Data center power densities have surged from 4 kW per rack to 12-30 kW and beyond, driven by AI workloads that demand 10× more electricity than traditional searches. Traditional air cooling can no longer handle these thermal loads, making liquid cooling not just preferable but essential for modern GPU and CPU clusters.
- Why Liquid Cooling Dominates
- Single-Phase Systems
- Two-Phase Systems
- Refrigerant Selection: Engineering Trade-Offs
- Safety Classification
- System Design Pressures
- Volumetric Flow Considerations
- Regulatory Landscape
- Hose, Tubing, and Connection Design
- Material Selection Criteria
- Rubber vs. Thermoplastic
- Quick Disconnect Couplings
- Sustainability Impact
- Implementation Guidance
Why Liquid Cooling Dominates
Liquid coolants carry over 1,000 times more heat than air while reducing energy consumption, water usage, noise, and operating costs. Two primary architectures have emerged: single-phase and two-phase direct-to-chip systems, each with distinct engineering considerations.
Single-Phase Systems
Single-phase systems circulate water-glycol mixtures or deionized water through cold plates mounted directly on processors. The coolant absorbs sensible heat without changing phase, requiring cooling distribution units (CDUs) and facility water loops with chillers and cooling towers.

Key characteristics:
- Coolant temperature rises as it absorbs heat
- Requires larger volumetric flow rates
- Well-understood infrastructure and maintenance
- Proven track record in hyperscale deployments
Two-Phase Systems
Two-phase systems leverage latent heat during refrigerant phase change, maintaining uniform chip temperatures through controlled boiling. Liquid refrigerant flows to cold plates where it evaporates, then returns to a condenser as vapor.

Advantages:
- Superior thermal uniformity on chip surfaces
- Reduced system weight and simplified plumbing
- Minimal water consumption (near-zero in many climates)
- Better thermal performance/cost ratio
- Lower pumping power due to efficient heat transfer
Refrigerant Selection: Engineering Trade-Offs
Choosing a refrigerant for two-phase systems involves balancing safety, environmental impact, and system design requirements.
Safety Classification
A1 refrigerants (non-flammable) avoid restrictions imposed on A2L (mildly flammable) options, simplifying compliance and reducing training requirements. R-1233zd(E) and R-1336mzz(Z) achieve A1 classification with GWP values of 1-2, representing 97-99% reductions versus older refrigerants.
System Design Pressures
Operating pressures directly impact component costs and reliability:
| Refrigerant | Design Pressure | Application Notes |
|---|---|---|
| R-513A, R-1234yf | 325-350 psig | Similar to R-134a systems |
| R-1234ze(E), R-515B | 250-275 psig | Moderate pressure reduction |
| R-1233zd(E), R-1336mzz(Z) | 70-80 psig | Very low pressure, large vapor volumes |
Volumetric Flow Considerations
Low-pressure refrigerants enable smaller liquid-side components but require 3-6× larger vapor return piping compared to R-134a-like options. At 70°F, R-1336mzz(Z) needs 9.86 ft³/min vapor flow per ton versus 1.43 ft³/min for R-134a.
Design implication: Cost savings from lower-pressure ratings must be weighed against upsized return manifolds and increased pressure drop sensitivity.
Regulatory Landscape
PFAS regulations will affect two-phase cooling systems, requiring proactive refrigerant strategy and potential operator certification. Engineers should monitor ANSI/ASHRAE Standard 34 updates and EPA SNAP approvals when specifying systems.
Hose, Tubing, and Connection Design
Flexible connections between components face demanding requirements in two-phase systems.
Material Selection Criteria
Permeability: Vapor-phase refrigerants can migrate through hose walls; barrier layers minimize loss but reduce flexibility.
Pressure rating: Design factors of 3:1 to 4:1 ensure burst pressure exceeds maximum working pressure.
Fluid compatibility: Materials must resist extraction and degradation; polychloroprene, NBR, HNBR, PTFE, polyamide, and PPS have proven compatibility with R-513A, R-515B, and their constituents.
Dimensions: Inside diameter affects pressure drop and heat transfer; outside diameter determines space claim in dense server racks.
Rubber vs. Thermoplastic
Rubber hose (EPDM, nitrile, polychloroprene) offers superior flexibility, tight bend radius, and simple barb connections with low assembly force.
Thermoplastic options (polyamide, PTFE, PFA) provide better chemical resistance, lower permeation, and thinner walls but may require O-ring seals and higher assembly forces.
Quick Disconnect Couplings
QD dry-break couplings enable hot-swapping servers without system shutdown. Critical performance metrics include:
- Low pressure drop at required flow rates
- Minimal fluid loss and air inclusion during connection
- Compact length for rack flexibility
- High vacuum capability
- Tool-free operation
Sustainability Impact
Traditional data centers consume billions of liters of water annually with projected double-digit growth. Two-phase systems eliminate most water consumption, offering a clear path to sustainable cooling infrastructure.
Implementation Guidance
When specifying direct-to-chip cooling:
- Match cooling architecture to rack density: Air cooling remains viable below 12 kW/rack; single-phase for 12-30 kW; two-phase for extreme densities
- Prioritize A1-classified refrigerants to simplify compliance and operational requirements
- Balance system pressure against component sizing – very low-pressure refrigerants reduce some costs but increase vapor piping complexity
- Evaluate material compatibility across the entire fluid loop, including seals and gaskets
- Design for serviceability with quick-disconnect couplings and accessible manifolds









