An air-cooled hall is vendor-neutral almost by accident. Racks take air from the room, and whoever made the rack, the server and the cooling plant never had to agree on anything more demanding than a footprint and a power connection.
Liquid cooling removes that accident. A loop is a mechanical system in which every component has to physically mate with the next one and tolerate the same fluid at the same conditions. The result is that choosing a cooling vendor commits a facility to considerably more than a purchase, and the commitments are hard to reverse because they are plumbed in.
Our guide to direct liquid cooling covers how the technology works and our overview of coolant distribution units covers the equipment. This is the procurement question neither addresses: what you are tied to afterwards, and which of those ties are avoidable.
Where the lock-in actually sits
It is worth being specific, because the exposure is not uniform across the system.
The couplings
Quick-disconnect couplings are the joint between a server and the rack manifold, and they are made and sold in incompatible variants. Different suppliers use different geometries, different sealing arrangements and different ratings. Two couplings can be the same nominal size and still not mate, and even where they mate the combination may not be qualified by either manufacturer.
This is the most physical form of lock-in on the list. Once a hall is fitted with manifolds presenting one coupling type, server hardware has to present the matching half or the hardware does not connect. Work is under way across the industry to standardize these interfaces, and it is making real progress, but a facility being built now should confirm the specific parts rather than rely on the direction of travel.
The fluid
This is the one that gets underestimated. The coolant in a direct-to-chip loop is a specified component, usually a water and glycol mixture carrying corrosion inhibitors and a biocide, and its formulation is matched to the metals, elastomers and plastics in the system.
Running a fluid outside the approved list is not a neutral substitution. It can affect warranty coverage, it can interact badly with materials the original fluid was chosen to protect, and the additive package degrades on its own schedule regardless of what you believe about the base fluid. Changing it later is a drain, flush and refill of the entire loop rather than a top-up, which is why the approved list is effectively a long-term commitment made at specification time. Our guide to liquid cooling maintenance covers the sampling and replacement regime that follows from it.
The service relationship
Less visible and frequently more binding. Many systems come with conditions about who may open the loop, who may replace a component, and what documentation is required to keep coverage intact. Those conditions determine whether your own technicians can act at two in the morning or whether a fault waits for a vendor engineer.
For a facility in the Gulf, where the supplier's nearest qualified engineer may not be in the country, the practical difference between a system your team can service and one it cannot is measured in hours of downtime. Our note on liquid cooling in UAE data centers covers the regional considerations that make this sharper here than in markets with denser vendor coverage.
The control layer
CDUs, leak detection and rack-level sensing all produce data, and that data has to reach the systems that already run the facility. Where a vendor exposes an open protocol this is routine. Where the interface is proprietary, the cooling system becomes an island, and integrating it later is a project rather than a configuration change. Our overview of leak detection covers why that particular signal is one you want in your own monitoring rather than only in the vendor's.
The question a colocation tenant has to ask
For anyone placing liquid-cooled hardware in someone else's building, all of the above arrives as a single practical question: will my equipment connect to your hall.
That question has to be answered in specifics. Which coupling type the facility's manifolds present. What fluid the loop runs and whether it is compatible with the cold plates on your servers. What supply temperature and pressure the facility provides, and what your hardware requires. Whether the facility will accept equipment it did not specify at all.
The answers determine whether a deployment is straightforward or impossible, and they are cheap to obtain early. Discovering a coupling mismatch after hardware has shipped to a GPU colocation facility is an expensive way to learn a dimension.
Keeping options open without avoiding the technology
None of this is an argument against liquid cooling, which for high-density AI hardware is increasingly not optional. It is an argument for making the commitments deliberately.
- Treat the system as layers rather than one purchase. Facility loop, CDUs, rack manifolds and server-side hardware are separable decisions. They do not have to come from one supplier, and the interfaces between them are where future freedom lives.
- Write the interfaces into the specification. Coupling type, fluid specification, supply temperature and pressure, and the control protocol. Stated explicitly, these become requirements a second supplier can quote against later.
- Get the approved fluid list in writing before commissioning, together with what happens to coverage if it is not followed.
- Settle who may service the loop, and specifically whether site staff may perform a hot-swap of a failed component without affecting coverage.
- Ask what a second source would involve. Not because you intend to use one, but because the answer reveals how much of the system is genuinely proprietary.
- Keep the as-built documentation. Fluid specification, component part numbers, loop drawings and commissioning data. A retrofit or a vendor change starts by establishing what is actually installed, and our guide to retrofitting liquid cooling covers why that record is worth keeping from day one.
What this is really about
Liquid cooling is at the point in its adoption where the interfaces are still settling. That is normal for an infrastructure technology and it will resolve, but it means decisions made now carry more consequence than the same decisions will in a few years.
The operators who handle this well are not the ones who find a vendor-neutral solution, because for the moment that is not quite available. They are the ones who know precisely which commitments they have made, wrote them down, and can tell you what changing any one of them would cost.
If you are specifying a liquid-cooled deployment, or evaluating whether your hardware will connect to a facility that already has one, get in touch. These are much better questions to resolve on a drawing than on a loading dock.
Frequently Asked Questions
Are liquid cooling components interchangeable between vendors?
Not reliably, and not yet by default. Quick-disconnect couplings, manifold interfaces and cold plate mounting differ between suppliers, and two components that look equivalent may not mate or may not be rated for the same conditions. Industry standardization work is under way and is genuinely improving the position, but a facility being built today should verify compatibility for the specific parts it intends to use rather than assume it.
What is actually locked in: the hardware or the fluid?
Both, and the fluid is the one people underestimate. The coolant is a specified part of the system, and running something outside the approved list can affect warranty, material compatibility and the behavior of corrosion inhibitors and biocides over time. Changing fluid in a filled loop is a flush and refill of the whole system, not a top-up.
Can a colocation tenant bring their own liquid-cooled servers?
It depends entirely on whether their cold plates and couplings interface with the facility's manifolds, and on whether the facility will accept the hydraulic conditions their hardware expects. This is the single most important question to settle before shipping equipment to a liquid-cooled hall, and it is answered in millimeters and pressures rather than in principle.
How do you reduce exposure without avoiding liquid cooling?
Separate the decisions. The facility loop, the CDUs and the in-rack distribution do not all have to come from one supplier, and the boundaries between them are where optionality lives. Specify interfaces explicitly, require the fluid specification and approved list in writing, and establish who may perform service before the system is commissioned.
A Real Energisation Date, Not a Brochure
Rax operates where transmission investment has run ahead of load, and we are direct about staged capacity and real dates. Talk to our team about sizing a deployment against a timeline.
Contact Rax Data & Energy