AI infrastructure is driving a rapid shift toward direct-to-chip (DTC) liquid cooling.
In a direct-to-chip architecture, coolant is circulated through cold plates mounted directly on high-heat components such as GPUs and CPUs. The coolant absorbs heat at the server and transfers it through the technology cooling system (TCS), typically through a coolant distribution unit (CDU), to the facility-side cooling system.
That makes coolant selection more than a fluid decision. Coolant properties affect heat transfer, pressure drop, pump selection, corrosion protection, system reliability, and long-term maintenance.
For systems requiring freeze protection, glycol-water coolants are commonly considered. The two most relevant options for data center cooling are propylene glycol (PG) and ethylene glycol (EG).
Why Glycol Is Used in Direct-to-Chip Cooling
Water provides excellent heat-transfer performance, but it does not provide freeze protection below 0°C.
Adding glycol can provide three important functions:
- Lowering the coolant freeze point
- Increasing boiling-point protection
- Providing corrosion protection when used with the correct inhibitor package
The tradeoff is that adding glycol changes the thermophysical properties of the fluid. Compared with water, glycol solutions generally increase viscosity and reduce heat-transfer performance.
For AI cooling systems, that means glycol concentration should be selected as part of the overall thermal and hydraulic design rather than treated as an isolated specification.
Coolant properties directly influence CDU and pump selection, flow requirements, pressure drop, and cold-plate thermal performance.
Propylene Glycol vs. Ethylene Glycol for AI Cooling
Propylene Glycol
Propylene glycol is particularly relevant for the server-side technology cooling system.
PG is less toxic than ethylene glycol and has become the preferred glycol in several industry frameworks for single-phase cold-plate cooling.
Open Compute Project guidance for propylene glycol-based heat-transfer fluids in single-phase cold-plate racks specifies the use of PG as the freeze-point depressant together with an inhibitor package compatible with the metals present in the TCS loop.
The tradeoff is hydraulic and thermal performance.
Compared with ethylene glycol at equivalent concentrations, propylene glycol generally has:
- Higher viscosity
- Lower thermal conductivity
- Greater pumping requirements, particularly as temperature decreases
Those differences matter in high-density AI environments because coolant flow, pressure drop, pump power, and cold-plate performance are closely connected.
The objective should therefore be to use only the glycol concentration necessary to meet the required freeze protection and system operating conditions.
PG25 Provides Biostatic Protection
An additional advantage of PG25 in direct-to-chip cooling is its biostatic nature.
At approximately 25% propylene glycol concentration, the coolant environment inhibits sustained microbial growth. This can help reduce the risk of biological fouling and biofilm formation in closed technology cooling systems, including CDUs, manifolds, piping, and cold-plate flow passages.
This distinction is important: biostatic does not mean biocidal. PG25 is intended to inhibit microbial growth rather than sterilize the cooling loop or kill all microorganisms that may be introduced.
Maintaining the specified glycol concentration is therefore important. At lower glycol concentrations, the biological protection can diminish, while increasing glycol concentration beyond what is required can impose additional viscosity and heat-transfer penalties.
For AI direct-to-chip systems, PG25 represents a useful balance between biological stability, freeze protection, corrosion protection, and thermal/hydraulic performance.
Ethylene Glycol
Ethylene glycol generally provides better heat-transfer and pumping performance than propylene glycol at equivalent concentrations.
Its lower viscosity and higher thermal conductivity can reduce some of the hydraulic penalties associated with glycol-water mixtures.
However, EG has a significantly different toxicological and regulatory profile than PG. That makes fluid location within the cooling architecture important.
Rather than asking whether PG or EG is universally better, AI data center designers should evaluate which fluid is appropriate for each cooling loop.
Separate the Facility and Technology Cooling Loops
One of the most important design principles in direct-to-chip cooling is separation between the facility water system and the technology cooling system.
A CDU typically provides the boundary between:
Facility Water System (FWS)
The facility-side loop that moves heat between the data hall and the broader heat-rejection infrastructure.
Technology Cooling System (TCS)
The server-side loop that circulates coolant through rack manifolds, hoses, connectors, and cold plates.
Separating these loops allows each side to use fluid chemistry appropriate for its specific operating environment.
This is particularly important for AI systems because the coolant circulating near GPUs, CPUs, cold plates, and other IT equipment has different requirements from fluid circulating through facility infrastructure.
PG may be selected for a technology loop because of its lower toxicity and suitability for cold-plate applications, while a facility loop may use a different fluid based on climate, freeze protection, materials, and thermal-performance requirements.
The key point is that coolant selection should be made loop by loop, not for the data center as a whole.
Glycol Concentration Directly Affects Cooling Performance
More glycol is not automatically better.
Increasing glycol concentration improves freeze protection, but it also changes:
- Specific heat
- Thermal conductivity
- Density
- Viscosity
- Pumping requirements
- Pressure drop
- Overall heat-transfer performance
These effects become especially important in AI systems, where high heat loads must be removed through relatively compact cold plates and liquid distribution networks.
Lower glycol concentrations generally provide better heat-transfer performance and lower pumping penalties than higher concentrations.
For this reason, glycol concentration should be based on the required operating and freeze-protection envelope rather than selected as a generic percentage.
Corrosion Protection Is Critical
Glycol alone should not be treated as a complete coolant formulation.
Uninhibited glycol can degrade and form acidic compounds that lower coolant pH and promote corrosion.
A properly formulated coolant therefore requires an inhibitor package compatible with the metals and materials present throughout the technology cooling loop.
That includes components such as:
- Cold plates
- CDU heat exchangers
- Pumps
- Manifolds
- Tubing and piping
- Valves
- Quick-disconnect couplings
- Other wetted components
Material compatibility should be evaluated across the entire liquid loop.
Water Quality Matters
The water used to prepare a glycol-water coolant can directly affect coolant stability, corrosion protection, and system life.
Poor water quality can introduce hardness, sulfate, or other contaminants that interfere with coolant chemistry or contribute to deposits and corrosion.
High-quality dilution and makeup water should therefore be treated as part of the coolant specification.
For direct-to-chip systems, fluid cleanliness is particularly important because contaminants can affect small flow passages, heat exchangers, and cold-plate performance.
Coolant Monitoring Should Be Part of AI Cooling Operations
Coolant condition changes over time.
ASHRAE guidance emphasizes coolant-quality monitoring and filtration because degraded fluid chemistry can reduce efficiency and increase energy consumption.
A monitoring program for glycol-based DTC cooling should include parameters such as:
- Glycol concentration
- pH
- Reserve alkalinity
- Visual clarity
- Filtration condition
Glycol monitoring should be integrated into the operating strategy for the liquid-cooling system rather than treated only as a maintenance response after a problem occurs.
What Matters Most for AI Direct-to-Chip Cooling
For AI data centers, the coolant should be evaluated as part of the complete thermal system.
The most important questions are:
Which loop is the coolant serving?
Facility-side and technology-side loops can have different requirements.
How much freeze protection is actually required?
Additional glycol can impose thermal and hydraulic penalties.
How does the coolant affect flow and pumping?
Viscosity, specific heat, thermal conductivity, and density influence CDU sizing and system performance.
Is the inhibitor chemistry compatible with every wetted material?
Cold plates, manifolds, pumps, heat exchangers, connectors, and piping all need to be considered.
How will coolant quality be maintained?
Water quality, filtration, concentration, pH, and inhibitor condition all influence long-term reliability.
Selecting Coolant for AI Cooling Infrastructure
There is no single glycol formulation that is optimal for every liquid-cooled AI deployment.
A well-designed direct-to-chip system balances:
Thermal performance + hydraulic performance + freeze protection + material compatibility + operational safety + coolant stability.
For server-side technology cooling systems, propylene glycol-based inhibited coolants have gained significant industry attention because of their lower toxicity and alignment with cold-plate cooling guidance.
The final concentration and formulation, however, should be selected against the requirements of the specific TCS architecture, CDU, cold plates, materials, operating temperatures, and environmental conditions.
For AI direct-to-chip cooling, the goal is not simply to choose a glycol.
The goal is to choose a coolant that helps the entire liquid-cooling system remove heat from high-power silicon reliably and efficiently.
