Material Compatibility Coolant Sheets Split on Aluminum Use

Lenovo’s Neptune direct water-cooling standard states that aluminum and aluminum alloys are not allowed in the loop, and a 2018 working-group draft for an open liquid-cooled rack puts aluminum on its list of materials to avoid. ASHRAE TC9.9’s 2019 white paper on water-cooled servers lists aluminum in both its facility water system (FWS) and technology cooling system (TCS) columns, with a condition attached. The blanket ban holds for a loop that wets aluminum and copper together and fails as a universal rule. The conditional wording deserves the close reading, because the damage it guards against does not need the two metals to touch.

Three sheets answer the aluminum question three ways

Table 1. Five wetted materials in three sheets (Listing status)

* Aluminum not directly exposed to the same liquid as other metals (coatings or other proprietary methods), or a non-conductive liquid in a TCS

Source: ASHRAE TC9.9, “Water-Cooled Servers: Common Designs, Components, and Processes” (2019); Open Standards Harmonization Working Group, “Open Specification for a Liquid Cooled Server Rack” draft (June 20, 2018); Lenovo Press, “Lenovo Neptune Direct Water-Cooling Standards” LP2018 (September 13, 2024)

ASHRAE’s footnote on aluminum describes two routes to a permitted use. One is surface treatment by coatings or other proprietary methods, so that the aluminum is not directly exposed to the same liquid as other metals. The other is a non-conductive liquid in a TCS. A water or water-glycol loop with copper and aluminum wetted together uses neither route, and that case is the one the other two documents avoid or forbid. The three sheets agree on the bare mixed-metal loop and split only on whether engineered isolation can rescue aluminum.

Elastomers split the same way. ASHRAE lists FEP among TCS materials. Lenovo’s standard does not allow FEP, requires EPDM as the inner lining of hoses, and prefers peroxide-cured hoses because they do not absorb triazoles. ASHRAE’s own hose section warns that hose material may absorb the chemicals that provide anticorrosion protection and strongly recommends a long-term study of the hose with the working fluid. A hose that takes up inhibitor lowers the protection in the loop with no leak to show for it, so a hose that passes a water compatibility check is not yet qualified for treated coolant.

The brass row shows how a shared number hides different scopes. Lenovo and ASHRAE both draw the line at 15% zinc, Lenovo as a ban above it and ASHRAE as a listing below it, while the 2018 draft lists brass with no zinc limit and puts zinc, brazing material included, on its avoid list.

Does galvanic corrosion stop where the metals stop touching?

No. Lenovo’s standard describes galvanic corrosion as two metals far apart in the galvanic series, in electrical contact, immersed in the same water. It then adds that aluminum can be attacked by copper without electrical contact, because copper ions dissolved at very low concentrations deposit on the aluminum surface and form the couple. A paper from the Corrosion ’98 conference, by Northrop Grumman and Corrosion Testing Laboratories, states that this attack needs neither electrical contact nor significant coolant conductivity. In its propylene glycol comparison, the inhibitor suite of a commercial inhibited coolant controlled the attack as well as an inhibited ethylene glycol coolant did, and uninhibited reagent propylene glycol did not. The OSTI record carries only the abstract, so the test matrix behind that comparison is not visible.

A dielectric fitting between an aluminum plate and a copper manifold removes the contact path and leaves the ion path open. On the evidence of the 1998 abstract, the control for the ion path is the inhibitor package, which belongs to the fluid and not to the fitting.

ASHRAE’s white paper adds temperature. It notes that building cooling loops may run at 5 °C to 15 °C, and that materials with a record there may not work at the warmer fluid temperatures of a TCS loop. It asks that compatibility be examined at the warmer temperatures.

Its water-quality table makes a related point. TCS limits sit tighter than FWS limits by a factor of exactly ten for sulfide, sulfate, chloride, total hardness and residue after evaporation, with chloride below 5 ppm in a TCS against below 50 ppm in an FWS. The paper says both tables have been misapplied, citing a building-level loop held to the TCS table by mistake, and that holding the IT side to the FWS specification puts the equipment at significant risk.

Mass loss over 14 days averages away pits that run 14 to 18 times deeper

A study in Materials by Miao, Shao and Zheng put 6061-T6 aluminum, alone and coupled to 304 stainless steel, into 25 vol% propylene glycol in deionized water for 14 days at 80 ± 0.5 °C. The setup was static and naturally aerated. The fluid is described as deionized water and analytical-grade propylene glycol with HCl or NaOH for pH, with no inhibitor package, and the authors flag that the acid and base add chloride and sodium.

At pH 4.8, aluminum alone lost 2.407 g/m², and coupling to stainless steel added 0.968 g/m², or 40%.

At pH 6.8 the loss was the lowest of the three settings, yet coupling added 0.864 g/m² on a base of 1.757 g/m², or 49%. Neutral pH reduced the base loss, yet the coupling penalty did not shrink in proportion: 49% against 40%.

Thickness lost equals mass per area divided by density. Aluminum density is 2.70 g/cm³ at 295 K in the NIST NCNR reference table, which is 2.70 × 10⁶ g/m³. The uncoupled specimen at pH 4.8 gives 2.407 g/m² ÷ 2.70 × 10⁶ g/m³ = 0.891 µm, and the coupled specimen gives 3.375 g/m² ÷ 2.70 × 10⁶ g/m³ = 1.250 µm. The paper’s 3D confocal scans at pH 4.8 report representative pit depths of 12.35 µm uncoupled and 22.63 µm coupled. The ratios are 12.35 ÷ 0.891 = 13.9 and 22.63 ÷ 1.250 = 18.1.

That ratio treats the mass loss as an even thinning of the whole surface. The pit depths come from representative pits in selected regions and not from the deepest pit, so the ratio is a derived illustration for the acid case, not a measured quantity. It still shows what a mass-loss rating cannot: a loop leaks at its deepest pit, and a coupon that passes on g/m² can sit beside a pit that does not.

Faraday’s law converts the measured galvanic current to the same units. The penetration rate is i × M ÷ (n × F × ρ), the relation Metrohm gives in its application note on electrochemical corrosion studies. For aluminum, M = 26.97 g/mol, n = 3 (assuming all current oxidizes Al to Al³⁺), F = 96,485.33 C/mol and ρ = 2.70 g/cm³. At i = 1 µA/cm² = 1 × 10⁻⁶ A/cm², the rate is 1 × 10⁻⁶ × 26.97 ÷ (3 × 96,485.33 × 2.70) = 3.45 × 10⁻¹¹ cm/s. Multiplied by 3.1536 × 10⁷ s/yr that is 1.088 × 10⁻³ cm/yr, or 10.9 µm/yr per µA/cm².

The study measured a galvanic current density of 3.49 µA/cm² at pH 4.8, which converts to 3.49 × 10.88 = 38.0 µm/yr. The coupled mass loss annualizes to 1.250 µm × 365 ÷ 14 = 32.6 µm/yr, 14% below the current-based figure. The two figures land within 14% of each other, but they count different things, total coupled mass loss over 14 days against galvanic current over 20 hours, so the match shows the conversion is of the right order of magnitude and validates neither number.

The current was logged over 20 hours on 1 cm² electrodes at a 1:1 area ratio, and a year-long extrapolation from either number is not a life prediction.

The test ran at 80 °C to simulate data center coolant, above the 50 °C to 60 °C that ASHRAE gives for W2 through W4 TCS loops, so the absolute rates overstate a loop at those temperatures.

Lenovo’s standard separately puts the typical mean time to failure of pitting leaks in copper tube at about 2 years and states no water chemistry or temperature behind that figure.

A materials list names single materials, and corrosion happens between them

Table 2. Verdict by case (Reading of the three sheets)

* W2 through W4 designs, 50 °C to 60 °C per ASHRAE

Source: ASHRAE TC9.9, “Water-Cooled Servers” (2019); Open Standards Harmonization Working Group draft (June 20, 2018); Lenovo Press LP2018 (September 13, 2024); Corrosion ’98 abstract, OSTI record 350035

Each row turns on a pair of metals, a temperature or an inhibitor package, and none of those fits in a column of permitted materials. A design that relies on one of the exceptions, whether a coating, an isolated aluminum part or an elastomer that one sheet lists and another forbids, needs the supplier’s wetted-materials list for the whole loop, which ASHRAE tells designers to require. It also needs qualified engineering review of the actual fluid, temperatures and materials.

References

• Water-Cooled Servers: Common Designs, Components, and Processes (white paper) / ASHRAE Technical Committee 9.9 / 2019 (errata dated 10/10/2019) / https://www.ashrae.org/file%20library/technical%20resources/bookstore/whitepaper_tc099-watercooledservers.pdf
• Open Specification for a Liquid Cooled Server Rack, Progress Update (draft) / Open Standards Harmonization Working Group, hosted by Lawrence Berkeley National Laboratory / June 20, 2018 / https://datacenters.lbl.gov/sites/default/files/OpenSpecification.pdf
• Lenovo Neptune Direct Water-Cooling Standards, Planning / Implementation (LP2018) / Lenovo Press / created or updated September 13, 2024 / https://lenovopress.lenovo.com/lp2018.pdf
• Inhibition of heavy metal ion corrosion on aluminum in fresh water cooling systems using propylene glycol anti-freeze (conference abstract) / Northrop Grumman Corp. and Corrosion Testing Laboratories, Corrosion ’98, San Diego, March 22-27, 1998 / OSTI / https://www.osti.gov/biblio/350035
• Effect of Propylene Glycol Coolant pH on the Galvanic Corrosion Behavior of 6061 Aluminum Alloy/304 Stainless Steel / Miao H., Shao C., Zheng J., Materials 19(13), 2898 / July 6, 2026 / https://pmc.ncbi.nlm.nih.gov/articles/PMC13363374/
• Electrochemical Corrosion Studies of Various Metals (application note AN-COR-010) / Metrohm Autolab / September 2019 / https://www.metrohm.com/content/dam/metrohm/shared/documents/application-notes/an-c/AN-COR-010.pdf
• Reference Tables, physical properties of selected metals at 295 K / NIST Center for Neutron Research / undated / https://www.nist.gov/ncnr/sample-environment/sample-mounting/reference-tables
• Faraday constant, 2022 CODATA recommended values / NIST / 2022 / https://physics.nist.gov/cgi-bin/cuu/Value?f