In this explainer
- Start at the chip, not the cooling tower
- Two loops can exchange heat without sharing water
- The heat still needs an outdoor destination
- Warmer coolant can change the options
- Fans do not necessarily disappear
- Draw the boundary before comparing projects
- Why it matters
- Related explainers
- Sources and further reading
In this explainer
Following the heat through a liquid-cooled data center
Trace heat from an AI chip through cold plates, coolant loops and heat exchangers to understand what liquid cooling actually changes.
Liquid cooling does not make heat disappear. It gives heat a route out of crowded computing equipment. That route starts at a chip, passes through several pieces of equipment and ends somewhere outside the computing system. Understanding the whole route is more useful than asking whether a building “uses water.”
This explainer follows direct-to-chip cooling: liquid circulates through plates attached to hot components. Immersion cooling, where equipment sits in a suitable liquid, is a different arrangement. Both move heat with liquid, but their plumbing, materials and maintenance procedures are not interchangeable. Neither name tells you by itself how much water a facility consumes.
Start at the chip, not the cooling tower
Electrical energy used by computing equipment becomes heat that must be removed. A cold plate provides a path from a hot component into a circulating liquid. The plate is part of a designed thermal connection; liquid is not simply sprayed onto an ordinary exposed circuit board.
The Department of Energy’s data-center design guidance describes liquid cooling as one of several linked design choices. Equipment temperatures, air management and electrical efficiency all affect the cooling task. The goal is to keep components within their operating requirements while avoiding unnecessary energy use.
Think of the system as a relay. The chip passes heat to a plate, the plate passes it to coolant, and the coolant carries it onward. Improving one handoff helps only if the next handoff can accept the load. A larger pump cannot fix every problem with a poorly designed heat path.
Two loops can exchange heat without sharing water
A coolant distribution unit, usually shortened to CDU, helps manage the equipment-side cooling circuit. In a common arrangement, a heat exchanger transfers heat between that circuit and a facility circuit. The liquids can remain physically separate while heat crosses between them.
Keeping circuits separate allows the equipment loop and the building loop to have different requirements. Coolant quality, pressure and materials compatibility matter close to expensive electronics. The DOE guide describes fluids ranging from treated water to glycol-based solutions. The actual choice depends on the equipment and the designed operating conditions.
A familiar comparison is a radiator: heat crosses a surface without the fluids on each side becoming one mixture. The comparison is about heat transfer, not a complete engineering drawing. A data-center system also needs pumps, valves, monitoring, controls and procedures for abnormal conditions.
The CDU is not automatically the final cooling device. It may hand heat to another circuit that must carry it farther. When a project says it has installed liquid cooling, ask what sits beyond that unit. Otherwise the description stops halfway through the journey.
The heat still needs an outdoor destination
Facility equipment ultimately rejects heat to the environment or transfers useful heat to another user. A dry cooler exchanges heat with outdoor air. Evaporative equipment uses the evaporation of water to help reject heat. Mechanical refrigeration can be part of the design when required temperatures cannot be achieved otherwise.
These choices explain why a sealed equipment loop does not establish zero site water consumption. The liquid close to the chips may circulate repeatedly, while a separate outdoor process consumes water. Conversely, describing the chip loop as water-based does not prove that its contents are continually being used up.
The DOE cooling-water guidance treats energy and water as related design questions. A design may reduce one kind of resource use while changing another. Local weather, required temperatures and the chosen heat-rejection equipment affect the result. There is no single water-use figure that follows automatically from the phrase liquid cooling.
Warmer coolant can change the options
Cooling does not always require making something extremely cold. It requires moving heat from a warmer place to a cooler one at a rate that meets the equipment’s needs. If servers can accept warmer coolant, the outdoor equipment may have more opportunities to reject heat without mechanical refrigeration.
The exact opportunity depends on the climate and the hardware limits. A facility cannot safely raise temperatures just because another project did so. The allowable range belongs to a particular system, including chips, memory and other parts. Operators need measurements and manufacturer requirements, not a general slogan about warm-water cooling.
Likewise, heat reuse is a possibility rather than an automatic benefit. A nearby user must need heat at a useful temperature and at the times the data center provides it. Pipes, heat exchangers and sometimes additional equipment are required. The best place to inspect that claim is the actual heat-use agreement and system design.
Fans do not necessarily disappear
Direct-to-chip cooling can capture heat from selected high-power components while air cooling handles other parts of a server or room. The DOE design guide explicitly discusses arrangements where liquid removes part of the heat and air removes the remainder. “Liquid-cooled” therefore need not mean “no fans.”
This matters for both efficiency and neighborhood expectations. A label about server cooling does not establish the noise of the entire campus. Outdoor fans, electrical equipment and backup systems have their own operating conditions. To evaluate a local proposal, look for the project’s noise study and equipment schedule.
Maintenance also remains important. Fluids must stay within specified conditions, connections must be inspected and controls must respond to faults. Introducing a liquid system is not permission to ignore leaks; it makes detection, isolation and service procedures part of the design. Ask how equipment can be maintained without disrupting neighboring systems.
Commissioning is the process of checking that the installed system behaves as intended. Berkeley Lab’s liquid-cooling commissioning resources emphasize testing rather than assuming that a completed installation automatically performs correctly. The handoffs between IT equipment and building systems are particularly important places to verify responsibilities.
Draw the boundary before comparing projects
A practical reading exercise is to draw three boxes: the server, the building and the outdoor heat-rejection system. Place each piece of equipment described in the proposal inside the appropriate box. If the announcement discusses only the server box, you do not yet have enough information to judge the whole site.
Next, draw arrows for heat rather than assuming the same liquid flows everywhere. One loop may hand heat to another through an exchanger. Label any place where water is evaporated separately from places where liquid simply circulates. This prevents the common mistake of counting circulating flow as continuous water consumption.
Finally, note the conditions attached to performance claims. Was the result measured in operation or projected during design? What weather and computing load did it cover? Were pumps, fans and refrigeration included? These questions do not require a reader to design the plumbing. They establish whether two published claims describe comparable systems and whether the evidence reaches all the way from the chip to the environment.
Why it matters
Liquid cooling helps explain how dense AI equipment can operate in a building, but the public consequences depend on the full system. Water use, electricity demand, noise and opportunities for heat reuse cannot be read from one photograph of a cooling plate. Follow the heat all the way outside before drawing conclusions.
To read a proposal, ask four connected questions. What components are liquid-cooled? Where does the equipment loop transfer its heat? What outdoor equipment rejects it? What measurements support the energy and water claims? These questions separate the engineering description from the environmental outcome.
The existing water explainer focuses on circulation versus consumption. This article’s different angle is the chain of equipment that connects those ideas. Once you can trace that chain, terms such as CDU, cold plate and facility loop stop sounding like separate mysteries and become parts of one heat-moving system.
Sources and further reading
Sources checked Oct 24, 2026. Worked examples and analogies are explanatory, not measurements of a real project.
