Data centers, as centralized, large-scale, park-like, and independently operated entities, are characterized by numerous energy storage systems, high power consumption, poor heat dissipation, and large amounts of flue gas. In the event of a fire, this poses a severe challenge to personal safety, system security, and business operations. As critical infrastructure, data centers are planned, constructed, and operated to the highest standards. However, publicly available information shows that data center fires have occurred frequently worldwide in recent years, with varying causes. Some data is compiled as follows:

Statistics of Data Center Fires in the Last Five Years (Partial)

From the results, it seems that data center fires are relatively less frequent in China, but in reality, they are not uncommon, especially battery fires, some of which have even affected the normal operation of business systems. However, because customer disaster recovery systems worked, they did not lead to public reports. Data center fires are basically Primarily electrical fires due to the existence of a large number of battery systems prone to fire , and air conditioning systems that can accelerate the spread of fire, characterized by strong concealment, high randomness, and rapid spread when they occur.

Analysis of the Causes of Data Center Fire Accidents:

There are many ways to classify data center fire accidents. From the cause of the fire (or the initial ignition point), they can be divided into electrical faults, thermal runaway, human error, and insufficient maintenance. 4 main aspects. Accidents with unclear specific causes UPS System in battery rooms are grouped under electrical faults, and accidents with no publicly available causes are grouped under unknown causes. The overall statistics are as follows:

(1) Electrical Faults

Electrical faults are the most common cause of data center fires, accounting for over 50% . These faults may stem from arc flashes, short circuits, lightning strikes, surges, overloads, and equipment malfunctions, each capable of generating sufficient heat to trigger a fire.

1. Arc Flash: Discharge caused by low-impedance connections in the electrical system. The high-intensity flash during an arc flash can easily ignite surrounding materials and equipment, especially in environments containing charged batteries and capacitors.

2. Short Circuit: Iron filings hidden in cabinets during the construction phase of electrical equipment in data centers, aging, insulation damage, and poor contact after prolonged use can all lead to electrical short circuits. The current during a short circuit can reach tens of times the rated current, easily causing electrical equipment to burn out and even triggering a fire.

3. Lightning Strike: Lightning strikes are more common during the rainy season, especially during the spring and summer when thunderstorms are frequent. If the enormous energy released by lightning cannot be completely handled by lightning protection facilities, it is very easy to cause energy overflow and trigger a fire. Although data centers have good lightning protection and grounding facilities and conduct regular lightning protection inspections every year, this does not guarantee that there will be no flaws.

4. Surges: Caused by a sudden and significant increase in voltage, surges can lead to circuit overload and generate intense heat. If such surges penetrate equipment that is not resistant to high voltage, such as switching power supplies, UPS System the risk of fire will increase. There are no clear direct cases of such accidents, but this does not mean they have not occurred, and they still need more attention.

5. Equipment Failure: Many accidents do not give specific reasons; they only mention a fire in a certain room or a failure in a certain system. This could also be due to short circuits, overloads, etc. Accidents without clear reasons but with identified equipment are categorized as equipment failures. Among them, power distribution, UPS System (excluding batteries), diesel generator systems, and rooftop photovoltaic systems have all experienced fires, indicating that any part of the electrical system poses a hidden danger.

6. Cable Problems: Cable tunnels, cable shafts, and connecting cables, due to long-term load, aging, and wear, are also a factor that cannot be ignored. Data center cable tunnels usually do not have fire alarm systems installed, and even temperature-sensitive optical fibers are rarely deployed, lacking early warning of initial fires.

(II) Thermal Runaway

Thermal runaway is another common cause of data center fires and can be divided into battery fires, IT equipment fires, and fires caused by cooling system failures. 3个方面。

1. Battery Fires: Data Center UPS System Systems use a large number of lead-acid batteries and lithium batteries to ensure business SLA ， EPS systems use batteries, and large-scale energy storage devices are also being used. These high-energy-density batteries pose a significant fire risk. Overheating or damage to these batteries may cause them to enter a thermal runaway state, leading to battery fires or even explosions. Densely packed batteries can trigger a chain reaction, causing the fire to spread further and eventually leading to an uncontrolled fire accident.

Compared to lead-acid batteries, lithium batteries have a smaller footprint, higher cost-effectiveness, easier maintenance, and longer lifespan, making them increasingly common in data centers. However, lithium batteries pose a greater fire risk than lead-acid batteries. Many battery fire incidents have occurred in China, but they have not been publicly reported because they did not have external impacts. We welcome any additional information.

2. IT Equipment Overheating: IT With increasing power density of equipment, hardware quality issues, long-term overload, high-load operation, and lack of maintenance can all lead to overheating of high-density servers and even fires.

3. Cooling System Failure: When the cooling system fails or is insufficient, the heat generated by servers and network equipment may cause a fire. Although many fatal cooling system failures have occurred in data centers, they were handled relatively promptly, only affecting the operation of business systems and not resulting in fires.

(III) Human Error

Human error mainly includes management and operational errors during project construction, upgrades, renovations, and maintenance, including poor management of special operations such as fire and electricity use, inadequate protection during battery installation, transportation, or discharge experiments, and insufficient safety preparations.

1. Special Operations Management. Fires caused by special operations during construction projects have also occurred, including 2022 the GuoKe data center under construction in Suzhou, Jiangsu Province in [year], and 2024 the Alibaba data center under construction in Guangdong Province in [year], both due to welding, cutting, and other special operations, inadequate on-site protection, and poor management, igniting nearby debris and causing fires.

2. Daily Supervision. Inadequate daily fire safety publicity and management, especially the lack of safety education and supervision for incoming customers, and the lack of control over smoking and electrical equipment. For example, 2022 the fire on the exterior wall of the second telecommunications building of Hunan Changsha Telecom in [year], and 2023 the Khawaja Tower Khawaja Tower Data Center ISP office fire in Bangladesh in [year].

(IV) Inadequate Maintenance

During the operation of a data center, equipment and facilities require continuous maintenance, upkeep, repair, upgrades, renovations, and replacements. Any flaw in any of these stages may pose a risk to safe and stable operation.

In reality, due to cost control, the number of data center maintenance personnel is decreasing, salaries are decreasing, and work is being reused. The professional skills and sense of responsibility of technical personnel are declining, and maintenance problems frequently occur. Many data centers lack effective management measures for aging and soon-to-be-aging equipment, which may lead to fire hazards being overlooked and unresolved.

Data center fires are " low-frequency, high-impact ” risks. We must not " ignore ” their potentially devastating consequences because they are 几 infrequent. Recent cases have shown us that any potential hazard or risk can ultimately lead to a fire. In the context of cost reduction and efficiency improvement, and energy conservation and emission reduction, how to balance cost and safety in design, construction, and operation is a question that data centers need to deeply consider.

-----------------------------------------------------------------------------------------------

Information Source: Analysis of Major Data Center Fire Accidents in the Past Decade (Part 1)

The only lesson we have learned from fire accidents is that we have never learned a lesson.

Data Center Security Management - Fire Accident Cause Analysis