Direct-to-Chip Coolant Guide: PG25 vs. Deionized Water, Water Quality and Galvanic Corrosion Control

AI chip power keeps climbing with every generation, and the channels inside direct-to-chip cold plates keep getting finer. Engineers at Motivair, part of Schneider Electric, note that fin spacing in today’s cold plates is around 50 microns, and future designs may go tighter. In passages that narrow, a few particles, a thin biofilm, or a little corrosion debris turn straight into added thermal resistance and pressure drop.

Whether a cold plate still performs to spec in year three often depends less on the plate itself than on the direct-to-chip coolant flowing through it: which fluid the loop runs, how well its chemistry is managed, and whether the mix of metals in the loop has set up galvanic corrosion.

This guide looks at coolant selection and care from a cold plate manufacturer’s point of view. It covers how to choose between PG25 and deionized (DI) water, which water quality parameters to watch, and how to keep galvanic corrosion under control.

What Is Direct-to-Chip Coolant?

Direct-to-chip coolant is the fluid that circulates through the technology cooling system (TCS), carrying heat from cold plates on CPUs and GPUs to the coolant distribution unit (CDU). Single-phase systems mostly use one of two water-based fluids: PG25, an inhibited 25% propylene glycol solution, or treated deionized water. In direct-to-chip cooling, dielectric fluids appear mainly in two-phase systems.

Key takeaways

• PG25 is the most widely used direct-to-chip coolant: biostatic at 25% glycol, freeze-protected to about −10 °C, and easier to maintain than treated water.
• Treated DI water transfers heat better, but untreated DI water is corrosive and needs continuous chemical control.
• Watch conductivity as a trend, not an absolute number, and filter the TCS loop to 25 µm or finer.
• Galvanic corrosion comes from mixed metals in a conductive coolant. Keep aluminum out of copper loops, or use an inhibitor that protects both.

The Two Mainstream Direct-to-Chip Coolants: PG25 and Treated DI Water

PG25: The Default Cold Plate Coolant for Data Centers

PG25 is a mix of 25% propylene glycol and 75% water, and commercial products add corrosion inhibitors and other additives. The 25% figure isn’t arbitrary. At around 10%, propylene glycol can actually feed microbial growth; at 25%, the solution becomes biostatic. It also provides freeze protection down to about −10 °C (14 °F).

Freeze protection matters more for cold plates than it might seem. Cold plates and server liquid cooling modules can see low temperatures in shipping and storage, and if residual water in the loop freezes, it can damage braze joints and fittings.

The ecosystem is also converging on PG25. In August 2026, Castrol announced that two of its PG25 coolants had been validated to NVIDIA’s requirements after a review covering corrosion resistance, thermal stability, dilution water quality, glycol purity, and compatibility with wetted materials. On AI platforms, which PG25 product you use may be a platform requirement in its own right.

Deionized Water: Best Heat Transfer, but Not Harmless

Deionized water contains almost no dissolved ions, and it transfers heat better than any other common coolant. That purity is also the problem: DI water keeps pulling ions from the metal surfaces it touches until it reaches chemical equilibrium. In a loop with copper, stainless steel, aluminum, and polymer seals, uninhibited DI water attacks the most reactive metal first and carries the corrosion products into cold plate channels.

So when a data center says it runs “water,” it means treated water: DI water with corrosion inhibitors and biocide, under continuous monitoring. Nobody should fill a loop with straight DI water.

Why Not Ethylene Glycol, and Where Dielectric Fluids Fit

Ethylene glycol (EG) has somewhat better thermal performance and lower viscosity than propylene glycol, but it’s far more toxic. Many quick disconnects leave a little fluid on their faces when a server is unplugged for service, so data centers stick almost entirely with propylene glycol.

In direct-to-chip cooling, dielectric fluids and low-GWP refrigerants show up mainly in two-phase cold plate systems. Single-phase cold plate loops almost always run water-based coolant. Dielectric fluids are far more common in immersion cooling; for how the two architectures compare, see direct-to-chip cooling vs. immersion cooling.

PG25 vs. Deionized Water: Side by Side

In short, PG25 is the default direct-to-chip coolant for most AI data centers because it resists biological growth, protects against freezing, and needs less maintenance. Treated DI water transfers heat better but only stays safe with continuous chemical treatment and monitoring.

How Much Thermal Performance Does PG25 Cost?

The gap comes down to two fluid properties:

• Lower specific heat. Take Dow’s DOWFROST LC 25 as an example: its specific heat is 3.94 kJ/kg·K at 50 °C, and its viscosity is 2.5 mPa·s at 20 °C. Water’s specific heat at 50 °C is about 4.18 kJ/kg·K, so PG25 carries roughly 6% less heat per kilogram per degree. To carry the same heat load at the same temperature rise, you need more flow.
• Higher viscosity. At more than twice the viscosity of water, PG25 lowers the heat transfer coefficient at the channel walls and raises pressure drop at a given flow rate, which puts more demand on the CDU pump.

OCP’s OAI liquid cooling guidelines show what that looks like at the cold plate. In one set of tests, a cold plate on a 1 kW OAM thermal test vehicle reached a case-to-inlet thermal resistance below 0.02 °C/W with PG25 under ideal conditions, and as low as 0.015–0.016 °C/W with DI water at high flow rates in the same setup.

That has two practical consequences:

1. Cold plate thermal resistance and pressure drop need to be evaluated with the coolant, inlet temperature, and flow rate the system will actually run. APALTEK runs its cold plate thermal tests on PG25 by default so the data reflects the conditions most data centers operate under.
2. If a project is going with PG25, size the cold plate channels and the CDU around PG25’s properties from the start. Don’t design for water and apply a derating factor afterward. For how thermal resistance is defined and compared, see our guide to cold plate thermal resistance.

How to Choose: Four Questions

• Is there any freeze risk? If the cold plates or servers could sit below 0 °C during shipping, storage, or downtime, go with PG25. For exposure colder than about −10 °C, ship the hardware dry or ask your coolant supplier about a higher glycol concentration.
• Has the platform or customer specified a coolant? Many AI platforms and hyperscale customers already have coolant requirements, sometimes down to the product. Follow the platform.
• Does the site have water treatment capability? A DI water program needs ongoing chemical treatment and monitoring. Without a dedicated water treatment team, PG25 is the lower-risk choice.
• How much thermal margin do you have? If the chip has no temperature headroom left on PG25, the extra performance from DI water may be worth it, as long as you accept the heavier maintenance load.

Water Quality: The Parameters That Keep a Direct-to-Chip Loop Healthy

The water quality parameters that matter most in a direct-to-chip loop are pH, the conductivity trend, glycol concentration, inhibitor level, microbiological activity, particle filtration, and dissolved metals.

The TCS loop sits closest to the chips and has the narrowest passages, so its water quality requirements are far stricter than those of the facility water system (FWS). OCP’s modular TCS white paper notes that microchannel heat exchangers call for filtration of 25 µm or finer, and that pipe materials prone to shedding corrosion products, such as carbon steel, are generally avoided.

The table below lists the parameters most direct-to-chip loops monitor. Values are examples only. Always follow the latest limits from your coolant supplier, platform OEM, and OCP or ASHRAE guidance.

Conductivity deserves a closer look. Inhibited PG25 is conductive by design, so judging it against a DI water limit will flag a healthy fluid as out of spec. What matters is how conductivity in the same loop changes over time. Researchers use it the same way: a University of Texas at Arlington thesis on copper cold plates under accelerated degradation tracked changes in pH, oxidation-reduction potential (ORP), and conductivity over time as indicators of corrosion in the loop.

Commissioning: Flush, Fill, and Test Often in the First Week

Commissioning is when a new loop is most likely to go wrong. DI water left over from flushing dilutes the PG25, and glycol concentration can drop noticeably after the first pass through the CDU. Motivair recommends a tiered sampling schedule: test daily, or even several times a day, right after commissioning; move to weekly once readings stabilize, usually after about a week; then step down to monthly.

Another easy-to-miss point: two fluids labeled “PG25” can have completely different additive packages. That’s why OCP recommends running the same fluid throughout the TCS loop. If new equipment arrives pre-filled, confirm the fluid matches the loop’s manufacturer, concentration, and formulation before you connect it.

In Service: Manage by Trend, Not by Failure

Once the loop stabilizes, routine monitoring typically covers pH, glycol concentration, metal ions, microbiological activity, and particles. When a parameter drifts, the usual fixes are to boost inhibitors, adjust pH, add filtration, or dose biocide. Severe contamination calls for a full drain, flush, and refill.

For background on the TCS and FWS loops and where the CDU fits, see What Is a Coolant Distribution Unit (CDU)? and our liquid cooling glossary.

Galvanic Corrosion in Direct-to-Chip Loops: Why Mixed Metals Are Risky

How Galvanic Corrosion Happens

Galvanic corrosion is the accelerated corrosion of one metal when it shares a conductive fluid with a more noble metal. When two metals with different electrochemical potentials sit in the same conductive fluid with an electrical path between them, they effectively form a battery. The more active metal (the anode) corrodes faster, while the more noble metal (the cathode) is protected. Every water-based coolant conducts electricity, so every direct-to-chip loop has the basic conditions for galvanic corrosion.

The classic pairing is copper and aluminum. Aluminum is far more active than copper in the galvanic series, so when the two share a loop, aluminum is the part that corrodes. Even without direct contact, copper ions dissolved in the coolant can plate out on aluminum surfaces, creating local micro-cells that cause pitting.

Aluminum Cold Plates Need a Matched Inhibitor

Aluminum cold plates aren’t off-limits in data centers, but they’re much more demanding about inhibitor chemistry. Many inhibitors designed for copper-based systems, such as traditional HVAC loops, are hard on aluminum. Saint-Gobain’s engineering team warns that an aluminum-containing loop that is under-inhibited or inhibited with the wrong chemistry can build up significant deposits in cold plate microchannels within weeks of commissioning.

If there’s any aluminum in the loop, the inhibitor package has to protect aluminum, copper, stainless steel, and any plating at the same time. Check the coolant against the bill of materials for the whole loop, not just the cold plate.

Other Corrosion Mechanisms to Watch

Galvanic corrosion isn’t the only failure mode. Direct-to-chip loops also need to control:

• Dezincification of brass. Zinc leaches selectively out of the brass, leaving weak, porous copper behind and releasing zinc-rich corrosion products. OCP recommends favoring brass with less than 15% zinc, and verifying any higher-zinc brass as dezincification-resistant per ISO 6509.
• Erosion corrosion. At high velocity, coolant continually scrubs away the protective film on channel walls. Particles, turbulence, and entrained air make it worse. Turning up the flow to chase performance can end up damaging cold plate channels over time.
• Oxygen ingress. In a sealed loop, dissolved oxygen should be consumed gradually. But if hoses are highly permeable or the expansion tank doesn’t seal well, dissolved oxygen stays high, and so does the corrosion rate.

Wetted Materials Checklist

What Coolant Choice Means for Cold Plate Design and Manufacturing

As a cold plate manufacturer, APALTEK treats the coolant as a design input from day one, not as an operations detail for the customer to sort out after delivery. That shows up in four ways.

1. Channel size is matched to filtration and fouling margin. In a microchannel cold plate, narrower fin gaps transfer more heat but are also more sensitive to particles and biofilm. Channel design should match the loop’s filtration rating, typically 25 µm or finer, and leave margin for long-term fouling.

2. Fewer dissimilar metals in the loop. APALTEK’s server CPU and GPU cold plates use vacuum-brazed copper construction. The wetted surfaces are predominantly copper, and there’s no gasket between the cover and base to become a leak path. Vacuum brazing takes place in a vacuum furnace without flux, so there’s no flux residue to end up in the coolant. For how brazing compares with other joining methods, see our overview of the cold plate manufacturing process.

3. Nickel-plated copper cold plates when you need them. If your loop is sensitive to copper leaching, or your platform specifies a surface finish, APALTEK can supply nickel-plated copper cold plates. Plating doesn’t replace the right inhibitor, but it adds a layer of protection for loops that need it.

4. Performance evaluated on real coolant. APALTEK runs cold plate thermal tests on PG25 by default, and test items and pass criteria are defined around each customer’s platform and qualification requirements.

When you send an RFQ to a cold plate supplier, give coolant details specific enough to design to:

• The exact coolant product and concentration, not just “PG25”
• Cold plate inlet temperature and available flow rate
• Allowable pressure drop
• Materials of the other wetted components in the loop (CDU, manifolds, quick disconnects, hoses)

For the full checklist, see how to prepare an RFQ for a custom cold plate.

Direct-to-Chip Coolant FAQ

What is the best coolant for direct-to-chip cooling?

For most single-phase direct-to-chip systems, an inhibited PG25 coolant is the best default: it protects against corrosion, biological growth, and freezing with modest maintenance. Treated DI water is the better choice only when you need every bit of thermal margin and have a water treatment program to support it. If your platform specifies a coolant, follow it.

Can you use tap water or distilled water in a direct-to-chip loop?

No. Tap water carries chloride, calcium and magnesium hardness, and microbes, which lead to pitting, scale, and biofilm. Distilled and deionized water are corrosive on their own and should only be used as treated water, with corrosion inhibitors and biocide. Never fill a loop with untreated water.

Can you mix PG25 and deionized water, or switch from one to the other?

Don’t mix them. Additive packages from different products may not be compatible, and dilution strips PG25 of its biostatic protection. To switch coolant types, fully drain and flush the loop, then confirm the new coolant is compatible with every wetted material.

How often should PG25 be tested, and when should it be topped up or replaced?

Test daily right after commissioning, move to weekly once readings stabilize, and step down to monthly after that. The exact schedule depends on how you balance risk against cost at your site. When pH, glycol concentration, inhibitor level, or metal ions drift out of spec, start by boosting inhibitors or adjusting concentration. Replace the coolant only if it’s badly contaminated or has degraded.

Can aluminum cold plates run on PG25?

Yes, as long as the inhibitor package is compatible with aluminum, and ideally without copper components in the same loop. If copper and aluminum do share a loop, the inhibitor package must protect both metals, and metal ion monitoring should be stepped up.

What happens if direct-to-chip coolant leaks?

Both PG25 and treated water are electrically conductive, so a leak onto a circuit board can cause a short. Single-phase direct-to-chip systems manage that risk with reliable sealing, such as brazed one-piece construction, pressure and leak testing before shipment, and leak detection at the rack and CDU level.

Make Coolant Part of Your Cold Plate Design

Coolant isn’t an operations detail to deal with after the cold plates ship. PG25 or DI water determines which fluid properties the channels are designed around. Water quality management determines how soon those channels start to foul. And the mix of metals in the loop determines whether corrosion shows up in a few weeks or a few years.

If you’re developing liquid cooling for an AI server or GPU platform, send APALTEK’s engineering team your exact coolant product, inlet temperature, and a list of the wetted materials in your loop. We’ll review material compatibility, run a cold plate thermal simulation under your coolant conditions, and recommend a channel structure and surface finish. Click Request a Quote to share your requirements, and we’ll get back to you within 1 business day. You can also browse APALTEK’s liquid cold plates to see the range of custom work we do.