Singapore Data Centre Safety: Managing Physical Risks as AI Infrastructure Scales
“Quick AI-Powered Insights on the Topic— Freshly Updated!”
|
Data centre safety in Singapore now depends on how well operators manage physical risks inside facilities that are getting denser, hotter and closer to full capacity. Data from AI workloads have pushed rack power beyond conventional cloud levels and upgrades are taking place in active sites, leading to a constant flow of contractors. The best way to observe the risks in real-time is to use AI video analytics on existing CCTV cameras, transforming the recorded footage into alerts for smoke, unauthorized entry, unsafe contractor activity, water leaks, and other safety and security issues.
Why is Data Centre Safety in Singapore Getting Harder?
AI infrastructure is changing the physical load of facilities. As per a study on Singapore's data centre construction market, GPU clusters may require rack densities of between 50 and 150 kW, compared to 5 to 10 kW for traditional cloud set-ups. Increased power consumption per rack leads to bigger electric rooms, larger batteries and UPS installations, more heat to remove, and in numerous new facilities, liquid cooling pipelines positioned nearby live machinery.
The second pressure is space. The same study shows that, Singapore's colocation vacancy rate at approximately 1.4% for 2025, is the lowest in the entire Asia Pacific region. When space is limited, the only growth option available is retrofitting existing halls. This is where electricians, mechanical fitters, cabling crews, cleaners and vendor engineers move through the facility, with the customer data centre being operated with servers up and running.
That kind of work carries its own risk profile. According to an analysis conducted in 2026, between 2/3rd and 4/5th of major IT and data centre outages involving some element of human error, and failures to follow established procedures remain the leading driver of those accidents. In a live facility where several contractor teams work across the same halls, every skipped step is both a safety risk and an uptime risk.
What Singapore’s Digital Infrastructure Act Means for Physical Risks?
On October 7, 2026, the Digital Infrastructure Bill was passed by Parliament of Singapore. When the Bill was presented by the Ministry of Digital Development and Information, it stated that the new legislation will have a wider range of operational risks, extending beyond cyber threats, covering technical failures, power and cooling issues, fire and other physical incidents and so on.
Licensed operators will need to have security risk management measures, as well as business continuity and disaster recovery plans, and must report on the incidents and disruptions to IMDA. Serious breaches will be penalized up to S$1 million or 10% of annual Singapore turnover, whatever is greater.
However, the detailed codes of practice are still to be published. Nevertheless, the direction is already clear. A fire in a battery room or a water leakage in a data hall is no longer considered a mere operational issue. It is now a reportable event and the operator will be responsible for explaining the circumstances of the event, specifying when it happened and who was present.
Where Physical Risks Concentrate Inside a Data Centre?
Most serious incidents in a data centre begin in small spaces. Each one has a different risk and different person responsible for it.
Battery Room Safety and UPS Halls
In September 2024, a fire broke out in a battery room at the Digital Realty’s SIN11 premises, in Singapore. This incident disrupted Alibaba Cloud services in the area, and water from firefighting operations entered a server room, leading to a power shutdown in one building. The root cause is yet unknown.
In Singapore, regulations concerning the design aspect of the risk is already in place. The Singapore Civil Defence Force (SCDF) Fire Code requirements for energy storage systems set controls like setback distances, fire-rated compartments, and storage limits and enhanced sprinkler systems.
What the design controls cannot cover is daily behaviour. Battery room safety also depends on who opens the door, whether packaging is left near battery strings and whether a technician is working alone without a permit.
Electrical Rooms and Switchgear
Switchgear rooms hold the danger of arc flash and electrocution and therefore should be accessed only by qualified and properly-equipped people. Since exposure is short and infrequent, it becomes hard to supervise. A five-minute task without arc-rated PPE, for instance, rarely shows up in a log and gets identified only in case an incident happens.
Cooling Plant and Liquid Cooling Lines
Higher rack densities bring more chilled water and, increasingly, direct liquid cooling into the white space. A slow leak near power distribution creates both an electrical hazard and a slip hazard, and it can grow unnoticed between inspection rounds.
Data Halls During Maintenance and Fit-Out
During maintenance works, data halls fill with ladders, lifted floor tiles, heavy equipment on trolleys and temporary cabling. Floor openings without barricade, unsafe ladder use and blocked aisles are common, and each on can turn a routine job into an injury or an outage.
Loading Bays, Corridors and Emergency Routes
New equipment arrives in crates, and packaging tends to collect in corridors. Blocked exits and narrowed evacuation routes are easy to miss in a building that rarely needs them, until the day it does.
What Can Existing CCTV Detect in a Data Centre?
Most data centres already have dense camera coverage for security. Singapore data centre safety teams, can turn these exiting cameras into an advanced safety system that can work around-the-clock tirelessly, but powering them with Computer Vision AI. Existing CCTV cameras powered by AI video analytics can help detect risks that data centres commonly face, without any separate camera network.
AI can be used to define virtual restricted zones in switchgear, UPS and battery rooms. When existing cameras pick up an unauthorised person entering these zones, AI sends an instant alert to both on-site and remote supervisors. This allows supervisors to stop unsafe access before anyone touches live electrical equipment or lithium-ion batteries.
Colocation sites host technicians from many customers who do not work for the operator. AI links camera feeds at restricted-zone entry points to the electronic permit to work system (e-PTW). It checks each contractor or vendor against an active permit for that area and time window, and flags anyone without permit. Every entry is recorded, so the operator knows exactly who went where. For facilities hosting financial institutions, this also supports Monetary Authority of Singapore (MAS) expectations that visitors are escorted at all times and that access to equipment racks is recorded, monitored and supervised.
CCTV covering battery rooms, cable trays and UPS halls is trained to recognize visible smoke and early fire. When viAct AI detects either, it triggers both on-site alerts (via. light and sound siren) and remote alerts (via. SMS. WhatsApp, Telegram and viAct mobile app) in real-time. This helps the team to know exactly where the fire is starting. In spaces packed with lithium-ion batteries and dense cabling, those first five-minutes decide whether a small fire stay contained or turns catastrophic.
In data centres, cameras already monitor the activities carried out in the hot and cold aisles. When powered by AI, these ordinary cameras become capable of identifying cabinet doors that have not been closed, obstructions in airflow path, and maintenance activities that have been left unattended. Although these occurrences do not lead to injuries very often, they create thermal hotspots and increase the risk of equipment failure as rack density grows.
Inside switch rooms and other energised zones, AI monitors if the workers are wearing necessary PPE, matched by their job and zone. It also recognizes when a technician operates alone, detects if the worker has fallen, or has become motionless. It is very important that the control room is alerted immediately about these happenings, especially during night shifts and after-hours work.
As liquid cooling moves closer to the IT load, chilled water lines now run through data halls and alongside power distribution. Cameras covering cooling plant, pipework and rack rows can detect water appearing on floors, above racks or near electrical equipment. AI flags the leak as soon as it appears, so teams can isolate it before a slow drip between inspection rounds becomes an electrical fault.
Cameras covering exits and corridors detect blocked doors, stored materials, packaging and trailing cables that narrow an evacuation route. Computer Vision can flag these in real-time during deliveries and fit-out work, so routes are cleared before an emergency.
Apart from these, the same cameras can monitor other risks across a data centre, like:
Data centre risk | How existing CCTV can help |
Vehicle and pedestrian collisions at loading bays and generator yards | AI can track forklifts, pallet jacks and delivery trucks around pedestrians and trigger instant alerts on unsafe proximity and on vehicles entering restricted zones. |
Open floor tiles and unsafe ladder use during fit-out | AI can flag raised-floor tiles left open without barricades. It can also spot unsafe ladder use near overhead cable trays before a fall happens. |
Climbing racks or cable trays and unsafe access | AI can detect workers climbing racks or cable trays, running in data halls, or reaching areas through unsafe routes. Supervisors can step in early through real-time alerts. |
Spills, clutter and improper storage in equipment rooms | AI can identify spills, clutter and packaging or spare parts stored where they add fire load or create trip hazards. |
Manual handling injuries during server and battery installation | AI can recognise unsafe lifting posture when teams move batteries, servers and heavy equipment. EHS teams can then correct handling practices. |
Perimeter intrusion and after-hours loitering | AI can detect people crossing fences or gates and loitering near entrances after hours. Security is alerted before access is attempted. |
The risks covered above are only part of what a data centre has to manage every day. viAct AI video analytics for data centre addresses these and many more through its 200+ pre-build AI modules for real data centre scenarios.
Across hyperscale, colocation, edge and enterprise facilities where it is deployed, viAct has helped operators cut data centre incidents by 60%. The gains come from three areas where data centres are most exposed:
Incident detection is 90% faster as arc flash risks, unauthorised access, and live-work violations are flagged the moment they occur.
Contractor compliance rises by 45% with permits, PPE and restricted zone entry verified for every contractor on every shift.
Human error downtime falls by 50% as unsafe actions are stopped before they trigger outages across switchgear, battery and cooling systems.
How Vision AI Supports Incident Reporting Under the Digital Infrastructure Act?
The Digital Infrastructure Act, makes it compulsory for the licensed operators to report specified incidents and disruptions to IMDA. Whenever a fire, leak or unauthorised entry lead to an outage, the first questions that arise are always the same: what happened, when did it start, and who was present?
In data centres, CCTV footage is recorded continuously but they are rarely watched in real-time. Without alerts that flag the moment an incident began, answering these questions means hours of manual footage review across multiple cameras.
viAct closes that gap. Every alert raised by viAct AI safety system is saved in viHUB, viAct centralised management platform, with a timestamped clip of the event. The record stays intact, so teams can return to it for investigation and incident reporting. Further, viGent, viAct AI safety co-pilot, converts video alerts into text and digitises manual reporting forms, reducing reporting time by up to 75%. Teams can also ask viGent what happened, where and who was involved. It can also analyse trends over time and then highlight recurring issues before they cause the next disruption.
Conclusion and Key Takeaways
As AI infrastructure scales, Singapore data centre safety is becoming harder to separate from uptime. Battery rooms, switchgear, cooling plant and live data halls all carry risks that design controls reduce but cannot remove. Meanwhile, the Digital Infrastructure Act now treats fires, power failures and cooling incidents as resilience matters. Continuous visual monitoring on existing cameras gives operations and EHS teams a shared, real-time view of these spaces, with evidence attached to every alert.
Key Takeaways
Singapore data centre safety is getting harder as AI racks reach 50 to 150 kW and upgrades take place inside live facilities. viAct addresses this with AI video analytics that runs on a data centre’s existing CCTV, with no new camera hardware.
Unauthorised entry into switchgear rooms, UPS halls and battery rooms is one of the most serious physical risks in a data centre, because a single wrong move near live equipment or lithium-ion batteries can cause an injury or an outage. viAct flags these entries in real-time and alerts the control room before anyone reaches the equipment.
In colocation facilities, technicians from many different customers enter the building frequently, and most of them donnot work for the operator. Verifying each one against an active permit at restricted-zone entry points, as viAct’s electronic permit to work system does, closes a gap that badge readers alone leave open.
Data centre operators using viAct across hyperscale, colocation, edge and enterprise facilities have cut incidents by 60%. They have also seen incident detection become 90% faster and contractor compliance rise by 45%.
Under Singapore’s Digital Infrastructure Act, operators must report specified incidents to IMDA. viAct saves every alert in viHUB with a timestamped clip, and viGent, its AI safety co-pilot, helps teams trace what happened, where and who was involved.
FAQs
1. Can AI manage data centre safety on its own?
Not entirely. AI can monitor cameras around the clock and alert teams in real-time when something goes wrong, such as unauthorised entry, smoke or a water leak. However, ultimately, it is the people who decide how to respond and carry out the fix.
2. Can AI video analytics replace VESDA?
No. VESDA (Very Early Smoke Detection Apparatus) detects smoke particles in the air, often before smoke is visible. AI video analytics adds visual confirmation of where the smoke is and whether anyone is in the room and trigger real-time alerts both on-site and remote.
3. Does AI detection work in dim battery rooms?
Yes, but accuracy depends on lighting, camera angle and how much racks or cable trays block the view.
4. How does viAct protect privacy in data centre monitoring?
viAct automatically masks faces in footage, encrypts video end to end and offers On-Prem deployment, so video is processed inside the facility rather than sent elsewhere.
5. How long does it take to set up viAct AI on existing data centre cameras?
A single site can typically go live in 3 to 7 days, and multi-site rollouts take 4 to 8 weeks.
Read More:
