As AI workloads push rack densities and sustained compute hours skyward, many data centers are turning to liquid cooling to keep chips within thermal limits. Propylene glycol-based coolants are common in secondary and row loop systems for good reason – they widen operating windows and are biostatic at the right concentrations. But when glycol loops are operated at new temperature and duty extremes, they expose a cluster of mechanical, chemical, and biological risks that deserve the same engineering discipline normally reserved for process plants.

Why glycol chemistry matters more at extreme duty

Glycol alters fluid properties in ways that matter to heat transfer and hydraulics. Higher glycol concentrations raise fluid viscosity and lower thermal conductivity compared to pure water; both effects reduce convective heat transfer and can require greater pump energy to maintain flow. Temperature swings and repeated thermal cycling can change inhibitor efficacy and accelerate reactions with wetted materials if chemistries aren’t carefully matched. In short, a loop that starts within spec can drift out of it under sustained, high-temperature operation.

That drift isn’t hypothetical. Field experience from commissioning programs shows that coolant concentration, inhibitor levels, and general chemistry can vary across totes and fills, and that variation becomes consequential when cold-plate channels are on the order of microns. Preventing that drift, and catching it early when it does occur, is the single most important operational control for a data center using a liquid-cooled deployment.

Real operational failure modes to watch for

Three failure modes dominate risk assessments:

• Particulate fouling and channel blockage: Construction debris, welding scale, or improperly cleaned temporary equipment can compromise sub-millimeter cold plates. These narrow passages are unforgiving; even small particles can reduce flow and local cooling capacity.
• Corrosion and inhibitor depletion: If glycol concentration or inhibitor packages are incompatible with pipe and connector metallurgy, corrosion rates can accelerate. In addition, the presence of infiltrants from above and high temperatures can increase corrosion rates. In some circumstances, corrosion products themselves become particulates that further promote fouling.
• Microbial growth and biofilm: Warm, low-flow niches, such as dead legs and stagnant pockets, provide ideal conditions for microbes. Biofilm reduces heat transfer and complicates remediation; it’s also stubborn to remove once established. While glycol-based coolants kept in spec can keep such outcomes at bay, things can go wrong quickly if coolant quality is not well monitored and maintained.

Each of these can translate into elevated delta-T variance, premature component replacement, warranty exposure, or unplanned downtime - all outcomes that are expensive and reputationally damaging to the data center hosts.

Pre-commissioning: the most cost-effective risk control

Mitigating these modes begins long before coolant touches the racks. The Open Compute Project’s pre-commissioning guidance for technology cooling systems codifies a pragmatic sequence: hydrotest, clean, flush, and prepare for commissioning, with documented acceptance criteria for water quality and instrumentation.

That guidance is not academic; it prescribes staged filtration down to a practical final rating (Open Compute’s white paper recommends a ~5-micron target as an industry compromise between capture efficiency and pressure drop) and minimum flushing velocities (roughly 3–5 ft/s) to mobilize sediments and ensure chemical mass transfer during cleaning. These specific steps help reduce day-one surprises and lower the risk of any early rework.

Other practical commissioning controls include: use of “procedure water” meeting documented conductivity and microbiological specs, loop-back plumbing to enable complete flushing, and immediate post-test flushing to avoid prolonged stagnation.

Instrumentation and monitoring: make coolant chemistry visible

The next layer is instrumentation. Basic handheld lab tests are necessary, but in high-duty deployments, they’re not sufficient by themselves. Operators should instrument flushing skids and final loops with calibrated flow meters, filter ΔP gauges, pressure gauges, turbidity (or colorimetric) sensors, conductivity, and pH meters, and sample ports at representative locations. A calibrated refractometer intended for coolant use is essential for verifying glycol concentration both at fill and during operations.

Beyond manual checks, continuous telemetry turns weekly or monthly sampling into real-time visibility: trending pH, conductivity, temperature, flow, and turbidity can reveal gradual drift, transient upset events, or incipient filter blockage. That visibility enables early corrective action – topping glycol concentration, dosing inhibitors, or scheduling a filter change – before hardware is impacted. Correlating this with any needed rework or cabinet maintenance can also reveal when more precautions are required during repairs.

Operational controls and environmental considerations

Good practice extends to the human and regulatory dimensions. Maintain chain-of-custody on coolant totes, review supplier certificates on receipt, and confirm how to top up concentration efficiently (rather than diluting and refilling). For pre-commissioning wastewater, capture and treat flush water where possible; engage local permitting authorities early to avoid discharge delays. If on-site discharge is unavoidable, document the procedure-water characteristics, and biocide use to speed approvals.

On the microbiology front, design to eliminate dead legs, maintain minimum flow velocities during layup, and, if delays occur before commissioning, use a controlled dry-layup and inert gas blanket to minimize contamination risk. Biocide selection should be handled with the site water-treatment partner to ensure effectiveness and compliance with discharge rules.

Beyond mitigation: analytics and repeatability

The greatest long-term gains come from turning monitoring data into repeatable practices. Predictive analytics trained on telemetry and lab data can forecast when inhibitor levels will fall below thresholds or when filters will reach end-of-life, enabling scheduled interventions rather than emergency responses. When organizations standardize instrumentation, sampling locations, and acceptance criteria, they convert a risky novelty into a predictable engineering discipline.

What operators should prioritise now

For teams planning D2C rollouts or upgrades, three priorities deliver outsized value:

1. Treat the loop like a process system: document, instrument, and require sign-offs.
2. Insist on thorough pre-commissioning with staged filtration and the flushing velocities and procedure-water specs referenced in industry guidance.
3. Deploy continuous monitoring for the key chemistry and hydraulic signals, supported by a clear escalation playbook.

When glycol loops are designed, filled, and monitored with these engineering controls, they enable the density and efficiency benefits that liquid cooling promises without turning coolant management into a business risk. The work is discipline-heavy, but the payoff is concrete: fewer surprises, predictable performance, and a pathway to scale AI workloads without compromising uptime or sustainability.