Direct-to-chip cooling is a liquid cooling method that uses a specialized water coolant or blended glycol mixture. This coolant/mixture is circulated in close proximity to the heat-generating components of data centers or electronic systems such as CPUs, GPUs, and other heat-generating hardware. While direct-to-chip cooling offers improved energy efficiency and thermal performance, it also raises a range of technical concerns that must be considered during the design, installation, and operation phases.

One of the primary environmental concerns with direct-to-chip cooling systems is related to design, installation, and maintenance. Even though these systems often recirculate the coolant/mixture via closed-loop systems, poor design, installation, and/or maintenance (e.g., improperly secured fittings, pump failures, etc.) can lead to leaks or spills, which can significantly increase water consumption/result in high consumption rates. This effect is especially problematic in regions facing water scarcity. Modern CDU (Coolant Distribution Unit) architecture addresses these risks through hydraulic isolation. The facility water loop never directly contacts IT equipment, instead transferring heat through brazed plate heat exchangers to secondary loops serving the racks. This isolation prevents cross-contamination while enabling independent chemistry control. Current designs typically employ redundant pump configurations (N+1), allowing partial maintenance without full system shutdown, as seen in Google’s internal Project Deschutes (deployed since 2020 with 99.999% uptime across 2,000 TPU pods, with a fifth-generation CDU announced for OCP contribution in 2025).

Another concern, especially in data centers or HPC environments, is the trade-off between using pure water and a blended glycol mixture (typically with ethylene glycol or propylene glycol, e.g., PG25). Each has its pros and cons. The primary benefits of using pure water are its low viscosity and the fact that it is non-toxic and environmentally friendly. Pure water has greater heat transfer efficiency and is easier to pump, which reduces power consumption and system strain—a net positive. Pure water is also virtually hazard-free/less toxic, with easy disposal options, which may allow for traditional disposal via the sanitary sewer. However, pure water requires “high-quality” water; it must be deionized (DI) or have strict filtration to help avoid scaling and corrosion. Corrosion inhibitors or well-selected materials of composition may also be needed (e.g., only copper or stainless steel) to protect against corrosion. Finally, pure water creates an environment where there is the potential for biological growth; pure water can easily support bacteria, algae, and/or mold growth.

Alternatively, a blended glycol mixture could be used instead of pure water. Unfortunately, blended glycol mixtures also have pros and cons. Positive aspects include the fact that blended glycol blends often include corrosion inhibitors like phosphates and molybdates to protect metal and mixed-metal surfaces. At the same time, anti-scalants such as polyacrylates help prevent mineral buildup. Also, suppose the glycol mixture is maintained at the optimal level of 25%. In that case, biocides are typically not needed because the concentration is biostatic—it inhibits the growth of bacteria on its own. A blended glycol mixture can further create greater system stability with more tolerance of leaks, oxygen ingress, and suboptimal maintenance, assuming proper maintenance and equipment replacement procedures are in place.

On the other hand, blended glycol mixtures have lower heat transfer efficiency and a high viscosity, which increases the pumping power/energy demand requirements and pressure drop in the system. From a thermal perspective, these differences are quantifiable: DI water at 20–40°C has a Cp ≈ 4.18 kJ/kg·K, while PG25 has a Cp ≈ 3.85 kJ/kg·K (~8% lower). Further, viscosity is ~2.5× higher at 20°C (2.5 vs 1.0 mPa·s), moderating to ~114% higher at 40°C, and thermal conductivity is ~19% lower (0.485 vs 0.60 W/m·K). The cumulative effect is that PG25 systems typically require ~15–20% higher flow to achieve equivalent cooling capacity, while pumping power consumption increases ~25–30% due to viscosity. System designers must account for these concerns when sizing and operating CDUs. Moreover, industry standards specify flow rates of 1.25–2.0 L/min/kW (with 1.5 L/min/kW as the typical design target) for maintaining ΔT ≤10°C at maximum thermal design power. Applications requiring tighter temperature control may operate at up to 2.88 L/min/kW for 5°C ΔT, particularly in high-performance computing scenarios. When using PG25, flow rates must be adjusted upward by 15–20% due to its reduced thermal capacity and higher viscosity.

Another consideration, ethylene glycol is toxic, while propylene glycol is safer (relatively non-toxic), but it is still regulated. For example, PG25 (Propylene Glycol 25%) and system flush require compliance with local environmental and wastewater regulations. Although some localities may allow, in rare cases, PG25 to be flushed into the sanitary sewer in small quantities with plenty of water, it is important to check with the local regulator because most localities often require disposal through a licensed waste management company. For system flush, PG25 should be drained from the system into a suitable, labeled container and flushed thoroughly with water or a recommended cleaning agent—check manufacturer guidelines.

More specifically, the initial flush water should be captured if it contains high concentrations of PG25. Dispose of it similarly as you would with the original fluid and continue flushing until water runs clear and tests neutral (if testing is available onsite). Finally, disposal options should include (if possible) recycling, as some waste disposal companies recycle glycol-based products. But remember to always follow your internal chemical waste protocols. If these water/blended glycol mixtures are not properly contained, managed, and disposed of, they can pose significant environmental hazards and may result in reportable events. For instance, accidental releases or improper discharges into natural water bodies most likely will result in a reportable event and can lead to toxicity in aquatic life, disruption of ecosystems, and/or contamination of drinking water sources. To mitigate blended glycol mixture risks, stringent containment practices, proper wastewater treatment and/or disposal procedures, and regulatory compliance practices are essential elements for any organization.

A final concern is thermal pollution, where warm water is discharged back into the environment, potentially impacting aquatic ecosystems. In these situations, data centers are often required to cool their blow-down prior to discharge, mitigating this issue. Overall, as data center or electronic system demands grow, balancing cooling efficiency with responsible use becomes increasingly critical.

From a health and safety perspective, the use of pressurized water systems inside or near data center white spaces introduces additional considerations such as personnel safety, electrical exposure and shorts, equipment damage, and fire hazards. Workers servicing or inspecting direct-to-chip cooling systems may face exposure to hot surfaces, high-pressure fluids, and hazardous chemicals used in water treatment. These risks can lead to burns, chemical exposure, or even sudden equipment failure. To prevent these types of injuries, it is essential to implement strict control of hazardous energy (CoHE)lockout/tagout (LOTO) procedures, ensure systems are depressurized before maintenance or repair, and use proper personal protective equipment (PPE) protocols. Such PPE includes, but is not limited to, insulated gloves, face shields, and chemical-resistant clothing.

In some instances, the introduction of biocides to prevent bacterial growth in warm, moist environments could pose additional inhalation or skin exposure risks if not managed properly and/or if proper PPE is not used. In addition, comprehensive and specialized training, clear signage, and well-maintained and utilized safety data sheets (SDS) further support personnel hazard awareness and safe handling practices, while regular inspection of valves, seals, and containment areas helps reduce the risk of leaks and unintentional exposure. Further, engineering controls can complement these procedural safeguards: automated isolation valves, pressure monitoring with trend alarms, and leak detection systems are common in commercial CDUs and OCP-related designs, though not explicitly mandated in current OCP specifications.

Although direct-to-chip cooling is more effective and efficient than traditional air cooling, direct-to-chip cooling has unique concerns that must be addressed. To mitigate these concerns, organizations must implement rigorous system design, regular maintenance protocols, leak detection systems, robust employee training programs, proper chemical management practices, and comprehensive emergency response protocols. Successful implementation requires systematic commissioning—documenting baseline conditions across all components and validating emergency sequences against ASHRAE TC 9.9 guidance. The 2024–25 ASHRAE publications emphasize hydraulic isolation, N+1 redundancy, and commissioning validation, practices equally relevant to facility-level CDUs and rack-level D2C deployments.

Environmental protection strategies should include the use of closed-loop systems, environmentally friendly water treatment options, and proper wastewater and/or hazardous waste disposal methods. Health and safety strategies should include comprehensive job hazard analysis for maintenance and repair activities, strong CoHE/LOTO practices, appropriate use of PPE, and rigorous spill response protocols. Ensuring adherence to EHS best practices will help maximize the benefits of using direct-to-chip cooling while minimizing its potential EHS downsides.

As rack densities rise toward and beyond 100 kW per rack, the industry will benefit from standardized hydraulic interfaces (manifolds, quick-disconnects), unified telemetry protocols, and shared training curricula bridging IT and facilities. Google’s Deschutes program highlights these needs at hyperscale, and OCP’s forthcoming contributions are expected to extend these principles to broader industry deployments.

# # #

About the Authors

Walter Leclerc is an independent consultant. As a published author and global speaker, he has addressed topics including EHS strategy, safety culture, immersion and direct-to-chip cooling, electrical safety, and water stewardship. His mission is to help organizations integrate people, planet, and performance into a unified, future-ready strategy.

Steve Barberi has more than three decades of experience in process cooling and hydraulic design. Adopting a polymathic approach, he combines technical expertise in fluid mechanics and control systems with insights from safety, compliance, and industrial operations. His work focuses on bridging industries — applying lessons learned in regulated environments to the rapidly evolving requirements of AI and high-performance computing data centers.