AI-ERA DATA CENTERS IN SOUTHEAST ASIA – PART 5 OF 6 The Fire Marshal in the Boardroom

James Soh

Two facilities, one decision. The first is a data center in the region that has been running reliably for fifteen years. VRLA batteries in a dedicated room, approved under the codes of its era, maintained by a team that knows every string and knows what a bad cell smells like before the alarm triggers. An AI tenant is now asking about in-hall ESS. The project team has responded with a procurement document. It says “Li-ion.”

The second facility is at concept design stage: a new AI campus in Johor, Jakarta, or the EEC, fifty pages into the Owner’s Project Requirements, with a UPS and battery line item in the capex schedule that also says “Li-ion.”

Neither team has made a battery decision. They have deferred one. Li-ion is a family of chemistries with meaningfully different safety profiles, suppression requirements, regulatory classifications, and operational demands. Specifying “Li-ion” without specifying which chemistry is the equivalent of specifying “steel” without specifying grade, and discovering the difference at fabrication. For the established facility, the transition from VRLA to the right lithium chemistry is not a product swap. It is a change of regime: regulatory, structural, operational, and governance. For the to-be-designed facility, the chemistry decision made at the 30-percent design stage will govern AHJ submissions, suppression design, iBMS specification, and operations procedures for the next fifteen years. In both cases, the fire engineering consequences of that decision are already in the room, whether or not the fire marshal has been invited to the meeting yet.

What was a framing scenario when this series began is now documented fact. In September 2024, a lithium-ion battery explosion at Digital Realty’s SIN11 data centre in Loyang, Singapore triggered more than 36 hours of firefighting operations, grounded Alibaba Cloud’s Singapore Availability Zone C, and required SCDF to deploy an unmanned firefighting robot. Six months later, in March 2025, a second Singapore data centre fire at Chai Chee sent one person to hospital for smoke inhalation. Both incidents are examined in Section 2. They are not cautionary analogies from other markets. They are the argument made visible, in Singapore, in the period since this series was drafted.

———

1. Design and procurement: where Design for Safety and Design for Operations converge

The Li-ion family: why the distinction matters for data centers

Li-ion is a family, not a chemistry. NMC and NCA are the high-energy chemistries familiar from automotive applications: higher energy density but requiring tighter Battery Management System (BMS) control and protection to manage fire risk. Lithium Iron Phosphate (LFP) is the specific chemistry that lines up for facility-scale Energy Storage Systems (ESS): better thermal and structural stability, a much lower propensity for thermal runaway, and a cycle life of 6,000 to 12,000 cycles at 80% depth of discharge compared with 300 to 500 cycles for VRLA. That cycle life supports a 10- to 15-year asset life in UPS and ESS applications, improving lifetime cost even where the upfront capex is higher than VRLA. For Southeast Asia data centers specifying UPS and short-duration ESS today, LFP is the right default, and increasingly the chemistry that regional Authorities Having Jurisdiction (AHJ) are calibrating their guidance toward.

For the existing VRLA facility, this distinction matters because the AHJ classification, suppression requirements, and compartmentation standards for LFP are different from those that governed the original approval. The transition is not a documentation update. It is a fresh regulatory engagement.

For the to-be-designed facility, the distinction matters because a procurement document that says “Li-ion” without specifying LFP has not closed the chemistry decision. It has left it open to be resolved downstream, by someone without the authority or context to understand its consequences. That downstream resolution has consequences that run well beyond the battery room.

Balancing client requirements, fire prevention, and suppression

The chemistry choice determines the suppression system. The suppression system determines the structural requirements: drainage, slab loading, compartment sizing. The structural requirements determine which halls can host which tenants and at what ESS density. These are not three separate workstreams. They are one integrated design decision that needs to be resolved before concept design is frozen, in both the existing facility retrofit and the new build.

For the existing facility, the question is whether the current compartmentation and suppression can support LFP, and if not, what structural work is required and what that means for the AHJ re-engagement timeline. For the to-be-designed facility, the question is whether the base-build assumptions will support not just LFP today but hybrid ESS configurations as they become commercially viable. Both questions have a monitoring dimension that the chemistry decision alone does not resolve.

The intelligent BMS as a procurement decision

Specifying intelligent BMS (iBMS) capability at procurement, not retrofitting it later, is one of the clearest distinctions between a facility that will govern its battery strategy well and one that will manage reactively. A well-specified iBMS covers cell-level monitoring, early thermal anomaly detection, integration with the facility BMS and Data Center Infrastructure Management (DCIM) platform, and alert workflows that reach the right people before an event escalates.

For the existing facility, the VRLA-era BMS almost certainly cannot integrate with a modern iBMS without significant work. The LFP transition forces this decision. Deferring it means inheriting the monitoring gap for another decade, and managing LFP with the visibility tools designed for a different chemistry. The iBMS gap and the AHJ gap are related: both surface at the same point in the project, and both need to be resolved before procurement is signed.

AHJ engagement at the design stage

What to bring to a pre-application meeting. How to frame LFP and ESS configurations for an authority with limited prior exposure to high-density AI campuses. The difference between an authority that has seen your layout and chemistry and given preliminary comfort in writing, and one that sees it for the first time at inspection. Early, transparent engagement, before the 30-percent design stage for new builds and before procurement is signed for existing facilities, is how operators avoid the rescheduled inspection and the declined certification that opened Part 4.

One material development since this series began: the 2026 edition of NFPA 855, the primary international reference standard for stationary ESS installations, has been published with significant updates from the 2023 version. The new edition introduces requirements for Thermal Runaway Propagation Protection (TRPP) systems, updated explosion control and prevention provisions, new requirements for project emergency response plans, and large-scale fire testing that goes beyond prior editions. AHJs in the region who reference NFPA 855 are now working from a materially more demanding standard than the one in place when many current-generation campus designs were started. Operators engaging AHJs in 2026 should clarify which edition the local authority is referencing and ensure their design team is aligned to the current version, not the prior one.

In Data Center Primer I frame this as the point where Design for Safety and Design for Operations stop being parallel lenses and become the same decision. The consequence runs further than most project teams anticipate: the operations team inherits all of it, the labelling, the alert thresholds, the drill scenarios, the replacement cycles, without having been in the room when any of those choices were made. Getting the chemistry decision right at procurement, for both the established facility and the new build, is how operators avoid spending the next decade managing the consequences of a decision that was made too low in the organisation and too late in the design process.

The design and procurement decisions establish the regime. What happens next depends entirely on whether the operations team that inherits that regime has been prepared for it, not for the regime they have been working in, but for the one they are about to enter.

———

2. Operations: diligence as the daily discipline

The regime change for the established facility team

A team that has spent years working with VRLA has internalised a specific set of responses, risk instincts, and procedural habits. LFP behaves differently in a fault condition. Thermal runaway in LFP has a different signature, a different progression, and a different first-response requirement. The procedures, labelling, escalation paths, and drills that served well under VRLA are not wrong. They are calibrated to the wrong chemistry. Recognising that gap, and closing it deliberately, is the first operational task of the LFP transition.

For the to-be-designed facility the same point applies in a different direction. The operations team that will inherit the new design is almost certainly coming from a VRLA background. The procedures they are handed on day one need to be written for LFP from the start, not adapted from the previous generation.

Diligent use of iBMS

The iBMS is only as valuable as the discipline applied to it. What diligent use looks like in practice: regular review rhythms that bring operations and asset management together, alert response that follows defined thresholds rather than judgment calls, and module replacement that is driven by data rather than deferred because the schedule is tight. The gap between a facility that has an iBMS and a facility that uses it seriously is where battery events accumulate.

Chapters 8 and 9 of Data Center Primer go into the procedural detail that this article can only summarise: the SOPs, EOPs, maintenance records, and shift handover disciplines that underpin diligent operations practice. The reason that detail matters here is not completeness. It is that an iBMS alert ignored on a Tuesday night, or a module replacement deferred because the schedule was tight, is how a well-designed facility accumulates the conditions for a serious event. The hardware does not make the facility safe. The people following the procedures do. For a team transitioning from VRLA to LFP, those procedures are not an update to the existing manual. They are a new manual, and the operations leadership that treats them as such will run a materially safer facility than one that assumes the old instincts transfer.

Fire drills for ESS events: the immersion test

Fire drills for battery ESS events are different from conventional fire drills in ways that most VRLA-trained teams have never encountered. The thermal runaway scenario for LFP requires specific first-response procedures, including immersion of battery modules in water as a containment measure, that demand both physical preparation and practised execution under pressure. Research published in February 2026 confirmed that LFP batteries in failure scenarios can produce hydrogen fluoride (HF) concentrations of 3,000 to 8,000 ppm. First-response procedures for LFP events must address toxic gas exposure, not only thermal containment. The drill is a governance signal as much as a training exercise: if the operations team cannot execute it confidently, the fire engineering design has not been translated into operational reality. For the existing facility transitioning to LFP, running this drill before the first LFP module is commissioned, not after, is the standard that mature operators set for themselves.

The three a.m. test

SOPs and EOPs that differentiate between chemistries and layouts. Escalation paths and authority on shift. The window between a manageable battery event and an evacuation is short. For LFP in an in-hall ESS configuration, it may be shorter than the team expects if they have calibrated their instincts on VRLA. Procedures and people need to be matched to the specific configuration of the facility. If the ESS configuration cannot be explained in a ten-minute shift handover, it is too complex to govern safely, from the operations floor or from the boardroom.

What that test looks like when the procedures are not matched to the configuration: Singapore, September 10, 2024.

At 7.45am, a fire alarm triggered at Digital Realty’s SIN11 data centre at 3 Loyang Way. By 8.15am, all twenty on-site personnel had evacuated. Alibaba Cloud confirmed the cause as an explosion of lithium batteries, which led to fire and elevated temperatures across its Singapore Availability Zone C. SCDF characterised the operation on arrival as likely prolonged. Four water jets were deployed. The sprinkler system activated. Water accumulated on the third floor, creating secondary short-circuit risk. Digital Realty shut down electrical systems on the advice of its Licensed Electrical Worker and SCDF. SCDF deployed an unmanned firefighting robot to cool the batteries, because the reignition and toxic gas risk made sustained manual proximity operations unsafe. Firefighting operations continued for more than 36 hours. Alibaba Cloud’s Availability Zone C remained impaired for days, with some hardware requiring careful drying before it could be safely accessed.

The facility opened in 2016, before SCDF’s June 2020 standard requiring ESS installations to be on the ground floor. The battery rooms were on the third floor of a four-storey building. The installation was pre-standard. The risk it carried was not.

Six months later, at approximately 6am on March 14, 2025, a second data centre fire occurred at 750C Chai Chee Road in Singapore. Operators reported an explosion. The fire involved a server rack, was extinguished with dry powder, and the building suppression system activated. One person was taken to Singapore General Hospital for smoke inhalation. Power to the building was shut off.

Two incidents. Two sites. Six months apart. Different scales, different operators, same signature: battery event, explosion, suppression activation, power shutdown, service disruption, casualty. That is not coincidence. It is a frequency that every established facility in Singapore and across the region should be reading as a direct signal about the gap between the configuration being managed and the procedures in place to manage it.

There is a data point that applies regardless of chemistry and regardless of market: a review of 35 documented large-scale BESS fire incidents in the United States found that nearly half of incidents where system age was known occurred within the first six months of operation, during commissioning and initial operations phases. The early-life risk is real. The procedures and drills must be in place before the first module is energised, not after the first event.

LFP will not be the last chemistry decision either facility makes. The discipline established for the VRLA-to-LFP transition, chemistry evaluation, AHJ engagement, suppression implications, iBMS specification, operations retraining, is the same discipline that applies to every technology decision that follows. The question is whether the organisation has built that capability once and can apply it again, or whether it will rediscover the hard way that the next chemistry change is also a regime change.

———

3. New battery technology: monitoring, evaluation, and timely replacement

The principle: engineering and financial sense, not novelty

The governance discipline established for the VRLA-to-LFP transition is the same discipline that applies to every subsequent technology decision. Chemistry evaluation against AHJ requirements, suppression and structural implications, iBMS compatibility, operations team retraining: these are not one-time exercises. They are the repeatable capability that allows an organisation to evaluate what comes next without being surprised by it. The question is not whether to monitor emerging chemistries. It is whether the organisation has built the institutional muscle to act on that monitoring at the right level.

The emerging landscape: a working map for Southeast Asia

LFP is the current default for facility-scale UPS and short-duration ESS, the right choice now for both facility types, and the chemistry against which all others should currently be benchmarked. Its position as the regional default is reinforced, not challenged, by the two Singapore incidents described above: both involved legacy lithium-ion installations predating LFP adoption, not modern LFP-specified facilities operating to current standards.

Sodium-ion has moved from a technology to monitor to a technology to actively evaluate. When this series was drafted, sodium-ion was an emerging chemistry being positioned for hybrid systems in large AI campuses. By early 2026 it has entered commercial deployment for AI data center applications. Peak Energy shipped the first grid-scale sodium-ion BESS in the United States in August 2025. Energy Vault announced a dedicated storage architecture for AI Neoclouds and AI-first data center operators using sodium-ion batteries in February 2026, citing higher safety and reliability relative to conventional BESS. MIT Technology Review named the sodium-ion battery one of its 10 Breakthrough Technologies for 2026, referencing data center backup power and utility-scale storage directly. CATL confirmed at its December 2025 supplier conference that sodium-ion and lithium-ion are expected to develop in parallel across multiple sectors from 2026.

For Southeast Asia, the most relevant development is the Hithium hybrid architecture: a system combining long-duration lithium storage with high-rate sodium-ion technology to cover both grid stability and instantaneous power response for AI data centers, with commercial deliveries scheduled for Q4 2026. The sodium-ion component carries a distinct safety profile relative to LFP: higher thermal runaway onset temperatures (220 to 260 degrees Celsius versus 170 to 220 degrees for NMC-based lithium-ion), lower heat release rates, and reduced hydrogen content in off-gases. Those characteristics are directly relevant to AHJ submissions and suppression design for high-density in-hall ESS configurations.

The practical instruction for operators in the design stage today: sodium-ion is no longer a horizon item. It is a near-term procurement consideration for hybrid architectures. The ESS architecture decisions made at the 30-percent design stage should explicitly accommodate hybrid LFP and sodium-ion configurations, not lock to a single chemistry. The design pattern that locked the previous generation into VRLA for a decade is the risk to avoid repeating.

LTO and high-rate Li-ion variants offer extreme cycle life and high C-rates for specialised UPS or grid-support roles. Higher cost keeps them out of mainstream hall-scale backup. Relevant for specific tenant requirements but not a default consideration for most Southeast Asia facilities.

Solid-state offers better energy density and no liquid electrolyte, eliminating thermal runaway in the conventional sense. Manufacturing cost and scalability constraints keep it off any credible near-term procurement shortlist in the region. Keep on the radar, not on the specification.

The replacement decision as governance

The replacement decision belongs at the same governance level as the original procurement. Who decides, on what basis, and how that decision is documented and communicated to the AHJ, insurers, and tenants are not questions to resolve on the operations floor. The proactive withdrawal logic from Part 4 applies here at a technology level rather than just an individual battery condition level: an organisation that has defined its withdrawal triggers for ageing VRLA strings should be equally clear about the conditions under which it would supplement or replace LFP with a successor chemistry. Culturally, proactive technology evaluation must be seen as good asset stewardship, not an admission that the last decision was wrong. The organisation that treats it as the former will stay ahead of regulatory and market expectations. The one that treats it as the latter will fall behind both.

The chemistry landscape will keep moving. What does not change is the governance discipline required to evaluate it, act on it at the right level, and communicate the consequences to the AHJ, insurers, tenants, and operations teams who will live with the result. The questions below are designed to test whether that discipline exists in your organisation today.

———

Four questions worth asking your teams this week

Chemistry and procurement: “Have we specified LFP explicitly in our UPS and ESS procurement, for both new builds and existing facility transitions, and does that specification include the iBMS integration requirements our operations team will need?”

AHJ and insurer engagement: “For each facility transitioning to LFP or planning in-hall ESS, have local authorities and insurers seen the layout and chemistry, and do we have their preliminary position in writing before procurement is signed? Are we referencing the 2026 edition of NFPA 855 in those conversations?”

Operations readiness: “Have our operations teams been retrained specifically for LFP, including thermal runaway procedures, toxic gas protocols for HF exposure, and the immersion drill, and when was that training last tested under realistic conditions?”

Technology horizon: “Do we have a defined process for evaluating emerging chemistries against our hall configurations and tenant mix, and does our new-build ESS architecture accommodate hybrid configurations, including sodium-ion, as they become commercially viable in the region?”

If those questions cannot be answered clearly and consistently across your portfolio, the organisation has more work to do, regardless of how long the facility has been running or how recently the concept design was frozen.

———

Looking ahead to Part 6

Parts 1 through 5 have examined the AI-era facility one domain at a time: the speculative build dilemma, operations and talent, the client’s view, battery governance, and fire engineering. Each part has shown how a decision made in one domain creates consequences in another. Part 6 takes the integrated view: what all of these decisions look like when they compound across a 15- to 20-year asset life, and what the operators and investors who get this right will own that those who do not will not.

The detailed frameworks behind what this series covers are in Data Center Primer (ISBN 9789819439768).