An AI infrastructure operator places an order for a liquid-cooled GPU cluster: several million dollars of accelerators, the racks, the cooling distribution units, and the fluid that makes the whole thing possible. The hardware is the expensive part and the part everyone focuses on. It is also the part that clears customs without much drama, because servers and networking gear are dry electronics with well-understood classifications. The part that gets held, or in a worse case cannot legally be used in the country at all, is the cheapest line on the purchase order. The coolant.
A liquid-cooled data centre deployment is not one import, it is two very different imports wearing the same purchase order. The servers are an IT hardware import. The coolant is a chemical import, subject to safety data sheet requirements, dangerous goods rules in some cases, and, for the highest-performing immersion fluids, a fast-moving set of PFAS restrictions that can make a fluid legal to buy today and restricted to place on the market tomorrow. The company that treats the cooling fluid as an accessory to the hardware is the company that finds this out at the border, where a customs hold on one line can strand an entire deployment.
This is a genuinely new problem for most technology importers, because until very recently data centres were air-cooled and the only fluid on site was in the chillers. Cooling systems now sit alongside the transformers, switchgear and prefabricated modules on the critical path of a data centre build. The move to liquid cooling, forced by AI hardware that runs too hot for air to handle, has quietly turned a hardware procurement into a chemical procurement, and the two are governed by completely different rules at the border.
Why AI Hardware Forced the Move to Liquid
The reason liquid cooling has gone from a niche to the default in high-density deployments is thermodynamic, not fashionable. Air has a low heat capacity. Past a certain rack power density, roughly the point where a rack draws tens of kilowatts, moving enough air to carry the heat away becomes physically impractical: the fans required are too large, too loud, and consume too much of the power budget to justify. The latest generation of AI accelerators pushes rack densities well beyond that threshold, and air simply cannot keep the silicon inside its thermal envelope. This is the same wave of data centre and AI hardware imports reshaping infrastructure procurement worldwide, driven by the AI server trade flows now running through a handful of manufacturing hubs.
Liquid carries heat far more effectively, and there are two dominant ways to use it. In direct-to-chip cooling, a coolant circulates through cold plates mounted directly on the processors, the GPUs and accelerators that generate the heat, pulling it through a metal plate without ever touching the electronics. In immersion cooling, entire servers are submerged in a dielectric fluid that does not conduct electricity, and heat passes straight from the components into the fluid. Both work. They differ enormously in how hard they are to import, and the reason is the fluid.
Importing Liquid Cooling for Data Centers: The Fluid Is the Regulated Part
Here is the distinction that decides how complicated your import becomes, and it maps directly onto the two cooling methods.
Direct-to-chip systems overwhelmingly use a water and glycol mixture, typically around a 25% propylene glycol blend. This is, chemically, a benign and familiar fluid. Anyone who has maintained a glycol loop in a building HVAC system or a chiller has handled it. It is not a PFAS substance, it is not generally a dangerous good for transport, and it moves across borders as an ordinary industrial fluid with a standard safety data sheet. If your deployment is direct-to-chip, the coolant is the easy part, and the harder compliance work sits with the servers themselves, covered in our guide to shipping AI servers internationally.
Immersion cooling is where the difficulty concentrates, and it splits again into two sub-types. Single-phase immersion uses synthetic hydrocarbon or highly refined mineral oils, which are PFAS-free, biodegradable, and relatively straightforward to move, though still a fluid with handling and documentation requirements. Two-phase immersion, the highest-performing cooling technology available and the one capable of the most extreme rack densities, has historically relied on a specific class of engineered fluorochemical fluids. Those fluids are PFAS. And PFAS is where the border problem lives.
| Cooling method | Typical fluid | PFAS exposure | Import complexity |
|---|---|---|---|
| Direct-to-chip | Water and propylene glycol blend | None | Low. Benign industrial fluid, standard SDS |
| Single-phase immersion | Synthetic hydrocarbon or mineral oil | None. PFAS-free | Moderate. A fluid with handling and documentation needs |
| Two-phase immersion | Engineered fluorochemical fluid | High. These fluids are PFAS | High. Supply shift plus market-by-market PFAS rules |
The pattern the table shows is the core of the procurement decision: cooling performance and import difficulty rise together. The method that handles the most heat is also the one sitting in the middle of the regulatory shift.
The PFAS Problem That Reshaped the Whole Market
In December 2022, 3M announced it would exit all PFAS manufacturing by the end of 2025. That single corporate decision, driven by a multi-billion-dollar legal settlement over PFAS contamination of drinking water, removed the supply chain for the fluorochemical fluids that made two-phase immersion cooling work. The specific engineered fluids that the technology depended on were wound down, with the final ordering window closing in 2025 and manufacturing lines shutting by the end of that year.
The effect on procurement is twofold, and both parts matter at the border. First, the established two-phase fluids are being replaced by a new generation of alternatives from other chemical manufacturers, which means the fluid in a system specified this year may be a recently commercialised chemistry with its own classification, its own safety data sheet, and its own regulatory questions still settling. Second, and more importantly, the regulatory pressure that drove the exit is not a single ban with a single date. It is a moving front, and it looks different in every market you might import into.
In the European Union, a universal PFAS restriction proposal has been working through the REACH process since 2023, covering thousands of substances. The proposal has been through revision, and the current version narrowed its scope and expanded the list of specific derogations, with some heat-transfer applications receiving long transition periods. What this means in practice for an importer is that the status of a given immersion fluid in the EU is not a fixed yes or no. It depends on the exact substance, the exact application, and where the restriction process has reached, and it is genuinely subject to change during the planning horizon of a large data centre build.
In the United States, there is no single federal PFAS import ban for these fluids, but there is a thickening layer of obligations: federal reporting requirements for PFAS substances, proposed rules that would classify certain PFAS compounds as hazardous under environmental law, and a growing patchwork of state regulations, some with manufacturer reporting deadlines and future prohibitions on intentionally added PFAS. The federal position and the position in the specific state where a facility is being built are not the same question, and both have to be answered.
Deploying liquid-cooled infrastructure across more than one country? The coolant that is straightforward to import into one market may be restricted or reportable in another, and the hardware and the fluid clear under completely different rules. Carra Globe acts as importer of record for data centre and AI infrastructure across 175+ countries, and handles the hardware import and the chemical import as one coordinated compliance workflow.
What Changes When Your Import Becomes a Chemical Import
Set the PFAS question aside for a moment, because even a perfectly benign coolant changes the shape of the import the moment it is a fluid rather than a box of electronics. An IT hardware importer moving into liquid cooling for the first time encounters a set of requirements it has never had to handle.
Every coolant needs a safety data sheet, and that document has to be correct, current, and available. The fluid needs to be classified for transport, and depending on the specific chemistry and its flash point, that classification determines whether it moves as an ordinary fluid or as regulated dangerous goods, with the packaging, labelling, and documentation that dangerous goods require, in the same way lithium batteries carry handling rules that dry hardware does not. The volumes involved in filling an immersion deployment are not trivial, which affects how the fluid can be shipped and held in bonded warehouse storage. And the coolant has its own tariff classification, entirely separate from the servers, sitting in the chemical chapters of the tariff schedule rather than alongside the IT hardware. Getting that classification right across a mixed shipment is exactly the kind of line-by-line work a data centre import demands. A customs entry for a liquid-cooled cluster is really two entries in one: the servers under their headings, the fluid under its own.
None of this is insurmountable. All of it is unfamiliar to a team whose entire experience is importing dry hardware, and unfamiliarity at a border is how timelines slip, on top of the certification and export-control gates the GPU hardware already has to pass. The classification of the fluid, the transport rules, and the destination-market chemical regulations all need to be settled before the shipment is booked, in exactly the way the hardware specification is settled before the order is placed.
The Decision Is a Procurement Decision, Made Early
The reason this matters at the planning stage, rather than at the shipping stage, is that the cooling architecture and the fluid are chosen when the facility is designed, and by the time hardware is shipping the choice is locked. It is one more item on the critical path of a data centre construction programme where equipment arrives in waves over months. If the highest-performing option, two-phase immersion, is the one with the most exposed regulatory position in your destination market, that is a fact you want to know while the design is still on paper, not after the tanks are installed and the fluid is held at the port or barred from the market.
The practical sequence for a cross-border liquid-cooled build is straightforward, and it front-loads the fluid question rather than treating it as a shipping detail.
- Identify the exact fluid, by substance, not by category. “Immersion fluid” is not an answer a regulator accepts. The specific chemistry determines the safety data sheet, the transport classification, and the PFAS status. Get the identity of the fluid from the cooling vendor before anything else.
- Check that fluid against the destination market, not a generic rule. A fluid that ships freely into one country may be reportable or restricted in another. For multi-country deployments, this is a per-market check, and the answer can differ across the same rollout.
- Classify the fluid for transport and for tariff separately from the hardware. Establish whether it moves as dangerous goods, and confirm its tariff heading in the chemical chapters. This is what tells you the true cost and the true lead time of the coolant, which the hardware quote does not include.
- Confirm the regulatory position holds for your whole timeline. Because the PFAS front is moving, a fluid that is compliant at design is worth re-confirming before shipment on a long build. This is the check that prevents a specification made in good faith from becoming a problem eighteen months later, and it is core trade compliance work rather than a logistics afterthought.
Frequently Asked Questions
Is importing data centre cooling fluid harder than importing the servers?
Often yes. Servers are dry electronics with well-understood classifications. Cooling fluid is a chemical: it needs a safety data sheet, a transport classification that may make it dangerous goods, and sometimes PFAS compliance that varies by market.
The hardware usually clears more predictably than the fluid, which is the opposite of what most teams expect.
Are all immersion cooling fluids affected by PFAS restrictions?
No. Single-phase immersion uses synthetic hydrocarbon or mineral oils, which are PFAS-free. The PFAS issue centres on the fluorochemical fluids used in two-phase immersion, the highest-performing option, whose supply and regulatory status have both shifted.
Direct-to-chip cooling, which uses a water-glycol mix, is not a PFAS concern at all.
Can a coolant be legal to buy but restricted to use in my country?
Yes, and this is the trap. A fluid can be available from a supplier while being restricted or reportable in the market where you are building. Availability and market-access compliance are different questions, and the second stops shipments.
This is why the fluid has to be checked against the destination market specifically, not assumed from its availability.
Does the coolant need its own customs classification?
Yes. The fluid is classified in the chemical chapters of the tariff schedule, separately from the servers in the IT hardware chapters. A liquid-cooled deployment is effectively two imports on one purchase order, each with its own requirements.
Treating the coolant as an accessory to the hardware is the most common way its requirements get missed.
The hardware is what the budget is built around and what the engineering conversation focuses on, and it is genuinely the easier part to bring across a border. The fluid that keeps it alive is a chemical, and for the highest-performing cooling it is a chemical sitting in the middle of a regulatory shift that is still moving. The operators who deploy liquid-cooled infrastructure across borders without a surprise are the ones who treated the coolant as its own import from the start: identified the exact substance, checked it market by market, classified it in its own right, and confirmed the position held for the length of the build. The ones who treated it as an accessory to the servers discover, at the port or in the destination market, that the cheapest line on the order was the one that needed the most attention.
Disclaimer: This article is for informational purposes only and does not constitute legal, customs, regulatory, or dangerous goods advice. The classification and regulatory status of cooling fluids vary by exact substance, application, jurisdiction, and date, and the PFAS regulatory landscape is changing. Always confirm the current transport classification and market-access requirements for your specific fluid and destination with the relevant authority, the fluid manufacturer’s safety data sheet, or qualified counsel before importing.