Data Center Cooling Methods Explained (Air, Liquid & Immersion Cooling)

[00:00:00] Speaker 1: In our last video, we followed electrical power from the utility grid all the way to the server rack. Now we need to look at the other half of the equation. Because every watt of power that enters a server becomes heat. Power and cooling are inseparable. If you can't remove the heat, the servers can't operate. Modern data centers are pushing more power into racks than ever before. Traditional enterprise racks once averaged 3 to 5 kilowatts. Today, many racks operate at 10 to 20 kilowatts. AI and GPU clusters can exceed 50, 80, or even 100 kilowatts per rack. And all that heat must be removed instantly, continuously, and reliably. In this video, we're going to break data center cooling into four major methods used in modern facilities. We'll look at how each works, when it makes sense, and why cooling architecture is rapidly evolving. The Core Challenge Rising Heat Density Heat density is the driver behind cooling innovation. When racks were low density, room-level air conditioning systems were sufficient. Large computer room air conditioners could push cold air into the space, and the servers would pull that air through their internal fans. But as rack power increased, air flow requirements increased dramatically. Moving large volumes of air becomes inefficient and physically limiting. Air has relatively low heat capacity compared to liquid. As densities rise, air systems must work harder, move faster, and consume more fan energy. This is why data center cooling has shifted from cooling entire rooms to targeting heat at the source. Now let's walk through the four primary cooling methods used today. Cooling method In this design, large computer room air conditioners, known as crack units, or computer room air handlers, known as CRAW units, deliver cold air into the room. Crack units use direct expansion refrigeration with compressors. CRAW units use chilled water supplied from a central plant. Many facilities use raised floors. The space beneath the floor acts as a supply air plenum. Cold air is pressurized under the floor and delivered through perforated tiles positioned in front of server racks. Server racks are arranged in rows forming cold aisles and hot aisles. Cold air enters the front of the rack, passes through the servers, absorbs heat, and exits the back into the hot aisle. The warm air rises and returns to the cooling unit. Containment strategies improve this design. Cold aisle containment encloses the cold aisle to prevent mixing. Hot aisle containment captures hot exhaust air and routes it directly back to the cooling system. Room-based air cooling works well for lower to moderate rack densities. It is relatively simple and familiar to many mechanical contractors. However, as densities increase, room-based systems begin to reach physical limits. The volume of air required becomes excessive, and airflow management becomes more complex. This leads to the next step in cooling evolution. Cooling method number two: Close coupled air. Cooling. Close coupled cooling brings the cooling source closer to the heat source. Instead of relying entirely on perimeter units serving the entire room, cooling equipment is positioned directly in the row or near the rack. In-row cooling units sit between server racks. They pull hot air from the hot aisle, cool it internally, and discharge cold air directly into the cold aisle. Because the cooling unit is closer to the servers, airflow paths are shorter and mixing is reduced. Rear door heat exchangers are another close-coupled solution. These mount directly on the back of a rack. As hot air exits the rack, it passes through a liquid-cooled coil in the rear door, removing heat before the air re-enters the room. Close coupled systems improve efficiency and support higher rack densities than traditional room-based designs. They also allow different rows to be cooled independently, which is useful when load profiles vary. However, these systems still rely on moving air through servers. At very high densities, even close coupled air begins to struggle. This brings us to liquid-based solutions. Cooling method. Direct-to-chip liquid cooling removes heat directly from the hottest components inside the server, typically CPUs and GPUs. Instead of relying on air to absorb heat from heat sinks, cold plates are mounted directly on the processors. Coolant circulates through these cold plates, absorbing heat and carrying it away through a closed loop. The heated liquid is routed to a cooling distribution unit, or CDU. The CDU contains heat exchangers and pumps. It separates the facility water loop from the server coolant loop, ensuring that water quality and pressures are controlled. From there, heat is rejected to the building's heat rejection system, often a cooling tower or dry cooler. Liquid has far greater heat transfer capacity than air. This allows racks to operate at 50, 80, or even over 100 kilowatts. Direct-to-chip systems are becoming the dominant solution for AI-focused data centers. Because the liquid loop can operate at higher temperatures than traditional chilled water systems, some facilities can eliminate compressors and use economized heat rejection for much of the year, improving efficiency. Direct-to-chip cooling represents a fundamental shift. Instead of cooling rooms, we are cooling components. Cooling Method #4: Immersion Cooling Immersion cooling takes liquid cooling even further. In immersion systems, entire servers are submerged in a dielectric fluid. This fluid is electrically non-conductive and absorbs heat directly from all components. There are two primary types: single-phase and two-phase immersion. In single-phase systems, the fluid absorbs heat and is pumped to a heat exchanger. In two-phase systems, the fluid boils at low temperature, absorbing heat through phase change, then condenses and returns to the tank. Immersion systems can support extremely high densities and eliminate the need for high-volume airflow entirely. However, they require specialized hardware design and operational considerations. While growing in adoption, immersion remains more niche compared to direct-to-chip liquid cooling. Air vs. liquid – when each makes sense. The choice of cooling method depends largely on rack density. For lower-density racks, room-based air systems remain viable. For moderate density, close-coupled air solutions provide improved performance. As densities climb beyond the practical limits of air, liquid cooling becomes necessary. Geography also matters. Energy costs, climate conditions, and water availability influence system design. Redundancy requirements influence equipment selection. Capital costs and long-term operating efficiency must be balanced. There is no single universal solution. Cooling architecture is driven by density, efficiency goals, and risk tolerance. Closing – what comes next? Cooling is evolving rapidly because compute demand is accelerating. As AI workloads grow and power densities increase, cooling systems are becoming more integrated with overall facility design. In our next video, we'll look at redundancy in data center cooling systems. We'll break down what N, N+1, and 2N mean, and how redundancy strategies impact both power and cooling infrastructure. Understanding cooling methods is the foundation. Understanding redundancy is what makes these systems reliable. If you want to see deeper mechanical breakdowns of crack units, cross systems, and chilled water infrastructure, we also have a full HVAC-focused video linked in the description. Power brings energy in. Cooling removes it. Together, they make modern data centers possible. If you found this video helpful, be sure to check out our HVAC and plumbing estimating spreadsheets to streamline your construction bidding process. Check out our HVAC, electrical, and plumbing construction forms to help you run your business and explore our online courses for in-depth training.