Lead-acid batteries fail politely. A VRLA string at end of life swells, leaks, loses capacity and eventually fails to support the load. It is an availability problem, and the facility team finds it during a scheduled impedance test.
Lithium-ion fails differently. Most of the time it also simply degrades, but it has one failure mode that lead-acid does not: a cell can enter a self-sustaining reaction that generates its own heat faster than any cooling system can remove it. That is thermal runaway, and it is the reason the battery room is now a fire engineering problem rather than an electrical one.
This matters more every year because lithium has won the argument on every other metric. It is roughly a third of the weight for the same energy, tolerates higher ambient temperature, charges far faster after a discharge event, and lasts two to three times as long. For a GPU hall where floor loading and recharge time both matter, the case is overwhelming. The question is no longer whether to use it but how to contain the one thing it does badly.
What Actually Happens Inside the Cell
A lithium-ion cell stores energy by moving lithium ions between electrodes through a flammable organic electrolyte. A thin polymer separator keeps the electrodes apart. Everything depends on that separator staying intact.
Thermal runaway begins when a cell reaches a temperature at which the solid electrolyte interphase layer starts to decompose, typically somewhere around 80 to 120 degrees Celsius depending on chemistry. That decomposition is exothermic. The heat it releases pushes the cell hotter, which accelerates the decomposition, which releases more heat.
Past roughly 130 degrees the separator begins to soften and shrink. When it fails, the electrodes touch and the cell short-circuits internally, dumping its stored energy as heat in seconds. The electrolyte vaporises, pressure rises, and the cell vents. What comes out is a mixture of flammable hydrocarbons, carbon monoxide, hydrogen and, in fluorinated chemistries, hydrogen fluoride.
The critical point for facility design is what happens next. A single cell in runaway releases enough energy to push its immediate neighbours past their own initiation temperature. Without a barrier, the event propagates cell to cell, then module to module. Propagation is the difference between a damaged battery unit and a lost building.
Chemistry Changes the Risk Profile
Not all lithium is the same, and the distinction is the single most important specification decision in the battery room.
Lithium iron phosphate has a markedly higher thermal runaway onset temperature and releases less energy when it does run away. It is less energy dense, which means a bigger footprint for the same backup time, and that trade is almost always worth making in a stationary application where space is less constrained than it is in a vehicle.
Nickel manganese cobalt chemistry is denser and more common in modules designed for the automotive supply chain. It initiates runaway at a lower temperature and releases considerably more energy. It is not unsafe, but it demands more from the containment design.
If a vendor proposal does not clearly state the cell chemistry, that is the first question to ask. The downstream consequences for room construction, separation distances and suppression are entirely different.
What the Codes Require
NFPA 855 is the reference standard for stationary energy storage and underpins most jurisdictions' requirements. The provisions that drive design are the limit on stored energy per fire area, minimum separation between battery units and between units and walls, fire-rated construction for dedicated rooms, explosion control or deflagration venting sized for the gas volume a worst-case event would produce, gas detection, and a documented commissioning and emergency response plan.
In the UAE, the local civil defence authority is the authority having jurisdiction, and approval is required at design stage rather than after installation. Practically, this means the battery room layout, the ventilation strategy and the suppression design all need sign-off before equipment is ordered. Retrofitting separation into a room that was already built is expensive and sometimes impossible.
Two details catch people out. First, the stored energy limit is per fire area, not per room, so two rooms separated by a non-rated partition may be treated as one. Second, the required separation applies between units from different manufacturers' listings as well as between units generally, which constrains mixing equipment.
Detection Is Where the Time Is
By the time a conventional smoke detector responds to a battery event, propagation is usually already under way. The useful warning comes earlier, and from different instruments.
Cell-Level Monitoring
Any credible lithium system includes a battery management system monitoring voltage and temperature at cell or module level. What matters operationally is whether those readings leave the battery cabinet. A BMS that alarms locally and reports a single summary status to the building management system is giving you almost nothing. Per-cell data trended over time reveals the outlier cell weeks before it becomes a problem, and a cell drifting away from its siblings in voltage or temperature is the clearest early signal available.
Off-Gas Detection
Cells vent electrolyte vapour before they produce smoke or flame, and the gap is often several minutes. A detector calibrated for the specific solvents used in the electrolyte, rather than a general combustible gas sensor, can alarm during that window. That is enough time to transfer load, shut down charging and get people out, which is the realistic objective.
Thermal Imaging on the Round
Periodic thermal survey of terminations and cell faces catches developing hot spots and loose connections that create them. It is not a substitute for continuous monitoring but it finds the mechanical problems that continuous monitoring misses. Sites already running predictive maintenance and autonomous operations can fold it into an existing round.
Containment, Not Extinguishment
The instinct is to ask which suppression agent puts out a lithium fire. It is the wrong question. Thermal runaway is driven by a reaction inside sealed cells, and no agent discharged into the room reaches it. A clean agent will knock down flame in the space and protect everything else in the room, then the pack will continue to generate heat and can reignite.
What actually limits damage is physical.
Separation. Distance between units, and barriers rated to resist the heat flux of an adjacent unit in full runaway, stop propagation. This is the single most effective measure available and it is decided at layout stage.
Water, in volume. Where codes permit it, a sprinkler system applying water at the density NFPA 855 specifies cools surrounding units enough to prevent them initiating. It does not stop the affected unit. It stops the next one.
Deflagration venting. Vented gases are flammable and will accumulate. Explosion control, whether by deflagration venting to outside or by mechanical exhaust sized for the event, prevents a gas explosion that would do far more structural damage than the fire.
Letting it finish. Fire service guidance for lithium installations increasingly accepts controlled burnout as the correct outcome for the affected unit, with resources directed at cooling neighbours. Designing on that assumption produces a better room than designing around the hope of extinguishment.
Choosing between agents still matters for the rest of the room, and the trade-offs are covered in our guide to clean agent fire suppression systems.
Operational Practices That Matter
Keep the battery room at the temperature the cells were specified for. Every sustained ten degrees above specification roughly halves calendar life and moves cells closer to initiation. A battery room running warm because it shares cooling with a dense hall is a slow-motion problem.
Do not mix ages or chemistries in a string. Replacement modules from a different production batch, let alone a different manufacturer, create imbalance that the management system then has to work around continuously.
Treat a cell that has been overcharged, deeply discharged or physically damaged as compromised even if it tests normally afterward. Internal damage from those events does not always show up in capacity or impedance tests, and the cell may initiate later under normal load.
Write the emergency response plan with the local fire service rather than for them, and make sure the plan says who isolates charging, what the evacuation distance is, and that nobody re-enters until gas readings are clear. Hydrogen fluoride exposure from a vented cell is a genuine hazard and it is invisible.
What This Means for Colocation Customers
If you are buying hosting rather than building, battery design is somebody else's problem until it is not. Three questions are worth asking any prospective provider: what cell chemistry is in the UPS batteries, how are the battery units separated and what is the suppression strategy, and does the battery management system expose per-cell data to the operations team or only a summary alarm.
A provider who answers those crisply has thought about it. One who does not know the chemistry has outsourced the question to a vendor and inherited whatever that vendor proposed. The same reasoning applies to the broader electrical design, which our guide to N+1 and 2N power redundancy configurations covers in more detail.
Rax Data and Energy designs its battery rooms to the separation and detection standards described here, with lithium iron phosphate chemistry and cell-level monitoring exposed to our operations team. If you want to walk through how that is implemented across our UAE facilities, talk to us.
Frequently Asked Questions
What is thermal runaway in a lithium battery?
Thermal runaway is a self-sustaining chain reaction inside a lithium-ion cell. Heat triggers exothermic breakdown of the cell chemistry, which releases more heat, which accelerates the breakdown. Once it starts the cell cannot be cooled fast enough to stop it, and the energy released can propagate to neighbouring cells.
Is lithium-ion safe to use in a data center UPS?
Yes, and it is now the mainstream choice. Modern lithium iron phosphate chemistry is considerably more tolerant than the cobalt-based chemistry used in consumer devices, and a properly specified system includes a battery management system, cell-level monitoring and a containment strategy. The risk is managed by design, not eliminated by chemistry alone.
What does NFPA 855 require for data center batteries?
NFPA 855 sets requirements for stationary energy storage, including maximum stored energy per fire area, minimum separation between battery units, dedicated rooms or enclosures with rated construction, explosion control or deflagration venting, gas detection, and a commissioning and emergency response plan. Local authorities having jurisdiction may apply it with amendments.
How do you detect thermal runaway before it spreads?
Early detection relies on cell-level voltage and temperature monitoring through the battery management system, combined with off-gas detection. Cells vent electrolyte vapour minutes before visible smoke or flame, so a gas sensor calibrated for electrolyte solvents gives considerably more warning than a conventional smoke detector.
Can a clean agent system put out a lithium battery fire?
A clean agent will suppress flame but it does not remove the heat driving the reaction inside the cells. Once runaway is established the pack continues generating heat and can reignite after the agent disperses. Clean agents protect the room and its other equipment; containing the battery event itself depends on separation, cooling and letting the affected unit burn out safely.
A Real Energisation Date, Not a Brochure
Rax operates where transmission investment has run ahead of load, and we are direct about staged capacity and real dates. Talk to our team about sizing a deployment against a timeline.
Contact Rax Data & Energy