Try Free

How Data Center Cooling Works — CRAC, CRAH, Hot & Cold Aisles Explained

NexLayer Tech November 18, 2026 22m 2,695 words
▶ Watch original video

About this transcript: This is a full AI-generated transcript of How Data Center Cooling Works — CRAC, CRAH, Hot & Cold Aisles Explained from NexLayer Tech, published November 18, 2026. The transcript contains 2,695 words with timestamps and was generated using Whisper AI.

"When we think about a data center, we usually think about servers, switches, storage, and maybe power. But one thing is happening all the time in the background. All of this equipment is producing heat. And that heat can become a serious problem very quickly. Servers are designed to operate within..."

[00:00:00] Speaker 1: When we think about a data center, we usually think about servers, switches, storage, and maybe power. But one thing is happening all the time in the background. All of this equipment is producing heat. And that heat can become a serious problem very quickly. Servers are designed to operate within a certain temperature range. If the environment becomes too hot, performance can drop. Hardware can become unstable. And eventually, equipment may shut itself down to protect its components. So, cooling is not just about making the room comfortable. Cooling is actually part of keeping the data center available. Without proper cooling, even a perfectly designed network and power system cannot keep the application running for long. So, where does all this heat actually come from? Almost every device inside the data center consumes electrical power. CPUs, GPUs, memory, disks, network switches, and power supplies all use electricity to perform work. But not all of that electrical energy becomes useful computing. A large part of it eventually turns into heat. Think about a high-performance CPU or GPU. When it is heavily loaded, it can generate a significant amount of heat in a very small area. Now, multiply that by dozens of servers inside one rack. And then hundreds of thousands of racks across a data center. That is why heat density becomes such an important challenge. The more powerful our equipment becomes, the more important cooling becomes as well. The cooling process is actually quite simple when we look at the big picture. First, cold air is delivered to the front of the server racks. The servers pull that cold air into the equipment using internal fans. As the air moves across components like CPUs, memory, and power supplies, it absorbs heat. The air then leaves the rear of the rack as warm or hot air. That hot air needs to be collected and sent back toward the cooling equipment. The cooling system removes the heat from that air and then sends cooled air back toward the servers. And then the cycle repeats. So, you can think of data center cooling as a continuous loop. Cold air in, heat absorbed, hot air out, heat removed, cold air back again. One of the most important ideas in data center cooling is the hot aisle and cold aisle design. The racks are normally arranged so that the front of two rows face each other. This creates the cold aisle. Cold air is supplied into this area and the servers pull that air through their front panels. On the other side, the backs of two rows face each other. This creates the hot aisles. This is where the heated exhaust air from the server is released. The important idea is separation. We want cold air to stay on the intake side and hot air to stay on the exhaust side. If those two air streams mix together, cooling becomes much less efficient. Now, let's look inside the rack itself. Most servers are designed for front-to-back airflow. Cool air enters from the front of the server. Internal fans pull it across the components. And then, the heated air exists from the rear. The same idea applies to network switches. But there, we need to be careful because now every switch has the same airflow direction. Some network equipment uses front-to-back airflow. Other equipment may use back-to-front airflow. If the airflow direction is installed incorrectly, hot air from one device could be pushed directly into the intake of another device. This can create a hot spot very quickly. Cable management is also important. If the rear of the rack is full of badly organized cables, airflow can become restricted. So, good rack cooling depends not only on the cooling system, but also on how the equipment and cables are installed. One traditional cooling design is the raised floor system. In this design, the actual data center floor is raised above the structural floor. The space underneath becomes an air plenum. Cooling units push cold air into this underflow space. The cold air then travels underneath the racks. In the cold aisles, perforated floor tiles allow that air to rise into the room. The servers pull the cold air into their front panels. The hot air exists from the rear and returns through the room toward the cooling units. What advantage of this design is that the floor itself becomes part of the air distribution system. But airflow needs to be carefully managed. If perforated tiles are placed in the wrong location, or there are too many openings, the air pressure under the floor can become unbalanced. Not every data center uses a raised floor. In many facilities, cooling air can also be distributed from above. With overhead air distribution, ducts or sealing systems deliver cool air directly to the cold aisles. The servers still pull the air through the racks in the same way. Hot air leaves the rear of the racks and is returned through another overhead path or through the room. The basic cooling principle does not change. The only difference is how the air is delivered and returned. Overheat systems can make the floor layout simpler because the space underneath the racks is not required for air distribution. You will often hear the term CRAC inside a data center. CRAC stands for Computer Room Air Conditioner. You can think of it as a specialized air conditioning system designed for IT environments. The CRAC takes warm return air from the data center. Inside the unit, the air is cooled using a refrigeration system. Then the cooled air is supplied back to the data center. A CRAC unit may also help control humidity. This is important because data centers need much more precise environmental control than a normal office. So, while a normal air conditioner focuses mainly on human comfort, a CRAC unit focuses on protecting IT equipment and maintaining stable operating conditions. A CRAH looks similar to a CRAC, but the way it removes heat is different. CRAH means Computer Room Air Handler. Instead of having its own refrigeration compressor, a CRAH normally uses chilled water provided by a central cooling system. Warm air from the data center enters the CRAH. Fans move that air across a chilled water cooling coil. The cold water inside the coil absorbs the heat. Then the CRAH sends the cooled air back into the data center. So, the main difference is simple. A CRAH usually creates cooling locally using refrigeration. A CRAH uses cold water that is produced somewhere else in the facility. This makes CRAH systems common in larger data centers. Now, let's follow the chilled water cooling process. The main component is the chiller. The chiller removes heat from the water and produces chilled water. Pumps move that cold water through insulated pipes toward the CRAH unit. Inside the CRAH, warm air from the data center passes over the chilled water coil. The water absorbs the heat from the air. The cooled air goes back into the data center. But now, the water is warmer. That warmer water travels back to the chiller. The chiller removes the heat again. And depending on the design, that heat is eventually rejected outside through equipment such as cooling towers or dry coolers. So, there are actually two connected loops. An air loop inside the data center and a water loop moving heat away from the room. Hot aisle containment improves the basic hot aisle and cold aisle design. Instead of allowing hot air to spread around the room, we physically enclose the hot aisle. The rear sides of the server racks face this contained area. All the hot exhaust air is trapped inside. That hot air is then directed back toward the cooling system. Meanwhile, the rest of the room remains relatively cool. The big advantage is that hot air and cold air are much less likely to mix. The cooling equipment receives warmer returned air. Which can also help the cooling system operate more efficiently. So, hot aisle containment is basically about controlling where the hot air is allowed to go. Cold aisle containment uses the same idea, but in the opposite way. Instead of enclosing the hot aisle, we enclose the cold aisle. Cold air is delivered directly into the contained area. The front of the servers face this cold space and receive a consistent supply of cool air. After the air passes through the servers, the hot exhaust enters the surrounding room. The main goal is still the same. Keep hot air and cool air separated. Whether we contain the hot aisle or the cold aisle depends on the facility design, cooling system, fire safety requirements, and operating strategy. A hot spot is an area where the temperature becomes much higher than the rest of the data center. Hot spots can happen for many reasons. One common reason is blocked air flow. For example, cables or equipment can stop hot air from leaving the rack properly. Missing blanking panels can also create problems. If there are empty rack spaces without blanking panels, hot exhaust air can move back toward the front of the rack. This causes hot air recirculation. High density servers can also create local hot spots because some racks may generate much more heat than others. The best way to prevent hot spot is good air flow design. That includes proper rack orientation, containment, blanking panels, cable management, equipment placement, and continuous monitoring. Cooling is only useful if we know whether it is actually working. That is why temperature monitoring is so important. We normally care most about the temperature of the air entering the servers. That is the real environment the equipment is experiencing. Sensors can be installed at the front of the rack at different heights. For example, one sensor near the bottom, one in the middle, and one near the top. This helps identify temperature differences within the rack. We can also monitor the hot air exhaust temperature at the rear. Monitoring software can then show trends over time. Instead of waiting until equipment overheats, operators can receive alarms when the temperature begins moving outside the expected range. This gives the team time to react before the problem becomes an outage. Temperature is not the only environmental condition we need to control. Humidity also matters. If the air becomes too dry, the risk of static electricity increases. Static discharge can damage sensitive electronic components, but too much humidity is also dangerous. High humidity can increase the risk of condensation or corrosion. So, data centers try to maintain humidity within a controlled operating range. The important point is stability. We do not want large changes in temperature or humidity. A well-designed cooling system helps control both. Cooling system can fail just like servers, switches, or power equipment. That is why redundancy is important. Imagine that the data center needs three cooling units to handle the full heat load. If we install exactly three units, that is an N design. If one fails, we may no longer have enough cooling capacity. With N+1, we add one extra cooling unit. Now, one unit can fail or be taken offline for maintenance, while the remaining units continue supporting the data center. Larger facilities may also use 2N or other highly redundant designs. And redundancy does not only apply to CRAC or CRAH units. It can also apply to chillers, pumps, cooling towers, fans, power feeds, and control systems. The goal is simple: no single cooling failure should immediately cause the data center to overheat. Cooling failure can become serious very quickly. At first, the room may look completely normal, but the servers continue producing heat every second. If the cooling equipment stops removing that heat, the temperature begins to rise. The internal fans in servers and switches may increase their speed. Then we may start seeing temperature alarms. If temperatures continue increasing, CPUs may reduce their performance through thermal throttling. Eventually, some equipment may automatically shut itself down to prevent permanent hardware damage. The actual speed of this process depends on many factors, including room size, IT load airflow, thermal mass, and how much cooling capacity has been lost. The important point is that cooling problems need a fast response. Cooling normally consumes a significant amount of energy. But sometimes, the outside environment can help us. This is called free cooling or economization. Imagine it is a cold winter day outside. Instead of running compressors or chillers at full capacity, the cooling system may be able to use the cool outside conditions to help remove heat. There are different ways to do this. Some systems use outside air directly. Others use heat exchangers so outside air never actually enters the data center. When conditions are suitable, mechanical cooling can be reduced. That saves energy and improves the overall efficiency of the facility. As servers become more powerful, air cooling becomes more difficult. This is especially true for high density GPU and AI systems. That is where liquid cooling becomes important. Liquid can move heat much more efficiently than air. One common design is direct to chip cooling. A cold plate is installed directly on top of components such as the CPU or GPU. Coolant flows through the plate and absorbs the heat. The heated liquid then travels through pipes to a heat exchanger. Another design is rear door heat exchanger where a liquid cooled heat exchanger is installed at the back of the rack. And the immersion cooling entire servers may be placed into a special non-conductive liquid. We are seeing more liquid cooling because modern computing is increasing much higher heat density than traditional servers. Cooling is necessary but it also consumes energy. That means efficient cooling can save a data center a lot of power. One common measurement is called PUE. PUE stands for Power Usage Effectiveness. The idea is simple. We compare the total energy used by the entire facility with the energy used by the IT equipment. For example, imagine the servers, storage, and network equipment consumes 1 MW but the entire facility consumes 1.5 MW. The additional energy is being used by cooling, power conversion, lighting, and other infrastructure. In this example, the PUE would be 1.5 MW. The closer we can move toward 1 MW, the less energy is being used outside the actual IT workload. Efficient cooling, good airflow, containment, and economization can all help improve PUE. Modern cooling systems are not managed only by people walking around the data center. Sensors are collecting information continuously. We can monitor temperature, humidity, fan speed, water pressure, water flow, power consumption, and many other values. This information may be collected by a building management system or BMS. Data center infrastructure management or DCIM may also provide monitoring across both facility and IT infrastructure. Automation can react to changing conditions. For example, if the IT load increases, cooling capacity can be adjusted. Fan speeds may increase, water flow may change. If a cooling unit fails, another unit can take over. And alarms can automatically notify the operations team. This makes the cooling environment more efficient and much more predictable. So now, let's bring everything together. Servers, storage systems, and network equipment consume power and generate heat. Cool air reaches the front of the racks. The equipment pulls that air through the chassis and releases hot air from the rear. Hot IL and cold IL layouts help control the airflow. Containment reduces the mixing of hot and cold air. CRAC or CRAH units remove heat from the air. In larger systems, chilled water transports that heat toward chillers, cooling towers, or dry coolers. Temperature and humidity sensors continuously check the environment. Redundant cooling equipment protects the facility when something fails. And automation helps the cooling system respond to changing IT loads. So, data center cooling is not just one air conditioning unit. It is a complete system. Air flow, cooling equipment, water systems, monitoring, redundancy, and efficiency all work together to keep the IT infrastructure available. And ultimately, that is the purpose of cooling. Keep the equipment cool, protect the hardware and keep the services running.

Transcribe Any Video or Podcast — Free

Paste a URL and get a full AI-powered transcript in minutes. Try ScribeHawk →