Cooling Water and Coolant Filtration for AI & Data Centers

Coolant filtration is becoming increasingly important in modern data centers. In the digital world, different types of data centers serve distinct purposes. Traditional data centers primarily address “storage and transmission” challenges, managing massive amounts of information like a warehouse or library.

AI compute centers, on the other hand, focus on “intelligent computing”, handling tasks such as deep learning, scientific simulations, and autonomous driving training, supporting applications like ChatGPT.

As computing demand continues to grow, a 500 kW AI compute center can produce 5–7 m³ of air-conditioning condensate per day. The combined pressures of expanding compute capacity, high reliability requirements, and water-efficient operations are driving rapid growth in the data center filtration market.

Why Is Cooling Important in Modern Data Centers?

As computing demands grow and server racks become more densely packed, traditional air-cooling systems increasingly struggle to manage the resulting heat load. To address this, data centers are adopting liquid cooling solutions.

In liquid cooling systems, a heat-transfer fluid transports heat away from servers and other equipment. The fluid can be water, used for cold plate cooling, or a dielectric fluid, used for immersion cooling. It then delivers the heat to heat exchangers, cooling towers, or other cooling sources.

Liquid cooling systems offer superior heat removal and allow for more compact designs. However, the performance of these systems depends heavily on fluid quality.

Contaminants such as particles, biofilm, corrosion byproducts, and sediment can reduce heat transfer efficiency, clog microchannels, or degrade dielectric fluids. They can also corrode metal components, ultimately affecting the overall reliability of the system.

For this reason, filtration and purification of the heat-transfer fluid is a vital part of designing high-reliability, high-efficiency liquid cooling systems, helping to maintain stable operation and protect equipment.

Cooling for AI Servers: What’s the Challenge?

Managing heat-transfer fluids in high-density AI server environments presents unique challenges compared to conventional industrial or commercial cooling systems. The combination of extremely high heat loads, continuous operation, and sensitive equipment requires specialized filtration solutions.

Industry-Specific Considerations

High-density AI servers generate substantial heat, necessitating cooling systems that operate at continuous high flow rates over extended periods. Any downtime for maintenance or filter replacement can lead to significant operational disruption and increased costs.

Filtration Requirements

• High Flow and Longevity: Filters must handle large volumes of fluid continuously without frequent replacement, ensuring long service life and minimizing operational interruptions.
• Water Quality Characteristics: Cooling water typically contains suspended particles, but microbiological contamination is generally low. The focus is therefore on absolute filtration of particulates with minimal pressure drop, rather than demineralization or sterilization.
• Chemical Compatibility: Although the chemical environment is relatively mild, water treatment chemicals such as antiscalants, corrosion inhibitors, and biocides are often used. Filters must be compatible with these chemicals without degradation, rather than being resistant to extreme pH or high salinity.

Types of Liquid Cooling Systems

Understanding the type of liquid cooling system is essential, as it determines the filtration requirements and fluid selection. In high-density AI server environments, liquid cooling systems generally fall into two types: cold plate cooling and immersion cooling.

Cold Plate Cooling uses liquid to carry heat away from CPUs or GPUs through plates attached to each chip. This method allows servers to run efficiently and keeps temperatures stable. The liquid needs to stay clean to avoid blocking the small channels in the system.

Immersion Cooling submerges entire servers in a special cooling liquid, letting heat escape evenly from all components. This approach makes server designs more compact and energy-efficient. Keeping the liquid clean is important to prevent problems and ensure long-term reliability.

Traditional data centers mostly use cold plate cooling, while AI-focused centers often use immersion cooling. Both rely on clean cooling liquid to protect sensitive equipment and keep servers running efficiently.

Water-Based vs. Non-Water-Based Coolants

Effective cooling water management ensures system reliability and efficiency. Proper fluid selection and multi-stage filtration prevent corrosion, scaling, microbial growth, and downtime in industrial and data center systems.

Cooling Fluid Types

Cooling fluids are categorized into water-based and non-water-based types. Water-based coolants include deionized water and glycol solutions, formulated to maintain low conductivity, controlled pH, and minimal hardness.

Glycol solutions provide freeze protection and are enhanced with corrosion inhibitors, antioxidants, and biocides, while propylene glycol variants are suitable for low-toxicity or food-grade applications.

Non-water-based coolants include organic oils for immersion cooling and fluorinated liquids for cold-plate or two-phase cooling, selected according to their boiling points.

Multi-Stage Cooling Water Filtration

Cooling water treatment in modern data centers typically follows a multi-stage process to ensure high reliability and water quality. In high-density AI server environments, filters must meet several critical requirements:

1. Pretreatment: Cooling and Coarse Filtration

Condensate from data centers usually ranges from 15–30°C, and can be hotter in AI compute centers with liquid cooling. Plate heat exchangers lower the temperature below 50°C, while coarse filters remove large particles like dust and rust, preparing the water for more precise treatment.

2. Primary Filtration: Precision Particle Removal

Fine filters, including high flow filter cartridges or bag filters, remove metal micro-particles and colloids, keeping turbidity under 1 NTU. Activated carbon filtration can further reduce organic residues and oil, ensuring the water meets quality requirements for sensitive cooling equipment.

3. Advanced Treatment: Ion Removal and Sterilization

Dual-stage reverse osmosis lowers conductivity to ≤5 μS/cm in AI compute centers, while single-stage RO or softening resins suffice for standard data centers. For ultra-pure water, electrodeionization or polishing mixed-bed units further reduce conductivity. UV or ozone sterilization controls microorganisms, keeping microbial counts below 1 CFU/mL.

4. Post-Treatment: Water Conditioning and Reuse

Finally, water is conditioned for reuse. pH is stabilized at 7.0–8.5 to prevent corrosion, and terminal 0.2 μm PP pleated filter cartridges removes residual particles. IoT sensors continuously monitor water quality, triggering alarms, switching to backup filters, and adjusting flow through the Cooling Distribution Unit (CDU) to ensure stable, reliable operation.

Conclusion

Choosing the right cooling fluids and applying multi-stage filtration help maintain stable and efficient operation in both traditional and AI-focused data centers. Clean liquids, controlled water quality, and protected components allow data centers to run smoothly, minimize downtime, and handle increasing computational demands. BOLEFIL can provide customized filtration solutions for modern data center cooling systems.