Data center construction is expanding rapidly to support cloud computing, artificial intelligence workloads and increasing power densities. In April 2026, the Pew Research Center reported that more than 3,000 data centers are currently in operation across the United States, with more than 1,500 more planned or under construction (https://pewrsr.ch/4eHlHBW).

What makes these facilities different is not only their scale, but their risk profile. Owners expect 24/7/365 operation, yet the equipment inside can be severely damaged by both fire and water. No single suppression system solves every problem, so project teams must weigh tradeoffs between protection, cost and operational impact.

Contractors play a larger role in these projects than they often do in conventional buildings. Design intent only works if it can be built, tested and maintained under real conditions. Field input on routing, sequencing and access often changes how systems are ultimately configured.

Codes, standards and decision-making

Data center projects still begin under the same code framework as other commercial buildings. The enforced building and fire codes establish the baseline requirements for construction type, occupancy, fire resistance, detection and sprinkler protection. On paper, many data centers resemble large industrial or business occupancies.

Suppression system selection is where data centers deviate. Instead of pointing to a single prescribed approach, the codes push designers toward a network of referenced standards: NFPA 13, Standard for the Installation of Sprinkler Systems; NFPA 75, Standard for the Fire Protection of Information Technology Equipment; NFPA 750, Standard on Water Mist Fire Protection Systems; and NFPA 2001, Standard on Clean Agent Fire Extinguishing Systems. Each standard allows multiple compliant approaches depending on the hazard and the owner’s priorities.

FM Global Data Sheet (FMDS) 5-32, Data Centers and Related Facilities, goes further by tying protection strategy to building geometry, airflow, power systems and system arrangement. The standard emphasizes system reliability and fire control performance over code compliance. Designers must match protection to actual operating conditions while maintaining dependable fire control.

Wet pipe: The starting point

Most projects start with wet pipe simply because it is familiar, reliable and cost-effective. In offices, support spaces and similar low-risk areas, it’s usually the right answer.

Wet pipe systems also perform well during actual fire events. Sprinklers activate individually, apply water directly to the fire and control it before it spreads. Real-world data consistently shows that only a small number of heads operate in most incidents, and accidental discharges are almost always tied to mechanical damage or installation issues, not spontaneous failure.

The concern in data halls shifts from fire performance to water release into sensitive equipment areas. FMDS 5-32 places greater weight on reliable fire control than on preventing water exposure because uncontrolled fire and smoke result in significantly greater loss than water damage. As rack densities increase and layouts become more complex, owners become less willing to accept that risk.

For that reason, wet pipe is typically limited to areas where water exposure is acceptable or where another system is the primary means of fire protection.

Preaction sprinkler: The most common solution

Preaction systems are widely used in data halls because they delay water entry into the piping until a detection event occurs. This reduces the chance of unintended discharge while keeping the basic performance of a sprinkler system. 

NFPA 13 and NFPA 75 allow several configurations, and FMDS 5-32 connects those options to risk tolerance and system reliability. The key differences between preaction system types are what is required to open the valve and whether water is present in the piping before a sprinkler operates. FMDS 5-32 explicitly ranks them in order of preference because complexity and water delay can significantly affect fire growth.

In a noninterlocked preaction system, the valve operates solely on detection. When the detection system activates, the preaction valve opens and fills the piping with water. At that point, the system effectively behaves like a wet pipe system — water is in the piping, but it will not discharge into the space unless a sprinkler opens. 

If a sprinkler operates before detection, the valve will open due to a loss of air pressure in the piping, allowing water to be delivered to the fire. This arrangement delivers water the fastest, but it offers the least protection against unintended sprinkler discharge.

A single-interlock preaction system requires a detection event to open the valve and admit water into the piping. Until detection occurs, the piping remains dry. If a sprinkler operates before detection, water will not enter the system, and discharge will not occur until the detection system activates. This reduces the chance of accidental water introduction into the piping but can delay water delivery if detection is slow or impaired.

A double-interlock preaction system requires two independent events before water enters the piping: detection must activate, and a sprinkler must open (resulting in loss of supervisory air pressure). Both conditions must be satisfied before the valve opens. This provides the highest level of protection against unintended water release because neither a false detection signal nor accidental sprinkler damage alone will introduce water into the system. 

The tradeoff is increased complexity and the potential for slower response if either input is delayed. FMDS 5-32 cautions that these delays can negatively impact fire suppression and control for these systems.

In all cases, once water enters the piping, discharge still depends on sprinkler operation. The differences are how easily water is allowed into the system before that point and how much the design prioritizes response speed over protection against accidental water exposure. 

These systems require reliable detection. Airflow, containment and return air paths all affect how quickly smoke reaches a detector. Aspirating systems are often used to pick up very early smoke movement, especially in return air. Spot detectors still handle code coverage and zoning, but rarely carry the full load on their own.

On one of this author’s projects, spot detection met coverage requirements on paper, but airflow patterns would have excessively delayed response times. The final design added aspirating detection in hot aisles to hasten system response to the double-interlock preaction system. This kind of adjustment is common.

Water mist: Not a universal solution

Water mist limits water damage by using very fine droplets to cool the fire and interrupt combustion. In controlled environments, it can limit both fire and water damage.

It works best where the hazard is well-defined and the environment is controlled, such as in generator enclosures, uninterrupted power supply rooms and similar spaces. In those conditions, the mist stays concentrated long enough to do its job.

 Water mist is difficult to apply in data halls because high airflow and large volumes prevent droplets from staying concentrated to suppress a fire. Air movement can carry mist away.

 FMDS 5-32 requires application-specific approval and test data that reflects the actual space. Outside those conditions, performance is less certain. Ceiling height, ventilation rate and equipment layout all factor in whether the system will work.

Water mist systems in data centers are typically engineered solutions. Designers must account for the airflow, room geometry and cooling system behavior. Installation also matters more than with traditional sprinklers because component spacing and configuration must closely match the tested arrangement.

When everything lines up, water mist can be effective. When it doesn’t, performance drops off quickly. This is why it tends to stay in targeted applications rather than replacing sprinklers.

Clean agents: ‘No water’ is not always better

Clean agent systems are appealing because they leave no residue and can suppress a fire early. In practice, they are highly sensitive to enclosure and airflow conditions.

These systems only work if the enclosure holds the agent long enough to reach and maintain the required concentration. Even small leaks, such as cable penetrations, floor openings and return air paths, reduce suppression effectiveness.

FMDS 5-32 addresses this directly by requiring room integrity testing. Leakage rates determine whether the system will perform.

Similar to water mist, airflow is a major challenge for clean agent suppression because data halls are not sealed rooms. Supply and return systems move large volumes of air unless they are actively shut down. If airflow continues after discharge, agent concentration can drop off almost immediately. As a result, water mist and clean agent systems must be integrated into the fire alarm sequence of operations to shut down cooling equipment before suppression agent release.

Clean agents still have a place, particularly as an early response layer ahead of water-based systems. However, their performance depends less on the agent itself and more on how well the space is controlled and maintained.

Lithium-ion batteries

Lithium-ion battery (LIB) use in data centers as distributed battery backup units within server racks is becoming more prevalent. These batteries are a hot topic in the fire protection industry due to concerns about thermal runaway and the difficulty of suppressing LIB fires.

FMDS 5-32 restricts some suppression methods when LIBs are present in server racks. Clean agent and water mist are not permitted as primary suppression methods. Instead, wet-pipe sprinklers with increased densities and specific layouts, along with vertical barriers, are recommended. 

This is a shift from previous suppression strategies, where clean agents were the preferred option. LIBs require the suppression system to control fire growth from the fire origin through cooling and exposure protection, even at the expense of water exposure to surrounding equipment.

As rack power densities increase and distributed battery systems become more common, suppression design is increasingly driven by LIB hazards rather than traditional cable and equipment risks. 

Choosing the right strategy

The right approach depends on what the owner is trying to protect and the level of acceptable risk.

Is the priority minimizing downtime, limiting equipment damage or controlling installation cost? Teams also need to decide where water discharge is acceptable and how much system complexity the facility can support over time.

Postconstruction modifications introduce additional challenges. Data centers rarely stay the same. Rack densities increase, equipment changes and cooling strategies evolve. Systems such as clean agent and water mist are sensitive to those changes, while sprinkler-based systems tend to tolerate them better, though not without limits.

Installation quality is as critical as system selection. Routing, sealing and obstruction control all influence performance. Small deviations from proper design, especially in mist and clean agent systems, can negatively impact performance.

Commissioning verifies design assumptions. Detection response, control sequences, airflow interaction and enclosure integrity all need to be verified under realistic conditions. After turnover, maintaining that performance becomes an ongoing responsibility.

Over time, performance depends on how well the system is installed and how conditions are maintained.