OAT vs NOAT vs HOAT Inhibitor Chemistry for Data Center Cooling

Specifying the wrong glycol inhibitor package for a liquid-cooled data center is not a configuration error you discover at commissioning — it's one you discover at year two or three, when a cold plate leaks, a heat exchanger pressure-drops 40% without explanation, or a corrosion coupon comes back from the lab looking like it spent a month in acid. The base glycol fluid — whether propylene or ethylene — is the carrier. The inhibitor package is the actual corrosion management system, and the three dominant technology families each use a fundamentally different electrochemical strategy to protect metal surfaces. OAT, NOAT, and HOAT are not interchangeable, they are not upgradeable through top-offs, and they are not forgiving when mixed. For engineers speccing fluid for a new B200 or MI300X deployment, or troubleshooting a cooling loop degrading faster than the manufacturer's change-out recommendation, understanding what these inhibitor families do at the molecular level is not academic — it is the difference between a six-year maintenance cycle and a three-year emergency drain-and-refill.

The Chemistry Behind Each Technology

Organic Acid Technology — OAT — derives its corrosion protection from carboxylate salts, most commonly sebacate, 2-ethylhexanoate, and their derivatives. These are long-chain organic molecules with carboxylic acid head groups that adsorb directly onto metal surfaces through chemisorption, forming a monomolecular passive film. The mechanism is fundamentally different from older inorganic inhibitor chemistries: OAT is not a sacrificial inhibitor that deposits a thick oxide layer on first contact. Instead, it forms a thin, self-repairing molecular film that persists as long as carboxylate concentration remains above the critical threshold — typically measured by reserve alkalinity titration. This sustained-release protection is why properly maintained OAT systems achieve nominal service lives of five to seven years in controlled environments. OAT inhibitor packages carry strong yellow-metal protection credentials. On copper and copper-nickel alloy surfaces, the carboxylate head groups coordinate with cuprous ions at the oxide interface, stabilizing the Cu₂O passive film and preventing the ongoing oxidative dissolution that causes copper ion contamination of the fluid stream. Brass fittings — which contain 15–35% zinc depending on alloy designation — require that the OAT package include specific benzotriazole-class azole inhibitors to address dezincification risk; not all OAT products include adequate azole concentrations, and this is a specification detail worth verifying explicitly.

Nitrited OAT — NOAT — adds sodium nitrite or a nitrite salt to an OAT carboxylate base. The nitrite functions as an anodic inhibitor for ferrous metals, reacting at anodic sites on iron and steel surfaces to form a protective iron-nitrite complex that impedes the oxidative dissolution that drives rust formation. This is the technology of choice for cooling systems that incorporate cast iron pump impeller housings, mild steel headers, or carbon steel piping in the building-level chilled water plant that feeds GPU racks through a CDU. The price of that ferrous protection is service life: nitrite oxidizes over time, and in hot, aerated systems, depletion is measurable within three years. NOAT systems carry a nominal four-to-five-year change-out recommendation, and the nitrite fraction must be specifically tested — it is not captured by a pH or conductivity reading. Nitrite also contributes to conductivity and can participate in reactions with secondary amines if the fluid is ever contaminated with certain corrosion inhibitor chemistries, making cross-contamination control critical.

Hybrid OAT — HOAT — blends OAT carboxylates with inorganic inhibitors, most commonly silicates, phosphates, or both, to extend metallurgical compatibility to aluminum. Silicate is particularly effective for aluminum: it deposits a thin, dense amorphous silica film on aluminum surfaces at pH 7–8, providing rapid passivation that OAT carboxylates alone cannot achieve quickly enough on reactive aluminum alloys. This matters because aluminum cold plates and manifolds used in some liquid-cooled GPU designs (including AMD Instinct reference configurations) will experience aggressive pitting within six to twelve months in an under-inhibited or wrong-chemistry environment. The tradeoff is that silicate concentration must be managed carefully — silicates deplete, they are sensitive to pH swings, and in narrow-channel architectures they present a scaling risk that carboxylate-only fluids do not.

Why Mixing Technologies Destroys System Integrity

The instruction to avoid mixing glycol inhibitor technologies appears in virtually every fluid manufacturer's technical data sheet, yet it remains one of the most common sources of catastrophic heat exchanger fouling in data center liquid cooling installations. The reason is not caution language inserted by a legal department — it is a straightforward consequence of the electrochemical incompatibility between inhibitor classes, and understanding the mechanism is the best prevention against the decision-making that leads to an "emergency top-off with whatever is in stock."

The most consequential mixing failure in data center contexts is silicate-carboxylate incompatibility. When a silicated HOAT fluid is added to a system already containing OAT fluid — or vice versa — the silicate oligomers in the HOAT encounter the lower-pH, carboxylate-stabilized environment of the OAT system. Silicates are stable as alkaline, polymerized anions (silicate chains with Si-O-Si linkages) within their designed pH window of 7.0–9.0. Outside that window, or when silicate concentration is diluted below its stability threshold by blending with a carboxylate-dominant fluid, the silicate anions undergo polycondensation — they cross-link into a three-dimensional silica gel network. This gel is amorphous, nearly transparent, has a specific gravity close to 1.0, and is highly adhesive to metal surfaces.

In practical facility terms, silicate gel formation manifests first as a gradual increase in differential pressure across the CDU-to-rack supply header, followed by reduced flow rate at the rack level and ultimately GPU thermal throttling or emergency shutdown as coolant flow falls below the manufacturer's minimum specification. By the time the pressure drop is obvious, gel has already deposited on heat transfer surfaces, reducing the U-value of the heat exchanger and compounding the thermal problem. The insidious aspect of this failure mode is that it is slow enough to be misattributed to pump degradation, filter fouling, or balancing valve drift — all of which get investigated before someone considers fluid chemistry.

The second mixing hazard is nitrite-amine incompatibility. Many OAT formulations use morpholine or other secondary amines as pH buffering agents. If an NOAT fluid — which contains nitrite — is added to an amine-containing OAT system, nitrite reacts with secondary amines to form N-nitrosamines. These compounds are both a system chemistry problem (they destabilize pH buffering capacity) and a chemical safety concern under OSHA hazard communication standards. A system contaminated in this way requires complete drain and disposal of the mixed fluid as hazardous waste, not simply a flush and recharge.

The operational protocol for an inhibitor technology change is non-negotiable: complete drain of the existing fluid, two or three DI water flushes until conductivity of the purge water falls below 50 µS/cm, a 1% citric acid flush held for 24 hours if silicate gel contamination is suspected, a final DI water rinse, and a full recharge with the new fluid at the correct dilution ratio. Attempting to "dilute out" the old inhibitor by progressive top-off with new fluid does not work — you will achieve a mixed system at an unknown ratio, with unknown inhibitor interactions, for an indefinite period.

Selection Rules by Cold-Plate Metallurgy

The first question in any fluid specification process should not be "what does the CDU manufacturer recommend?" — it should be "what metals does this fluid touch between the CDU outlet and the CDU return?" The answer determines the inhibitor class. Everything else is secondary.

Pure copper cold plates are the dominant architecture in enterprise GPU liquid cooling, and they drive the majority of data center deployments. Copper is an electrochemically noble metal relative to steel and aluminum, which means it is generally more resistant to oxidative dissolution — but it is not passive in the absence of inhibitors. In high-flow-rate systems with dissolved oxygen present, uninhibited copper surfaces will develop a cuprite (Cu₂O) layer that is porous and continues to corrode underneath. OAT carboxylate inhibitors, particularly those containing sebacate and 2-ethylhexanoate, coordinate with cuprous sites at the oxide interface and produce a dense, self-repairing passive film. For copper-only and copper-nickel alloy systems, OAT is the specification of choice. The five-to-seven-year service life under controlled thermal conditions provides the lowest long-term ownership cost per rack.

Aluminum cold plates and manifolds change the equation entirely. Aluminum's native oxide (Al₂O₃) is amphoteric — it dissolves in both acidic and basic conditions — and aluminum is electrochemically active enough to pit rapidly when the passive film is disrupted. OAT carboxylates are not effective aluminum inhibitors at the passivation speed required; aluminum surfaces can develop pitting in under-inhibited OAT systems within six to twelve months. The correct specification is either a silicated HOAT product or a specialized OAT formulation that incorporates aluminum-specific inhibitors such as tolyltriazole derivatives or proprietary aluminum-protective packages. AMD's Instinct MI300X thermal and mechanical design guide specifies a pH range of 7–9 and an inhibitor package explicitly compatible with aluminum alloys — this is pointing to HOAT or aluminum-OAT, not standard OAT.

NVIDIA's GB200 NVL72 liquid-cooled rack architecture uses copper cold plates with a nickel-plated surface on the GPU-contact side. NVIDIA's NVLink liquid cooling documentation references ASHRAE Guideline 21-compatible fluids and specifies an OAT-class inhibitor package. Nickel is compatible with OAT chemistry and does not introduce the aluminum pitting risk. Specifiers should confirm the full material bill for manifolds, fittings, and secondary loop components — nickel-plated copper cold plates paired with aluminum manifold blocks would require HOAT chemistry, not pure OAT, and field installations have made this mistake.

Mixed-metal systems — which describe the majority of real-world data center installations, where copper GPU cold plates connect through stainless or brass manifolds to building-level chillers with cast iron pump housings — are the most common scenario and the most difficult to specify. HOAT is the pragmatic choice for mixed-metal systems because it provides the broadest metallurgical coverage, but it demands tighter monitoring: silicate concentration, reserve alkalinity, and metal ion content must be tracked quarterly rather than annually. The shorter effective service life (three to five years for silicated HOAT versus five to seven for OAT in a copper-only system) should be factored explicitly into the maintenance budget.

Performance Under High GPU Thermal Density

Inhibitor service life estimates from fluid manufacturers are developed under reference conditions: ASTM D1384 corrosion testing at 88°C with a defined metal coupon set, or ASHRAE Guideline 21 conditions assuming 35–45°C supply fluid temperature and moderate heat load. Neither set of conditions reflects what happens inside a cold plate serving a B200 GPU at 700 W TDP or a GB200 module approaching 900 W. The gap between reference conditions and actual high-density deployment is where inhibitor depletion predictions fail, and where facilities teams relying on annual testing cadences find themselves behind the curve.

At 500–600 W/cm² heat flux — the range applicable to NVIDIA B200 and GB200 GPUs with cold-plate contact areas in the 60–70 cm² range — the fluid layer immediately adjacent to the cold-plate wall experiences a temperature 15–25°C above the bulk fluid temperature at typical 20–25 L/min flow rates. This localized thermal boundary layer is where carboxylate oxidation occurs. The chemistry is straightforward: long-chain dicarboxylate molecules (C8–C12 chain length in most OAT products) undergo oxidative cleavage at elevated temperature, breaking into shorter-chain species that are less effective at film formation and more acidic in solution. The reaction rate follows Arrhenius kinetics — for every 10°C increase in temperature, oxidation rate approximately doubles. At a wall temperature of 70°C versus the 45°C reference temperature, the effective oxidation rate at the metal-fluid interface is 4–6 times the reference rate.

An 85 kW GPU rack running 32 × B200 GPUs generates approximately 289,000 BTU/hr through the liquid cooling circuit. At a 15°C delta-T across the rack CDU and 20 L/min per server node, each cold plate pass exposes the fluid to sustained elevated temperatures that compress the effective inhibitor service life from five to seven years down to two to three years — without any change in bulk fluid pH or conductivity.

The flow rate dimension adds a second mechanism: at 20+ L/min through a cold plate with 1–2 mm microchannels, the Reynolds number is high enough to generate turbulent or transitional flow regimes at the channel wall. Turbulent flow increases mass transfer at the metal surface, which means inhibitor molecules are consumed at the passivation sites faster than in laminar-flow reference conditions. The passive film that OAT chemistry builds is maintained by dynamic adsorption-desorption equilibrium — as the film thickness fluctuates under turbulent shear, more inhibitor molecules must be consumed to repair and maintain it. Higher flow rates accelerate this consumption.

For hyperscaler facilities and colocation operators running B200 or GB200 clusters, this has a direct operational implication: the CDU manufacturer's change-out recommendation of five years, published in the product data sheet, was not developed for 85–130 kW rack densities with 20 L/min per node. It was developed for HPC clusters running 10–30 kW racks at 10 L/min flow rates. The thermal and hydraulic conditions of modern AI training infrastructure are categorically different, and the maintenance schedule must reflect that. Quarterly reserve alkalinity testing, which we cover in the next section, is not conservatism — it is the minimum viable monitoring cadence for these thermal densities.

Reserve Alkalinity Testing

pH monitoring is the most commonly deployed real-time parameter in data center liquid cooling loops, and it is also the most misleading single indicator of inhibitor health. A pH reading within specification — typically 7.5–10.5 for OAT and 7.0–9.0 for HOAT — does not mean the fluid has adequate inhibitor capacity. It means the fluid has adequate buffering capacity. These are not the same thing, and the difference between them is the conceptual gap responsible for most preventable corrosion failures in systems that were, on paper, being monitored.

OAT carboxylate inhibitors are weak acid salts. In solution, they participate in the acid-base buffering equilibrium of the fluid, resisting pH change as they are consumed. This is useful for system stability — it means pH does not crash immediately as inhibitors deplete — but it creates a measurement problem. Because the inhibitors themselves are buffering the pH, the pH remains within specification until approximately 50% of the inhibitor reserve has been consumed. At that point, buffering capacity has degraded enough that pH begins to drift, and it can move outside specification quickly. By the time a pH alarm fires, you have already lost half your protection chemistry. In a high-density GPU cooling loop, that transition from 50% depletion to <10% depletion can take less than eight months.

"Reserve alkalinity is the honest number. By the time pH shifts outside the 7–10 band on an OAT system, you've already lost roughly half your inhibitor capacity — often 18 months before that pH reading would have flagged anything. Engineers who rely on pH alone are flying blind from the moment the fluid goes in."

Reserve alkalinity is measured by potentiometric titration: a 20 mL fluid sample is titrated with 0.1N hydrochloric acid to the inflection point at approximately pH 5.5. The volume of HCl consumed in milliliters is the reserve alkalinity value. A fresh OAT product at operating concentration will typically show reserve alkalinity of 8–14 mL depending on inhibitor concentration. At 4–5 mL, the system is approaching the depletion threshold and a change-out should be planned. Below 3 mL, the fluid should be considered at end of useful life regardless of pH.

ASTM D2570, the standard test method for simulated service corrosion testing, provides a more comprehensive assessment by exposing metal coupons from the system's actual metallurgy set to conditioned fluid under controlled temperature and flow. This test takes two weeks to run and produces corrosion rate data in mg/cm²/week — the same units used in the original fluid qualification testing. ASTM D2619, the hydrolytic stability method, evaluates whether the base fluid and inhibitor package remain chemically stable when exposed to water at elevated temperature, which is relevant for systems that experience wide temperature swings during power cycling. Both tests should be included in a full annual laboratory panel, but they complement — not replace — the quarterly reserve alkalinity measurement.

The recommended monitoring framework for high-density GPU liquid cooling loops is tiered. Continuous in-line sensors monitor bulk pH and conductivity, with alerts at defined thresholds. On a quarterly basis, a 100–200 mL grab sample goes to an accredited laboratory for reserve alkalinity titration, ICP-OES analysis of dissolved metal ions (copper, iron, aluminum, zinc), and inhibitor-specific colorimetric or chromatographic assay. Annually, the full ASTM D1384 coupon set runs alongside the quarterly panel. Metal ion concentrations are diagnostic as well as prognostic: copper above 0.5 ppm suggests copper surface oxidation is occurring; aluminum above 0.1 ppm in an aluminum-containing system suggests pitting is underway; iron above 0.5 ppm in a system with ferrous components suggests the NOAT nitrite has depleted below its protective threshold.

Real-World Failure Modes

Copper oxidation at low inhibitor reserve typically presents first as a color shift in the fluid — from clear or pale yellow to blue-green — as Cu²⁺ ions dissolve into solution. Facilities teams often attribute this to contamination rather than corrosion chemistry, and the response is frequently a filter change rather than a fluid analysis. By the time the fluid is visibly colored, copper ion concentration is already high enough to cause secondary problems: copper ions in solution can plate out electrochemically on aluminum surfaces, forming Cu⁰ deposits that create galvanic couples between the copper deposit and the surrounding aluminum. These galvanic pairs drive accelerated aluminum dissolution at the cathodic boundary — pitting that has nothing to do with the aluminum's direct contact with the bulk fluid chemistry and that progresses even after the fluid is refreshed, because the copper deposit is now permanently present on the aluminum surface.

Consider a 42-rack GPU compute cluster commissioned with an OAT product selected correctly for its copper cold plates and stainless steel manifolds. Annual pH testing returned values of 8.1, 7.9, 8.2, and 8.0 over four years — all within specification, all generating no action. In month 47, a pinhole leak developed at a brass compression fitting in the secondary loop return header. Post-failure analysis revealed dezincification of the brass fitting: the zinc had been selectively leached from the Cu-Zn alloy matrix, leaving a porous, copper-rich plug that lacked the mechanical strength of the original material. The failure was not a surprise in retrospect — the OAT product specified did not include adequate azole inhibitor concentration for brass dezincification protection, and reserve alkalinity had fallen below 2.5 mL by month 36. That depletion was never measured. Had quarterly reserve alkalinity testing been in place, the curve would have shown the inflection at month 28–30, allowing a mid-cycle inhibitor top-off or a planned change-out at month 36. The pinhole leak at month 47 required emergency water removal, CDU offline time, and component replacement — none of which would have been necessary with a $150 quarterly laboratory test.

Silicate gel formation presents as a diagnostic puzzle because its earliest indicator is not visible in the fluid. The first measurable signal is an increase in differential pressure across the rack-level heat exchanger or CDU — typically a 15–25% rise over two to four weeks — that does not resolve with filter replacement. Flow rate at the rack manifold drops correspondingly. GPU inlet temperature rises because the CDU is no longer transferring heat at its design rate. The progression is linear and worsening, not episodic. Disassembly of the heat exchanger reveals grey-white gel deposits on the plate surfaces and in the port headers. The gel is insoluble in water and requires a weak alkaline or acid wash to remove. In one documented pattern, a facility that had replaced 40% PG OAT with 40% PG silicated HOAT via progressive top-off — rather than drain-flush-refill — experienced complete blockage of a brazed aluminum plate heat exchanger within nine months, requiring heat exchanger replacement at approximately $8,000 per unit.

Aluminum pitting is the failure mode with the most severe structural consequences. Because pitting is a localized electrochemical process driven by chloride ion attack or inhibitor film breakdown, it does not produce visible discoloration or conductivity changes until the pit has perforated the wall. A 2 mm thick aluminum cold plate wall can be penetrated by pitting in 18–24 months in an under-inhibited system. The geometry of the cold plate assembly means that a pinhole-scale perforation introduces coolant into the GPU module, causing immediate GPU failure and potentially catastrophic electrical damage to adjacent components. Coolant-on-GPU events routinely result in entire server node replacement — not just cold plate replacement — at costs in the range of $40,000–$100,000 per node depending on GPU configuration.

Spec-Sheet Checklist for Procurement

Glycol coolant procurement for liquid-cooled data center infrastructure is not a commodity purchasing decision. The fluid will be in contact with $500,000–$1,000,000 worth of GPU compute per rack, and a specification error will not present itself immediately. The following checklist captures the questions that separate an adequate supplier conversation from a specification you can actually defend to an operations team two years later.

The first and most important question is inhibitor class — not brand name, not product color, not freeze-point chart. Ask the supplier to explicitly state whether their product is OAT, NOAT, HOAT, or a proprietary hybrid. Ask for the specific inhibitor compounds included and their approximate concentrations. If the supplier cannot or will not provide this at the pre-purchase stage, the product is not appropriate for critical infrastructure fluid specification.

Silicate content must be provided in parts per million for HOAT products, not described generically as "silicate-containing." Silicate concentration determines both the level of aluminum protection and the risk of gel formation in narrow channels. Products with silicate concentrations above 300 ppm as SiO₂ require tighter pH management (stay above 7.0) and more frequent monitoring. Ask specifically whether the product contains phosphates in addition to or instead of silicates, as the two behave differently and have different compatibility profiles.

Conductivity ceiling at your operating concentration is a critical parameter for systems with any electrical safety requirements or systems monitored via in-line conductivity sensors. Ask for a conductivity curve across the concentration range (20–50% glycol), not just a single point. HOAT products with high inorganic inhibitor loading can exceed 1,500 µS/cm at 40% concentration — a value that may interfere with leak detection systems designed around lower-conductivity OAT products.

Baseline reserve alkalinity must be provided for a fresh product at your specified concentration. This is your starting point for the depletion curve you will track quarterly. If a supplier cannot provide this value, they are not equipped to support a meaningful testing program. Ask what reserve alkalinity value they recommend as the change-out trigger, and how that value was determined — ideally from coupon corrosion testing rather than field experience alone.

Change-out interval recommendations must be accompanied by the conditions under which the estimate was generated. An OAT product claiming a seven-year service life based on 45°C fluid temperature, 200 L/min building-loop flow rate, and copper-only metallurgy is not giving you a seven-year number for a 130 kW GPU rack at 25°C supply with 20 L/min per node and mixed copper-aluminum metallurgy. Ask for the thermal and hydraulic conditions behind the service life claim, and ask whether the supplier has data from high-density AI computing deployments specifically.

Testing support is the final differentiator between a supplier and a chemistry partner. Ask whether the supplier offers a sampling program with prepaid laboratory kits, what analytes are included in their standard panel, what the turnaround time is for reserve alkalinity and ICP-OES results, and whether they provide a written recommendation with each result or simply return data. In a segment where the most expensive failure modes are chemical in origin, ongoing analytical support is worth weighting heavily in supplier selection.

Every quarter that a reserve alkalinity result comes back above 5 mL is a quarter where no emergency action is required. Every quarter it is not measured is a quarter where the system is operating without an honest assessment of its protection margin. The chemistry that protects a $40,000 GPU cold plate from pitting failure costs less to test per quarter than the synthetic motor oil in a company fleet vehicle. Specify the right inhibitor class for your metallurgy, never mix technologies, and measure what actually matters — not just the parameters that are easiest to instrument.

For deeper treatment of the topics introduced here, the related sub-pillar articles cover direct-to-chip versus immersion fluid type selection, GPU thermal density and coolant flow rate modeling, and CDU-to-rack manifold design for high-density racks. All sub-pillars are organized under the Alliance Chemical AI & Data Center Cooling Chemicals pillar page, which includes the full glycol product catalog with inhibitor class, metallurgical compatibility, and available safety data sheets for each product in the lineup.