Artificial intelligence is changing more than software. It is also changing the physical infrastructure required to support modern computing.
Every search, cloud application, streaming service, business platform, and artificial intelligence model depends on data centers filled with servers, storage equipment, networking devices, and power systems. As computing demand grows, these facilities require more electricity and generate more heat. That heat must be removed continuously to keep equipment operating safely and reliably.
For the HVAC industry, this creates a major long term opportunity. Data center cooling is no longer a small specialty market limited to a few large technology companies. It is becoming an important application for chillers, air handling equipment, heat rejection systems, pumps, controls, heat exchangers, and advanced thermal management technologies.
Data Center Electricity Demand Is Rising Rapidly
The growth of artificial intelligence is accelerating an expansion that was already underway.
According to the International Energy Agency, data centers consumed approximately 415 terawatt hours of electricity worldwide in 2024. In its base case projection, the IEA expects that consumption to more than double to around 945 terawatt hours by 2030. That would represent just under 3 percent of global electricity demand. The agency identifies artificial intelligence as the most important driver of this increase, alongside continued growth in cloud computing and other digital services.
The trend is especially significant in the United States. A report released by the United States Department of Energy found that data centers consumed about 176 terawatt hours of electricity in 2023, equal to approximately 4.4 percent of total United States electricity consumption. The report estimates that data center electricity use could reach between 325 and 580 terawatt hours by 2028, representing between 6.7 percent and 12 percent of national electricity consumption.
These numbers include electricity used by servers as well as the supporting infrastructure required to keep them running. Cooling is one of the largest parts of that supporting load.
More Computing Power Means More Heat
Servers consume electricity to process, transfer, and store information. Most of that electrical energy eventually becomes heat inside the computing equipment and surrounding facility.
Traditional enterprise data centers were often designed around central processing unit workloads with relatively moderate rack densities. Artificial intelligence systems depend heavily on graphics processing units, which can process many calculations simultaneously but also concentrate far more electrical power and heat into a smaller area.
ASHRAE notes that the transition from CPU based computing to GPU based computing is creating new power densities, thermal loads, and infrastructure requirements. Modern AI facilities may also need to support both high density GPU workloads and lower density traditional computing, which means the cooling system must handle different heat profiles within the same building.
This creates a difficult engineering challenge. It is not enough to calculate the total cooling capacity of the building. Designers must also understand where the heat is generated, how quickly the load can change, how air and liquid move through the facility, and how much additional computing capacity may be added in the future.
A data center may have enough total cooling capacity on paper and still experience overheating if cold air does not reach the correct equipment or if hot exhaust air returns to the server inlets.
Data Center Cooling Is Different From Comfort Air Conditioning
A conventional commercial HVAC system is mainly designed to maintain comfortable conditions for people. The load changes with occupancy, outdoor temperature, solar heat, lighting, ventilation, and indoor humidity.
A data center has a very different operating profile. Its cooling load is dominated by electronic equipment, and that equipment may operate every hour of the day throughout the year. Even when outdoor temperatures are low, servers continue to generate heat.
Temperature control also has a direct relationship with equipment performance and reliability. Excessive server inlet temperatures can cause internal fans to work harder, reduce processing performance, or trigger protective shutdowns. Poor humidity management may also increase the risk of condensation or electrostatic discharge.
However, this does not mean every data center should be kept extremely cold. Excessive cooling wastes energy and may increase operating costs without improving reliability. ASHRAE publishes recommended and allowable environmental ranges for different classes of information technology equipment. The appropriate operating range depends on the server specifications, cooling architecture, facility design, and risk requirements.
The goal is not simply to create the coldest possible room. The goal is to deliver the correct thermal conditions to the equipment while using energy efficiently.
Cooling Can Represent a Major Share of Facility Energy Use
The amount of energy used for cooling varies considerably between data centers. Climate, cooling technology, server density, airflow management, operating temperature, water availability, and system controls all affect performance.
The United States Department of Energy states that HVAC systems may account for approximately 25 percent to 40 percent of electricity consumption in some data centers. The exact percentage is not universal, but the figure demonstrates why cooling efficiency has such a large effect on facility operating costs.
One common way to evaluate data center infrastructure efficiency is Power Usage Effectiveness, usually known as PUE. It compares the total energy entering the data center with the energy used directly by information technology equipment.
A PUE of 1.0 would mean that every unit of energy entering the facility goes directly to computing equipment, with no additional energy required for cooling, lighting, power conversion, or other supporting systems. That represents a theoretical ideal rather than a typical operating condition.
A lower PUE generally indicates that less energy is being used by supporting infrastructure. However, PUE should not be used alone to compare every facility. Climate, operating load, building size, redundancy requirements, water consumption, and computing productivity can all influence the result. A well designed cooling strategy should consider energy, water, reliability, and actual computing performance together.
Air Cooling Still Has an Important Role
Air cooling remains widely used in data centers, particularly in facilities with conventional rack densities.
A typical air cooled system supplies conditioned air to the front of server racks. Internal server fans pull the air through the equipment and discharge heated air from the rear. The cooling system then collects the hot return air, removes the heat, and sends cooled air back toward the server inlets.
Proper air management is essential. When hot exhaust air mixes with cold supply air, the system may need to produce more airflow or lower supply temperatures to protect the hottest equipment. This increases fan and compressor energy use.
Hot aisle and cold aisle arrangements help separate supply air from exhaust air. Containment systems can improve this separation further by enclosing either the hot aisle or the cold aisle. Sealing unused rack openings and cable penetrations also reduces recirculation.
The United States Department of Energy recommends separating hot exhaust air from cool supply air because higher return temperatures can improve cooling capacity, reduce required airflow, and increase the number of hours when economizer cooling is available.
Air cooling may include computer room air conditioners, computer room air handlers, central air handling units, in row cooling equipment, overhead systems, and rear door heat exchangers. The best arrangement depends on rack density, room layout, existing infrastructure, and future expansion plans.
Chilled Water Systems Support Larger Cooling Loads
Larger data centers often use centralized chilled water systems.
In this type of design, chillers remove heat from a water loop. Pumps distribute the chilled water to computer room air handlers, air handling units, in row cooling equipment, or heat exchangers located closer to the information technology load.
Chilled water systems can provide large cooling capacities and support centralized equipment management. They can also be designed with multiple chillers, pumps, cooling circuits, and heat rejection devices to meet redundancy requirements.
Modular chiller configurations may allow capacity to be added as the computing load grows. Variable speed compressors, pumps, and fans can help the system respond more efficiently when the facility is operating below its full design load.
The performance of a chilled water system depends on more than the rated efficiency of the chiller. Chilled water temperatures, condenser conditions, pump pressure, equipment staging, heat exchanger performance, water flow, and control sequences all influence actual energy consumption.
A highly efficient chiller cannot deliver its expected performance if the pumps operate at unnecessary speeds, the return water temperature is too low, or multiple units run inefficiently at partial load. Data center cooling therefore requires coordinated system design rather than isolated equipment selection.
Economizer Cooling Can Reduce Compressor Operation
Economizers use favorable outdoor conditions to reduce the amount of mechanical refrigeration required.
An air side economizer introduces filtered outdoor air when temperature and humidity conditions are suitable. A water side economizer transfers heat through cooling towers, dry coolers, fluid coolers, or heat exchangers without relying entirely on compressor based cooling.
The potential savings depend strongly on the local climate. A data center in a cool, dry region may have many hours of economizer operation. A facility in a hot and humid location may have fewer opportunities and may require more mechanical cooling throughout the year.
Outdoor air quality, humidity control, contamination risks, water use, freezing protection, and local design temperatures must all be evaluated. Economizer cooling is not simply free cooling without consequences. It must be integrated carefully with the rest of the thermal management system.
Liquid Cooling Is Expanding With AI
As rack densities increase, moving enough air through the server room becomes more difficult. Fans require additional energy, ducts and floor openings have physical limits, and high velocity airflow may still struggle to remove heat from the most concentrated components.
Liquid cooling brings the cooling medium closer to the heat source. Because liquids can transport much more heat than the same volume of air, they can support higher equipment densities and reduce the amount of room airflow required.
Direct to chip cooling circulates liquid through cold plates attached to processors and other high heat components. Rear door heat exchangers remove heat from server exhaust as it leaves the rack. Immersion cooling places electronic components in a specially designed dielectric fluid.
These systems are not interchangeable. Each approach has different requirements for server compatibility, pumps, piping, leak management, fluid quality, controls, maintenance, and heat rejection.
ASHRAE’s 2026 guidance recognizes that high density AI systems are driving greater use of liquid cooling. It also emphasizes that modern facilities may need to support a combination of cooling profiles instead of relying on a single technology throughout the entire building.
Liquid Cooling Does Not Eliminate HVAC
Liquid cooling is sometimes presented as a replacement for traditional air conditioning, but that description is incomplete.
A direct to chip system removes heat from processors, but the heat still needs to be transported away from the server and rejected outside the facility. The system may require cooling distribution units, pumps, heat exchangers, facility water loops, dry coolers, cooling towers, or chillers.
Not every component in a server rack is connected to a liquid cooling loop. Memory, storage equipment, networking hardware, power supplies, and other components may continue to release heat into the surrounding air. Electrical rooms, battery rooms, offices, and support spaces also require ventilation or air conditioning.
For this reason, many AI data centers are expected to use hybrid cooling systems. Liquid cooling can serve the most concentrated heat sources, while air cooling continues to manage the remaining room and equipment loads.
ASHRAE’s retrofit guidance describes a hybrid approach as an effective way to introduce direct to chip cooling without abandoning existing air cooling infrastructure. This is particularly important for facilities that must support both new GPU systems and older air cooled equipment.
Reliability Is as Important as Efficiency
Cooling efficiency matters, but data center operators cannot reduce energy use at the expense of uptime.
A cooling system failure can cause room temperatures to rise quickly, especially in facilities with high rack densities. Unlike a conventional office, a data center cannot simply stop cooling when the building is unoccupied.
Critical facilities often use redundant chillers, pumps, fans, cooling distribution units, controls, and electrical supplies. An N+1 configuration provides one additional unit beyond the number needed to meet the design load. More demanding facilities may use greater levels of redundancy or physically separate cooling paths.
The appropriate redundancy level depends on the business impact of downtime, maintenance strategy, data center classification, available budget, and ability to transfer computing workloads to another location.
Controls must also respond correctly when equipment fails. Starting a backup chiller is not enough if isolation valves remain closed, pumps do not provide the required flow, or control systems do not recognize the failure.
Reliability therefore depends on system architecture, commissioning, monitoring, maintenance, spare parts, operating procedures, and trained service personnel as much as it depends on equipment quality.
Intelligent Controls Create Additional Efficiency Opportunities
Data center cooling loads are not always constant. Artificial intelligence training jobs, cloud traffic, maintenance activities, and server utilization can create significant changes in heat output.
A well designed control system can adjust cooling capacity as these loads change. It can stage chillers, reset water temperatures, vary pump speeds, control fan airflow, monitor rack inlet conditions, and select the most efficient heat rejection method for current outdoor conditions.
Sensors should be placed where they represent actual equipment conditions. Measuring only the average room temperature may hide local hot spots near high density racks.
Integrated monitoring can also identify declining heat exchanger performance, unusual temperature differences, excessive pump pressure, blocked airflow, and equipment that is operating outside its expected efficiency range.
ASHRAE’s current AI data center framework emphasizes integrated thermal management rather than treating power, cooling, computing load, and controls as separate systems. This approach helps facilities improve scalability, resilience, and energy performance at the same time.
Water Use Must Be Considered Alongside Energy Use
Some highly efficient cooling systems rely on evaporative heat rejection, which can reduce electricity consumption but increase water use.
This creates a tradeoff. In regions with limited water supplies, operators may prefer air cooled chillers, dry coolers, or hybrid heat rejection systems. These options can reduce site water consumption, but they may require more electricity or larger equipment during hot weather.
The best solution depends on local climate, electricity prices, water availability, environmental regulations, operating temperatures, and sustainability targets.
NREL has demonstrated that advanced liquid cooling, heat recovery, and hybrid heat rejection can achieve very low PUE while also reducing water use. Its Energy Systems Integration Facility uses warm water liquid cooling and captures heat from its computing equipment for reuse in other parts of the building. The project demonstrates the value of designing computing, cooling, heat rejection, and heat recovery as one integrated energy system.
Not every project can reproduce this exact design, but the principle is broadly relevant. Data center cooling should be evaluated as part of a larger energy and resource strategy.
Existing Data Centers Create a Large Retrofit Market
New construction receives significant attention, but many cooling opportunities will come from existing facilities.
Older data centers may have been designed for much lower rack densities. Their chillers may have enough total capacity, but the air distribution system may be unable to deliver cooling to new GPU racks. Pumps and fans may operate at fixed speeds. Controls may respond to room temperature instead of server inlet conditions. Hot and cold air may mix throughout the space.
Upgrades can include aisle containment, airflow improvements, variable speed drives, higher efficiency chillers, better control sequences, improved sensors, rear door heat exchangers, cooling distribution units, and direct to chip liquid cooling.
A retrofit does not always require replacing the entire cooling plant. In many cases, improving air separation and control can release capacity that already exists but is being used inefficiently.
However, retrofit decisions must consider the full system. Adding liquid cooled racks may require new piping, structural reinforcement, leak detection, water treatment, heat exchangers, pumps, and outdoor heat rejection capacity. It may also change the temperature and flow requirements of the existing chilled water plant.
Data Center Cooling Is Becoming a Complete System Business
The growth of data center cooling does not only create demand for more equipment. It changes what customers expect from HVAC suppliers, contractors, and engineering teams.
Projects increasingly require coordination between mechanical systems, electrical infrastructure, information technology equipment, building controls, water systems, and facility operations. Equipment must be selected around actual operating conditions, future load growth, redundancy requirements, and maintenance needs.
This creates opportunities across the HVAC value chain. Chiller manufacturers can support larger and more modular cooling plants. Air handling equipment suppliers can develop solutions for high sensible heat loads and precise airflow management. Controls companies can coordinate equipment performance in real time. Contractors and service providers can support commissioning, preventive maintenance, upgrades, and emergency response.
The strongest opportunities will belong to companies that understand how these parts work together.
What HVAC Buyers Should Evaluate
Choosing a data center cooling system begins with the information technology load, but it cannot end there.
The design team must understand current and expected rack densities, total heat load, server inlet requirements, operating schedule, local design temperatures, available electrical capacity, water availability, redundancy expectations, installation space, maintenance access, future expansion plans, and control requirements.
The team must also determine whether the facility will use air cooling, liquid cooling, or a hybrid architecture. That decision affects nearly every part of the mechanical system, from indoor heat collection to outdoor heat rejection.
Equipment ratings should be evaluated at actual project conditions rather than only at standard test conditions. This is especially important in hot climates, where high outdoor temperatures can reduce cooling capacity and efficiency.
For mission critical projects, technical support and service capability should also be considered during equipment selection. A highly efficient system provides limited value if replacement parts, trained technicians, or emergency support are unavailable.
A Long Term Opportunity for Commercial HVAC
Data center cooling is becoming one of the most demanding applications in the commercial HVAC market.
Artificial intelligence is increasing computing density. Cloud services are expanding data center capacity. Cooling represents a significant operating cost. Existing facilities require upgrades, while new projects must prepare for technologies that may change throughout the life of the building.
The market will not be served by one universal cooling solution. Air cooling will continue to support many facilities and equipment types. Chilled water systems will remain important for centralized cooling. Liquid cooling will expand as rack densities rise. Economizers, heat recovery, advanced controls, and hybrid heat rejection will become more valuable as operators balance reliability, energy use, and water consumption.
For the HVAC industry, the opportunity extends beyond selling more cooling capacity. It is an opportunity to provide more intelligent, scalable, efficient, and resilient thermal management systems.
ZERO provides commercial HVAC solutions including air cooled chillers, modular chilled water systems, and customized air handling equipment for a range of project conditions. Data center and mission critical applications require project specific engineering, so equipment selection should always be based on operating temperatures, load profiles, redundancy requirements, controls, and applicable technical standards: zerohvacr.com
As digital infrastructure continues to expand, cooling will remain essential to its performance. The companies that can remove heat reliably and efficiently will play an increasingly important role in the future of computing.
