes impractical. You would need massive fans, elevated floor heights, and enormous CRAC units that consume significant energy. Liquid carries heat 3,000 times more efficiently than air by volume, which is why every hyperscaler building AI clusters has moved toward liquid-based data center cooling methods.
Direct-to-Chip vs Immersion Cooling: Two Liquid Approaches
Direct-to-chip cooling, also called cold plate cooling, attaches metal plates with internal fluid channels directly to CPUs and GPUs. Warm coolant flows through facility piping to a CDU. This method targets only the highest heat-generating components while the rest of the server uses conventional air. Dell, HPE, and Lenovo all ship servers with direct-to-chip options, and NVIDIA requires liquid cooling for its GB200 NVL72 racks rated at 120 kW.
Immersion cooling submerges entire servers in a tank of dielectric fluid. Single-phase immersion keeps the fluid below its boiling point. Two-phase immersion allows the fluid to boil on hot surfaces, with the vapour condensing on a built-in condenser and dripping back into the tank. When comparing immersion cooling vs air cooling, immersion eliminates fans entirely, reduces energy consumption by 30 to 50%, and enables rack densities above 100 kW.
Liquid Cooling Methods Compared
| Method | Max Rack Density | PUE Range | Deployment Complexity |
|---|---|---|---|
| Air Cooling | 15-25 kW | 1.3-1.5 | Low |
| Direct-to-Chip | 80-120 kW | 1.05-1.15 | Medium |
| Single-Phase Immersion | 100+ kW | 1.02-1.06 | High |
| Two-Phase Immersion | 150+ kW | 1.01-1.03 | High |
Who Uses Liquid Cooling for Data Centers Today
Microsoft uses direct-to-chip cooling across its Azure AI infrastructure and has tested two-phase immersion cooling since 2021. Google deploys custom liquid cooling in its TPU training clusters. Meta requires liquid cooling for all new AI training facilities supporting its Llama model development. NVIDIA’s reference architecture for the DGX SuperPOD mandates rear-door heat exchangers or direct-to-chip cooling at minimum.
Colocation providers are retrofitting for liquid cooling demand. Equinix offers liquid-ready cabinets in over 60 facilities globally. Vertiv, the leading thermal management vendor, reported a 50% year-over-year increase in liquid cooling orders through Q3 2025. CoolIT Systems, the largest direct-to-chip OEM, has shipped over 5 million cold plates. The market for liquid cooling in data centers is projected to reach $10.2 billion by 2028, according to Dell’Oro Group.
Cost and ROI of Liquid Cooling Deployment
A liquid cooling retrofit costs $5,000 to $15,000 per rack depending on density requirements and whether you choose direct-to-chip or immersion. That investment pays back within 18 to 30 months through reduced electricity consumption. A 10 MW facility dropping from 1.4 PUE to 1.1 PUE saves approximately $1.5 million per year in cooling energy costs at $0.06 per kWh. You also recover floor space because liquid-cooled racks pack more compute into fewer cabinets, reducing your real estate footprint by 30 to 60% compared to air-cooled equivalents. For facilities supporting AI workloads with growing AI data center power consumption, liquid cooling is not optional.
Frequently Asked Questions
What is liquid cooling in a data center?
Liquid cooling in a data center is a thermal management system that uses water or dielectric fluid to absorb and remove heat from servers. The coolant circulates through cold plates attached to processors or fully submerges hardware in fluid tanks, achieving PUE values between 1.01 and 1.15 depending on the method.
Is liquid cooling better than air cooling for data centers?
Liquid cooling outperforms air cooling above 25 kW per rack. It reduces cooling energy by 30 to 50%, supports densities up to 150 kW per rack, and lowers PUE from 1.3 to as low as 1.03. For AI workloads running GPUs like the NVIDIA B200, liquid cooling is the only viable thermal solution.
How much does liquid cooling cost for a data center?
Direct-to-chip liquid cooling costs $5,000 to $10,000 per rack for retrofits. Full immersion cooling runs $10,000 to $15,000 per rack including tanks and CDUs. A 10 MW facility typically recovers the investment within 18 to 30 months through reduced electricity costs and higher compute density per square foot.