Liquid Cooling for Data Centers: DLC, Immersion, and Rear-Door Heat Exchangers

Liquid cooling for data centers uses circulating fluid to remove heat from servers and IT equipment, replacing or supplementing traditional air-based cooling. Three main approaches exist: direct liquid cooling (DLC) with cold plates on processors, immersion cooling that submerges hardware in dielectric fluid, and rear-door heat exchangers (RDHx) that intercept hot exhaust air at the rack boundary. The data center liquid cooling market roughly doubled in 2025 to ~$3 billion and is projected to reach $7 billion by 2029, driven by AI rack densities that now routinely exceed 100 kW, well past air cooling's practical ceiling.

This post covers why air cooling fails above 30 kW per rack, how each liquid cooling method works, a side-by-side comparison of cost, density, and PUE impact, and why factory-integrated modular data centers simplify the transition.

Why Air Cooling Fails Above 30 kW per Rack

Air-based cooling was engineered for racks drawing 6 to 15 kW. That worked fine for general-purpose servers. It does not work for AI accelerators. NVIDIA's GB200 NVL72 reference architecture demands 120 to 132 kW per rack. Meta's Catalina design reaches 140 kW. Air cannot keep up.

The physics are clear. Cooling a 50 kW rack requires approximately 7,850 CFM of airflow at a 20°F temperature differential. At 100 kW, that volume doubles, and fan power scales with the cube of fan speed. A 2021 industry whitepaper on liquid cooling places air cooling's effectiveness limit at 20 kW per rack. Industry consensus across multiple sources puts the practical ceiling at 20 to 35 kW, depending on containment strategy and room design.

Cooling already consumes 30 to 40% of total data center energy in typical enterprise facilities, according to Deloitte's 2024 analysis of data center energy consumption. The global average PUE (power usage effectiveness) has stagnated around 1.54 for six consecutive years per the Uptime Institute. Every watt spent moving air is a watt not doing compute.

What Is Direct Liquid Cooling and How Does It Work?

Direct liquid cooling, also called direct-to-chip cooling, places cold plates directly on CPUs and GPUs. Coolant circulates through a closed secondary loop managed by a coolant distribution unit (CDU), which transfers heat to the facility water system via a liquid-to-liquid heat exchanger. Cold plates capture roughly 75% of total server heat at the chip level. The remaining 25%, generated by memory, voltage regulators, and storage, dissipates through residual airflow.

DLC systems operate with warm water inlet temperatures up to 45°C, frequently eliminating mechanical chillers entirely. Outlet temperatures of 50 to 60°C make waste heat directly usable for district heating. Current cold plate technology supports rack densities of 60 to 200 kW, with CoolIT Systems' latest designs handling up to 4,000W per processor. Well-designed DLC systems achieve PUE of 1.05 to 1.15 in production.

Capital costs for DLC run approximately $1,000 to $2,500 per kW cooled, a moderate premium over air. Liquid cooling overhead is just 0.1 to 0.3 kW per kW of IT load versus 0.5 to 1.2 kW for air. DLC has become the de facto standard for large AI clusters, with DLC revenue surging 156% year-over-year in Q2 2025.

Immersion Cooling: Maximum Density, Higher Complexity

Immersion cooling submerges entire servers in dielectric fluid. Two variants exist. Single-phase immersion keeps the fluid in liquid state throughout: warm fluid circulates to an external heat exchanger and returns cooled. Two-phase immersion uses a low-boiling-point fluid that vaporizes on contact with hot components, rises to a condenser, reliquefies, and returns.

Single-phase systems support up to 200 kW per rack with verified PUE as low as 1.02 in production. Sealed immersion systems achieve near-zero water usage, a significant advantage in water-stressed regions or under regulations that penalize evaporative cooling.

The trade-offs are real. Upfront costs run $3,000 to $5,000 per kW above DLC, with dielectric fluid adding $3 to $35 per liter depending on type. Serviceability requires extracting components from fluid, with single-phase systems needing roughly 30 minutes of drip time before handling. IT vendor warranties remain a friction point: not all GPU manufacturers endorse immersion for their latest silicon. And the EU's pending PFAS restrictions, expected to tighten around 2030, threaten the two-phase fluorocarbon fluids that enable the highest density ratings.

Immersion makes the most sense for greenfield ultra-high-density deployments above 100 kW per rack, zero-water mandates, and edge sites in harsh environments where sealed enclosures double as environmental protection.

Rear-Door Heat Exchangers: The Lowest-Barrier Entry Point

A rear-door heat exchanger replaces a standard rack's rear door with a liquid-cooled coil. Passive RDHx units rely solely on server fans to push air through, consuming zero additional energy. Active units add variable-speed fans for higher capacity. Chilled water or water-glycol circulates through the coil, absorbs heat, and returns to a CDU or facility loop.

Commercial RDHx products support 30 to 80 kW per rack, with some active units claiming up to 200 kW under ideal conditions. Facility-level PUE typically falls to 1.2 to 1.3, a meaningful improvement over the 1.54 industry average.

Retrofit is RDHx's defining advantage. The Uptime Institute's 2025 Cooling Systems Survey found that 46% of respondents rank ease of retrofit as the most important factor for liquid cooling adoption. Starting at approximately $5,000 per rack, RDHx is the lowest-cost liquid cooling entry point and the natural first step for existing facilities that need incremental density increases without structural changes.

Liquid Cooling Comparison: DLC vs. Immersion vs. Rear-Door Heat Exchanger

How EU Regulation Drives Liquid Cooling Adoption

Two regulatory frameworks matter. The EU Energy Efficiency Directive (EED, Directive 2023/1791) mandates annual reporting of PUE, water usage, and energy reuse for data centers with IT power demand of 500 kW or more. The EED creates transparency pressure but does not set binding PUE targets.

Germany's Energy Efficiency Act (EnEfG) goes further. New facilities commissioned from July 1, 2026 must achieve PUE of 1.2 or better within two years. Waste heat reuse requirements escalate from 10% in 2026 to 20% by 2028. Air-cooled facilities averaging PUE 1.7 in Germany cannot meet these thresholds without fundamental redesign.

Liquid cooling directly enables compliance. DLC systems routinely hit PUE 1.05 to 1.15. Liquid loops deliver waste heat at 40 to 60°C versus 25 to 35°C for air, which dramatically improves heat pump efficiency and makes district heating integration economically viable. Cities like Stockholm already recover enough data center waste heat to warm approximately 30,000 apartments annually.

Why Factory-Built Modules Simplify Liquid Cooling Deployment

Integrating liquid cooling in the field means routing piping through a live facility, coordinating CDU placement, running pressure tests, and managing leak risk across dozens of connection points. It is complex, slow, and error-prone.

Factory-built modular data centers shift this work into a controlled manufacturing environment. CDUs, manifolds, piping, cold plates, leak detection, and power distribution are pre-assembled, interconnected, and pressure-tested before shipping. Site preparation happens in parallel with factory assembly. Industry data shows modular approaches reduce construction timelines by approximately 30%, with some liquid-cooled modular deployments compressing 18-month builds to under six months.

For operators evaluating liquid cooling, a factory-integrated module eliminates the highest-risk variable: on-site plumbing. The cooling architecture arrives validated. The module connects to site power and network. Compute goes live.

FAQ

At what rack density does air cooling stop working?

Air cooling becomes impractical between 20 and 35 kW per rack, depending on containment design and room layout. Above 30 kW, fan power requirements and airflow volumes escalate to the point where liquid cooling becomes both more effective and more energy-efficient.

What is the difference between direct liquid cooling and immersion cooling?

Direct liquid cooling places cold plates on individual processors and circulates coolant through a closed loop. Only the hottest components contact liquid. Immersion cooling submerges entire servers in dielectric fluid, cooling all components simultaneously. DLC is operationally simpler and more widely deployed; immersion supports higher densities but introduces fluid handling complexity.

How much does data center liquid cooling cost?

DLC systems cost approximately $1,000 to $2,500 per kW of cooling capacity. Immersion cooling carries a premium of $3,000 to $5,000 per kW above DLC, largely driven by fluid costs and specialized enclosures. Rear-door heat exchangers start at roughly $5,000 per rack, the lowest liquid cooling entry point.

Can liquid cooling be retrofitted into an existing data center?

Yes. Rear-door heat exchangers are specifically designed for retrofit: they replace the rear door of a standard rack and connect to chilled water. DLC can also be retrofitted with CDU installation and rack-level piping. Immersion cooling is the hardest to retrofit because it requires purpose-built tanks and structural load assessment.

Does liquid cooling help meet EU energy efficiency regulations?

Yes. Germany's EnEfG requires new data centers from July 2026 to achieve PUE of 1.2 or better, with escalating waste heat reuse mandates. DLC routinely achieves PUE 1.05 to 1.15 and delivers waste heat at temperatures suitable for district heating integration, directly enabling compliance on both fronts.

What is a rear-door heat exchanger and when should you use one?

A rear-door heat exchanger (RDHx) replaces a standard rack's rear door with a liquid-cooled coil that intercepts hot exhaust air. It supports 30 to 80 kW per rack with minimal infrastructure changes. RDHx is the right choice for brownfield facilities that need incremental density increases without the complexity of full DLC or immersion.

How does liquid cooling improve PUE in a data center?

Liquid cooling removes heat more efficiently than air, reducing the energy spent on cooling infrastructure. DLC systems achieve PUE of 1.05 to 1.15 versus the 1.54 global average for air-cooled facilities. Immersion cooling can push PUE below 1.05. This translates directly to lower electricity costs per unit of compute delivered.

Which liquid cooling technology partners integrate with modular data centers?

Innovators such as Iceotope (precision chassis-level cooling), ZutaCore (waterless two-phase direct-to-chip), and Nexalus (microjet impingement cold plates) all produce rack-compatible solutions that can be factory-integrated into modular enclosures. Pre-integration in a controlled factory environment eliminates the on-site plumbing risk that makes field deployment of liquid cooling complex.

‍