Liquid cooling is a way of removing heat from servers and other IT equipment using a liquid, such as treated water or a specialized fluid, instead of relying on air alone. Liquids absorb and carry heat far more effectively than air, so liquid cooling lets a data center support racks that draw much more power, such as dense GPU servers used for AI training and inference. It ranges from coils on the back of a rack that cool the exhaust air, to cold plates mounted directly on processors, to tanks where equipment is fully submerged.
At a glance
- Uses liquid to carry heat away from equipment, either near the rack, at the chip or by immersion.
- Supports higher rack power density than air cooling typically can.
- Increasingly relevant for GPU and AI hardware, some of which is designed to be liquid-cooled.
- Needs facility piping, heat-rejection capacity and procedures that many data centers do not yet have everywhere.
- Changes the conversation with providers about space, power, maintenance and responsibility.
What problem it solves
Almost all the electricity IT equipment uses becomes heat. Traditional data centers remove that heat with air: cold air goes in the front of the racks, hot air comes out the back, and air handlers and chillers carry it away. That works for the densities most facilities were designed around, but it struggles as racks draw more power. Moving enough air through a very dense rack takes large fans, careful containment and a lot of energy, and at some point it stops being practical.
Accelerated hardware has made this a pressing issue. Racks full of GPU servers can draw several times the power of general-purpose racks. Liquid cooling lets a facility host that hardware without spreading it across many half-empty cabinets, and in some cases it is the only way the hardware can run at full performance.
How it works
Rear-door heat exchangers. A door filled with liquid-cooled coils replaces the back door of a rack. Hot exhaust air passes through the coils and is cooled before re-entering the room. Servers stay air-cooled, which often makes this the least disruptive option for existing equipment; how much heat it can remove depends on the design.
Direct-to-chip cooling. Cold plates sit on the hottest components, such as processors and accelerators, with liquid circulating through them. A coolant distribution unit (CDU) manages the loop and exchanges heat with the facility’s water system. Other components may still need some air cooling, so these are often hybrid deployments. Many high-end GPU servers are offered in direct-to-chip versions.
Immersion cooling. Servers are submerged in a non-conductive (dielectric) fluid in tanks. Heat moves into the fluid, which is cooled by a heat exchanger. Single-phase systems keep the fluid liquid; two-phase systems let it boil and condense. Immersion can support high densities, depending on the design, but changes how equipment is installed, serviced and warrantied.
Facility side. Every approach needs some way to reject heat from the building, and capacity depends on the design rather than a fixed ranking of methods. Systems connected to facility water may need new piping and water-quality controls; self-contained units and dielectric-fluid systems have different heat-rejection, fluid-management and leak-detection requirements. All of them need trained staff and procedures. Liquid can also allow warmer operating temperatures and less mechanical chilling, which may improve power usage effectiveness (PUE), depending on the design.
For help finding facilities that can host dense AI hardware, see our Colocation solution page.
When it matters for buyers
- Planning AI or GPU deployments. Check whether the hardware you plan to buy needs liquid cooling, and whether target facilities support it in the space they are offering. Renting capacity through GPU as a service moves that problem to the provider.
- Choosing a colocation provider. Liquid cooling capability differs widely between facilities and even between halls in one facility.
- Sustainability targets. Liquid cooling may reduce cooling energy and enable heat reuse, depending on the design.
- Hardware warranties and support. Immersion and direct-to-chip setups may affect vendor warranties and who can service the equipment.
Questions to ask vendors
- Which liquid cooling methods do you support, in which halls, and at what maximum density per rack?
- Who provides and maintains the coolant distribution units, piping and connections to our equipment?
- How are leaks detected and contained, and who is liable if our equipment is damaged?
- What are the water quality, temperature and flow specifications, and do they match our hardware’s requirements?
- How is liquid-cooled capacity priced compared with air-cooled space?
- If you do not support liquid cooling today, what is your plan and timeline, and is it in writing?
How it differs from air cooling
Air cooling moves heat with air: fans in the equipment and air handlers in the room, often with hot-aisle or cold-aisle containment. It is simpler, widely supported and suits most general-purpose equipment. Liquid cooling moves heat with liquid, which carries much more heat per unit of volume, so it supports higher rack power density at the cost of new infrastructure, procedures and sometimes equipment changes. Many facilities use both, with liquid cooling for dense racks and air for everything else.