Data Centre Insurance UK 2026: Uptime, Fire & Liability
General information — not advice. This guide explains how insurance for data centres generally works. It is not personalised advice or a recommendation, and it does not take account of your own circumstances. Policy wordings differ between insurers, so always check your own documents and speak to a broker before acting. Regulatory status is set out at the end of this guide.
What is data centre insurance and why is it not just property cover?
Data centre insurance covers the facility, the plant that keeps it running, and — critically — the consequences of it stopping. It differs from standard commercial property insurance in three ways: the business interruption exposure usually exceeds the property damage by an order of magnitude; the operator holds third-party equipment in care, custody and control, which standard liability excludes; and the severity is driven by plant failure and lithium-ion fire rather than by conventional perils. Since September 2024 UK data centres are also Critical National Infrastructure, which raises the resilience bar.
Key facts at a glance
- UK data centres were formally designated Critical National Infrastructure in September 2024 — the first CNI designation in almost a decade, placing them alongside energy, water and emergency services.
- The Cyber Security and Resilience (Network and Information Systems) Bill, introduced in November 2025, is expected to receive Royal Assent later in 2026. It amends the NIS Regulations 2018 to make data centre services an essential service in a new data infrastructure subsector, with Ofcom as operational regulator.
- Indicative scope thresholds are 1MW of capacity per site, or 10MW across multiple locations. Above these, operators will face statutory duties on security measures, incident reporting and regulator engagement, with commencement following secondary legislation and DSIT Codes of Practice.
- Business interruption, not property damage, is the dominant loss. A UPS or switchgear fire may cause modest structural damage while taking a hall offline for months — and the lead time on replacement transformers and switchgear now runs well beyond a standard twelve-month indemnity period.
- Lithium-ion UPS has displaced VRLA in most new build. It is lighter, smaller and longer-lived, and it introduces thermal runaway — a self-accelerating cell reaction that propagates and is not extinguished by the gaseous suppression most halls were designed around.
- Customers' servers are third-party property in your care, custody and control, which is excluded by standard public liability. Colocation operators need that exposure addressed explicitly, and the SLA credit regime is not a substitute for it.
- AI workloads are driving direct liquid cooling into halls designed for air. That puts a pressurised water circuit directly over high-value compute — a fundamentally different escape-of-water profile from a CRAC unit in an aisle.
A data centre is a building whose entire economic purpose is not stopping. That single fact reorganises the insurance around it. In an ordinary commercial property risk, the building and its contents are the value and business interruption is a secondary consideration. In a data centre the physical shell is often the least valuable component, the mechanical and electrical plant is where the capital sits, and the loss that matters is measured in hours of downtime multiplied by the obligations owed to every customer in the hall.
This creates a placement problem. General commercial property insurers can rate a warehouse. Rating a Tier III colocation facility requires understanding N+1 redundancy, concurrent maintainability, UPS chemistry, suppression design, and what a service level agreement actually obliges the operator to pay when uptime is missed. Relatively few markets do that well, and the gap between a policy that looks adequate and one that responds properly is usually invisible until an incident.
This guide is written for colocation providers, enterprise operators, edge facility owners and the developers building new capacity. It sits in our AI and tech insights hub and connects closely to cyber insurance and commercial property insurance, both of which form part of a complete data centre programme.
How does the Insurability Framework apply to a data centre?
Data centres are high-value, low-frequency, extreme-severity risks placed into a market with limited genuine appetite and a lot of polite interest. The difference between a competitive programme and an expensive one is almost entirely the quality of the risk presentation — which is what The Insurability Framework™ exists to build.
The Miller & Partner Method
Four pillars applied to every data centre placement. The objective is to present a documented resilience story — redundancy, suppression, maintenance regime, single points of failure — because underwriters in this class price the engineering, not the square footage.
Underwriter Intelligence
A handful of markets genuinely write data centre risk, and their appetites diverge sharply on UPS chemistry, suppression type, single-tenant concentration and whether liquid cooling has been retrofitted. Knowing which underwriter needs which evidence before submission is the difference between a structured programme and a stack of partial quotes.
Difficult Risk Expertise
Retrofitted lithium UPS in a hall designed for VRLA, direct liquid cooling added to an air-cooled hall, converted industrial buildings, single-customer concentration, flood-zone or high-water-table sites, and operators with a prior outage claim are all placements general property markets decline or price defensively. These are ordinary submissions here.
Risk Assessment
The uninsured losses here are structural rather than accidental. Indemnity periods set at twelve months against transformer lead times measured in years. Customers' equipment sitting in a custody exclusion. SLA credits treated as an insured loss when they are a contractual one. Contingent exposure to a single grid connection or fuel supplier. We identify these before renewal.
Claims Advocacy
A data centre incident runs as a technical investigation, a customer relations exercise, a regulatory notification and an insurance claim simultaneously. We manage the insurer and adjuster relationship — including the forensic engineering that determines whether a failure was sudden and unforeseen or a maintenance issue — while the operator deals with customers and, increasingly, the regulator.
What does Critical National Infrastructure status change?
It converts resilience from a commercial choice into a regulatory expectation. CNI designation in September 2024 signalled that disruption of a major UK data centre is treated as a national security event. The Cyber Security and Resilience Bill turns that signal into statute — bringing in-scope operators under the NIS Regulations with duties to register with Ofcom, implement appropriate technical and organisational security measures, and report significant incidents, backed by significantly increased penalties.
The practical significance for insurance is threefold, and none of it is obvious from the headlines.
First, the evidence base improves. Operators preparing for the regime will be documenting resilience, incident response and supply chain security to a standard they may not have held before. That documentation is exactly what a specialist underwriter wants to see, so compliance work has a direct rating benefit — one of the few places where regulation reduces insurance cost rather than raising it.
Second, the notification landscape gets more crowded. An incident may trigger a regulatory report to Ofcom, contractual notifications to every affected customer, a possible ICO notification where personal data is involved, and an insurance claim — on different timescales and to different standards. Answering them inconsistently is a genuine risk, and it is a reason to coordinate the regulatory and insurance responses from the first hour rather than sequentially.
Third, penalties are not insurable. Regulatory fines under the expanded regime, like every other penalty discussed in our guides, cannot be insured under any UK policy. Legal expenses and specialist regulatory response cover can fund the defence and the investigation; the fine itself always lands on the business.
Why are lithium-ion UPS batteries the dominant fire exposure?
Because thermal runaway behaves unlike any fire the building was designed for. A lithium cell in runaway generates its own oxygen and heat, propagates to adjacent cells, and cannot be reliably extinguished by the gaseous suppression most halls use — inert gas and clean agents work by displacing oxygen or absorbing heat, and a runaway reaction supplies both. It also produces flammable and toxic off-gassing that can create an explosion risk in an enclosed battery room.
The industry shift from valve-regulated lead acid to lithium-ion in UPS installations is close to complete in new build, and for sound reasons: significantly smaller footprint, lower weight, longer service life, faster recharge and better cycling. It is a genuine operational improvement. It is also a materially different fire risk in the same physical space, and a large number of facilities have retrofitted lithium into rooms whose fire strategy was written for lead acid.
What underwriters examine, in rough order of weight:
- Separation and compartmentation. Whether the battery room is a dedicated fire-rated compartment away from the white space, and what the rating actually is. This is the single most influential feature.
- Suppression appropriate to lithium rather than inherited from the previous chemistry. Water mist, and increasingly full water-based approaches, are treated more favourably for lithium than gaseous systems on their own.
- Off-gas detection. Early detection of electrolyte vapour before thermal runaway is established gives the only meaningful intervention window, and its presence changes the loss estimate significantly.
- Explosion venting or deflagration management in the battery room, given the flammable off-gas accumulation risk.
- Battery management system quality and monitoring, including cell-level telemetry and alarm escalation to a staffed position out of hours.
- Cell quality and provenance — tier-one manufacturers with documented testing attract materially better terms than commodity cells, exactly as they do in the battery energy storage market. Our battery storage facility insurance guide covers the thermal runaway underwriting question in depth and the analysis carries directly across.
What happens when cooling fails — and what does liquid cooling change?
Cooling failure is the fastest route to an outage: high-density halls reach damaging temperatures within minutes, not hours, so a chiller or CRAC failure with inadequate redundancy causes thermal shutdown before an engineer can attend. Direct liquid cooling for AI and GPU workloads changes the risk profile fundamentally by placing a pressurised water circuit directly over high-value compute — an escape-of-water exposure inside the rack rather than in the aisle.
Air-cooled halls have a familiar failure mode. Chilled water plant, CRAC or CRAH units, and the associated pumps and pipework fail; temperature rises; equipment throttles then shuts down to protect itself. The insurance questions are redundancy configuration, whether cooling is on UPS or generator-backed supply, and how quickly the thermal ride-through window closes.
Liquid cooling is the current structural change and it is being driven hard by AI compute. GPU-dense racks cannot be air-cooled economically at the power densities now being deployed, so direct-to-chip liquid cooling is being installed — including as a retrofit into halls that were never designed for it. From an underwriting standpoint this creates several new questions:
- Leak detection and automatic isolation at rack and manifold level, with response fast enough to matter given what sits beneath.
- Coolant distribution unit redundancy, and what happens to the load when a CDU is taken out of service.
- Whether the escape of water exclusion or sub-limit in the property section contemplates water inside the rack. Many wordings were drafted when the only water in a hall was in a pipe overhead, and the sub-limit reflects that.
- Mixed-mode halls where air and liquid cooled racks share a space, and the failure of one system affects the thermal envelope of the other.
- Retrofit quality — whether the installation was designed by the original M&E consultant or added later, which is a recurring theme in leak claims.
The 2026 regulatory backdrop is pushing the same direction: Climate Change Agreement efficiency targets and the energy-density gap created by GPU clusters make liquid cooling close to unavoidable at the high end. Operators should expect it to become a standard underwriting question rather than an unusual one.
Who is liable for a colocation customer's servers?
You are, to the extent your contract says so — and standard public liability will not respond, because customers' equipment is third-party property in your care, custody and control, which is a standard exclusion. Colocation contracts normally limit liability heavily and exclude consequential loss, with SLA credits as the agreed remedy. But credits are a contractual cost you pay, not an insured loss, and a limitation clause is only as good as its enforceability.
This is where colocation operators most often find a gap they did not know they had, and it has three separate components worth untangling.
The customers' equipment itself. Servers, storage and network kit in your halls belong to your customers. If a suppression discharge, a leak or a fire damages them, the claim against you sits outside standard public liability because of the custody exclusion. Whether you need to insure it depends on what your contract says — most colocation agreements place the risk of physical loss on the customer and require them to insure their own equipment, which is the right allocation. But it needs to be actually written that way, and checked against what your policy assumes.
SLA credits. Almost every colocation contract has an uptime commitment backed by service credits. These are a contractual price adjustment, not damages, and they are not an insured loss under a standard business interruption section — your own lost revenue is insurable; the discount you owe your customer generally is not. On a large facility with a multi-hour outage across many customers, the aggregate credit exposure is material and it should be modelled rather than discovered.
Consequential loss claims that break the cap. Limitation clauses generally hold, but they are tested when a customer's loss is very large, and enforceability turns on drafting, incorporation and reasonableness. The realistic exposure is not that every cap fails, but that one significant customer with a catastrophic loss litigates the point. That is a legal defence cost exposure whatever the outcome.
The mirror image of this is worth noting for completeness: your customers face their own dependency exposure on you, which is what contingent business interruption insurance addresses, and it is increasingly requested during procurement. Understanding both sides of that conversation is commercially useful when tendering.
How should business interruption be structured for a data centre?
Around plant replacement lead times, not around building reinstatement. The controlling constraint after a serious electrical incident is rarely construction — it is the delivery time for transformers, switchgear and UPS modules, which now runs to many months and in some cases beyond two years. A twelve-month indemnity period against that lead time leaves the tail uninsured no matter how generous the monetary limit looks.
Three structural decisions matter more than the headline sum:
- Indemnity period. Twenty-four months should be the starting assumption for any facility whose critical path runs through bespoke electrical plant, and thirty-six is defensible where the equipment is long-lead or non-standard. Model it from actual current supplier lead times, refreshed annually, rather than from a figure set when the policy was first written.
- Gross profit basis and customer churn. Revenue does not simply resume when the hall does. Colocation customers who migrate during an extended outage may not come back, and that revenue loss extends well past physical reinstatement. The indemnity period needs to cover the recovery of the customer base, not just the restoration of the facility.
- Denial of access and utility failure extensions. A data centre can be entirely undamaged and entirely useless — a grid failure, a fibre cut outside the boundary, or a cordon following a neighbouring incident all stop trading without touching your property. These extensions are frequently sub-limited to a level that bears no relation to the exposure.
Alongside these sit the standard property disciplines that matter more here than almost anywhere else, because of the concentration of value: sums insured need to reflect current replacement cost of M&E plant in an inflationary and supply-constrained market. Our guides to underinsurance and the condition of average and day one reinstatement set out the mechanisms, and business interruption insurance covers the calculation methodology in general terms.
Data centre insurance cover checker
Select the facility type that best matches yours to see the cover architecture we would normally build. For multi-site estates our commercial team can structure the whole programme on one submission.
Resilience and insurability self-check
Tick every item you could evidence today. This is the list we work through before taking a data centre to market — and much of it overlaps with what the incoming regulatory regime will expect.
- Documented redundancy configuration with single points of failure identified and recorded, not just an N+1 claim
- Load transfer and black-building tests actually performed, with reports including what failed
- Generator run and load-bank testing on schedule, with fuel quality testing and polishing records
- UPS chemistry declared to insurers, with any lithium retrofit notified at the time it was carried out
- Battery room in a dedicated fire-rated compartment with suppression appropriate to the chemistry in use
- Off-gas or very early smoke detection in battery and plant areas, alarming to a staffed position out of hours
- Leak detection and isolation on any liquid cooling, at rack and manifold level
- Planned preventive maintenance on all critical plant, with records and rectification evidence
- BI indemnity period modelled from current plant lead times, refreshed annually against supplier quotes
- SLA credit exposure modelled for a realistic multi-hour outage across the customer base
- Customer contracts reviewed against the policy — who insures customers' equipment, and what the liability cap says
- OT and building management systems segmented from IT, with access controlled and monitored
Data centre placement risk assessor
An indicative view of how your facility will be received in the market, and what we would prioritise in the submission. This is general information only — not personalised advice, not a recommendation and not a quotation.
How do cyber and physical risk converge in a data centre?
Through the operational technology that runs the building. Building management systems, DCIM platforms, chiller controllers, UPS management and access control are networked, often legacy, and frequently reachable from the corporate network or a vendor's remote support connection. A compromise of that OT layer produces a physical outcome — cooling stopped, power switched, doors unlocked — which is a property and business interruption loss arising from a cyber cause.
This is the exposure that falls between two policies, and it is where operators most often discover a gap that neither insurer thought they had written.
The problem is structural. Traditional property policies contain cyber exclusions, drafted to keep cyber-caused losses out of the property market. Cyber policies are built around data breach, business interruption from network outage and extortion, and are not always designed to respond to physical damage to plant. An attacker who manipulates a chiller controller to cause a thermal event, or who disables cooling entirely, causes a loss that both policies can plausibly disclaim.
What to establish, explicitly and in writing:
- Whether the property policy's cyber exclusion is absolute, or whether it carves back resulting physical damage from a non-malicious or even a malicious cyber cause. The wording varies significantly and the difference is enormous.
- Whether the cyber policy extends to physical damage and to OT systems, or is confined to data and network interruption.
- How the two policies interact on business interruption, and whether a single event could produce two claims with each insurer pointing at the other.
- Whether the cyber policy covers regulatory investigation costs under the expanded NIS regime, which will become more relevant as the Cyber Security and Resilience Bill commences.
Operationally, the controls that matter are segmentation of OT from IT, controlled and logged vendor remote access, patching regimes for building systems that were never designed to be patched, and monitoring that covers the BMS as well as the corporate estate. Our cyber insurance page covers the core product and the cyber insights hub covers the wider threat landscape.
What cover does a data centre construction or expansion need?
Construction all risks for the works, plus delay in start-up — which is usually the larger exposure. If commissioning slips, pre-let capacity earns nothing while financing costs continue, and the usual cause of the slip is long-lead electrical plant. Add transit cover for transformers, switchgear and UPS modules, and verify the M&E designer's professional indemnity rather than assuming it.
The UK is in a sustained data centre buildout, accelerated by AI compute demand and supported by planning reform — the Infrastructure Planning (Business or Commercial Projects) (Amendment) Regulations 2026 now allow qualifying schemes to be determined outside the standard local authority route. That means more operators are running construction risk alongside operational risk, frequently for the first time.
The specific features that distinguish data centre construction from ordinary commercial development:
- Value concentration in plant, not structure. A single transformer or switchgear lineup can represent a large share of the contract value, and it spends time in transit and in storage before installation. Transit and off-site storage cover need limits that reflect that.
- Testing and commissioning is the peak loss period. A disproportionate share of plant damage occurs when equipment is first energised and load tested. Confirm testing and commissioning is explicitly within the contract works cover rather than excluded or sub-limited.
- Expansion of a live facility. Building a new hall next to an operating one puts the revenue-generating asset within the construction risk envelope. Existing structures cover and a clear delineation of responsibility between the contractor's and the operator's policies are essential.
- Design liability sits with the M&E consultant. The most consequential errors in a data centre are design errors — undersized cooling, inadequate redundancy, a fire strategy that does not match the battery chemistry installed. Verify the designer's professional indemnity limit and that it is maintained after practical completion.
- Delay in start-up alignment. The DSU indemnity period and sum insured should reflect the financing structure and any pre-let commitments, not a generic figure.
Cyber risk during construction is an underrated element — project data, design documents and contractor systems are all targets, and our construction cyber insurance guide covers that specific intersection.
What drives the cost of data centre insurance in 2026?
Engineering evidence above everything else. The twelve factors below move the premium most, and the strongest single lever is documented, genuinely tested resilience — failover and black-building test reports that include what went wrong. UPS chemistry and battery room fire strategy, cooling type, indemnity period realism and customer concentration follow.
| Rating factor | Why it moves the premium | Mitigation |
|---|---|---|
| Tested vs asserted redundancy | Underwriters have seen enough N+1 design documents to discount them. Test evidence is what carries weight. | Run real load transfer and black-building tests; supply the reports including the failures found. |
| UPS chemistry and battery room design | Lithium-ion thermal runaway is the dominant severity driver and behaves unlike the fires halls were designed for. | Dedicated fire-rated compartment, chemistry-appropriate suppression, off-gas detection, explosion venting. |
| Suppression type and legacy systems | Gaseous systems alone are viewed less favourably for lithium; high-pressure inert discharge can itself damage HDDs. | Review the fire strategy against current chemistry; consider water mist for battery areas. |
| Cooling architecture | Liquid cooling puts water inside the rack; retrofits into air-designed halls are scrutinised hardest. | Leak detection and isolation at rack and manifold level, with documented design review. |
| Business interruption indemnity period | An unrealistic period is both an uninsured exposure and a signal that the risk has not been thought through. | Model from current supplier lead times annually — see our BI guide. |
| Sums insured adequacy | M&E replacement costs have risen sharply and lead times distort valuations; average applies here as anywhere. | Revalue plant regularly — see our underinsurance guide. |
| Customer concentration | A single dominant tenant converts a diversified revenue stream into a single point of commercial failure. | Disclose the concentration; consider contingent BI and trade credit alongside. |
| Maintenance regime and records | Machinery breakdown claims turn on whether a failure was sudden and unforeseen or a maintenance omission. | PPM on schedule with rectification evidence retained and available at claim. |
| Site and natural perils | Flood zone, water table, proximity to industrial hazard and grid resilience all feed directly into rating. | Flood and site hazard assessment, with mitigation documented where exposure exists. |
| OT security posture | BMS and DCIM compromise produces physical loss, and the property/cyber boundary is contested. | Segment OT from IT, control vendor remote access, and align the two policies deliberately. |
| Regulatory readiness | Preparation for the incoming NIS regime produces exactly the documentation underwriters want. | Use the compliance work as the underwriting submission — it is the same evidence. |
| Claims history | A prior outage, fire or water incident is heavily weighted and narrows an already small panel. | Evidence root cause analysis and corrective action — see our claims history guide. |
Data centre programmes are bespoke rather than packaged, so meaningful premium ranges are harder to give than in other classes and depend heavily on total insured values, indemnity period and limits purchased. As a broad indication for 2026: a small enterprise facility or edge estate may sit within a wider commercial programme at modest incremental cost, while a single well-engineered colocation facility typically runs into the tens of thousands annually across property, machinery breakdown, business interruption and liability. Hyperscale, AI-density and multi-site portfolio programmes are placed as structured schedules, frequently with layered limits and a Lloyd's component. What moves the number most is not size but the quality of the engineering evidence — a facility that can demonstrate tested resilience is priced on a different basis from one that asserts it.
What are a data centre operator's core legal duties?
Beyond ordinary occupier and employer duties, four regimes matter: the NIS Regulations as expanded by the Cyber Security and Resilience Bill, with registration, security measures and incident reporting to Ofcom; UK GDPR and the Data (Use and Access) Act 2025 where personal data is processed or hosted; fire safety under the Regulatory Reform (Fire Safety) Order 2005; and pressure systems and electrical safety for the plant itself.
Cyber security and resilience
The Cyber Security and Resilience (Network and Information Systems) Bill, introduced in November 2025 and expected to receive Royal Assent later in 2026, amends the NIS Regulations 2018 to bring data centre services into scope as an essential service within a new data infrastructure subsector, with Ofcom as regulator. In-scope operators will need to notify the regulator, satisfy structured information requirements, implement appropriate technical and organisational measures, and report significant incidents. Commencement follows secondary legislation and DSIT Codes of Practice, with a lead-in period.
Data protection
Where you process personal data — including as a processor for customers — UK GDPR obligations apply, including security of processing and personal data breach notification to the Information Commissioner's Office within 72 hours where the threshold is met. The Data (Use and Access) Act 2025 reached full implementation in February 2026 and affects the treatment of UK sovereign data.
Fire safety
The responsible person must carry out and maintain a fire risk assessment covering the whole premises, and maintain the fire safety measures. In a data centre this must specifically address battery chemistry, suppression system design and the interaction between suppression and the equipment being protected.
Plant, pressure and electrical safety
Chilled water systems and pressure vessels fall under the Pressure Systems Safety Regulations 2000 with a written scheme of examination; electrical installations engage the Electricity at Work Regulations 1989; and generators, fuel storage and standby plant carry their own maintenance and environmental duties. Our engineering inspection insurance guide covers the statutory inspection framework in detail.
Environmental and energy
Climate Change Agreement targets, energy efficiency reporting and, from 2026, building lifecycle carbon disclosure apply to the sector, alongside standard duties on fuel storage, refrigerant handling under F-Gas rules, and water discharge where evaporative cooling is used.
Real data centre claims and what they cost
The three composites below reflect the pattern we see in this class: one lithium UPS fire where the physical damage was trivial next to the downtime, one liquid cooling leak in a retrofitted hall, and one OT compromise that produced a physical loss and a coverage dispute. These are illustrative composites written for this guide — not real clients and not actual claims — and the figures are indicative examples rather than settlements we have handled.
Case study 1 — Lithium UPS thermal runaway in a retrofitted battery room
The situation: A colocation operator replaced end-of-life VRLA strings with lithium-ion modules across two UPS rooms, gaining footprint and service life. The rooms retained their original inert gas suppression and the fire strategy was not revisited. Eighteen months later a cell entered thermal runaway overnight. The gas discharged, failed to arrest the reaction, and propagation continued through the module. The room was lost, one UPS system was destroyed, and the affected hall ran on the redundant path for six weeks before load could be safely rebalanced.
Claim type: Property damage, machinery breakdown and business interruption.
Settlement: Battery room reinstatement and UPS replacement £780,000, with a nineteen-week lead time on the replacement modules. Smoke and particulate remediation across adjacent plant areas £145,000. Business interruption, including capacity that could not be sold during the constrained period, £610,000. Total £1,535,000. The insurer initially reserved its position on the chemistry change, which had not been notified; cover was ultimately confirmed after the operator evidenced that the retrofit had been carried out to manufacturer specification, but the reservation added eleven weeks to settlement.
Lessons: The retrofit was a material change of risk and should have been notified when it happened. Separately, the fire strategy was never reassessed for the new chemistry — gaseous suppression cannot arrest a reaction that supplies its own oxygen and heat. Off-gas detection would have provided the only meaningful intervention window and was absent.
Renewal impact: Premium increased approximately 75%, with conditions requiring off-gas detection, a revised fire strategy for all battery areas, and notification of any further chemistry change.
Case study 2 — Coolant leak over a GPU cluster
The situation: An operator retrofitted direct-to-chip liquid cooling into two halls to accommodate customer GPU deployments. A manifold connection installed during the retrofit developed a slow leak above a populated rack. Leak detection had been specified at CDU level but not at rack level, and the escape was found on a routine walk-round several hours after it began. Coolant had entered three racks of GPU compute belonging to a single customer.
Claim type: Property damage, third-party equipment damage and business interruption.
Settlement: The operator's own infrastructure damage was modest at £41,000. The customer's GPU hardware loss was assessed at £2.1m. The colocation agreement placed the risk of physical loss of customer equipment on the customer, who claimed under their own policy — but their insurer pursued a subrogated recovery against the operator alleging negligent installation and inadequate leak detection. Settled at £680,000 contribution plus £94,000 defence costs, against a business interruption element of £180,000. Total exposure £995,000.
Lessons: The contractual allocation of risk worked exactly as intended and did not prevent a subrogated recovery based on negligence. Rack-level leak detection was the control that would have limited the loss, and its absence was the central allegation. Note also that the escape of water sub-limit in the property section had never been reviewed after the retrofit and would have been inadequate had the operator's own equipment been more affected.
Renewal impact: Premium increased approximately 55%, with a condition requiring rack-level leak detection and automatic isolation across all liquid-cooled halls.
Case study 3 — BMS compromise, physical outcome, contested coverage
The situation: An attacker obtained access to a building management system through a chiller vendor's remote support connection, which used shared credentials and was not segmented from the wider estate. Cooling setpoints were altered and alarms suppressed. Temperatures rose in one hall until equipment began thermal shutdown. The intrusion was identified within four hours and cooling restored, but the hall was down for eleven hours in total and some customer hardware sustained heat damage.
Claim type: Business interruption and property damage from a cyber cause — presented to both the property and cyber insurers.
Settlement: Business interruption £320,000. Own equipment damage £86,000. Customer claims and goodwill credits £240,000, of which the SLA credit element of £155,000 was a contractual cost falling outside both policies. The property insurer pointed to its cyber exclusion; the cyber insurer's wording was focused on data and network interruption rather than physical damage to plant. Resolved after four months with a negotiated split between the two insurers, settling £406,000 of a £646,000 total exposure and leaving the credits and part of the equipment loss uninsured.
Lessons: This is the classic property/cyber gap and it is entirely foreseeable at placement. The two wordings should be read side by side, with the property policy's cyber exclusion and the cyber policy's physical damage scope reconciled deliberately rather than left to be argued after an event. Separately, vendor remote access with shared credentials into an unsegmented OT network was the root cause and is a control failure, not bad luck.
Renewal impact: Both policies restructured under one broker with the boundary explicitly aligned; combined cost rose approximately 30% while removing the disputed gap.
How should you manage a data centre insurance claim?
Run four workstreams in parallel from the first hour: technical restoration, customer communication, regulatory notification and the insurance claim. The two things that most affect the outcome are preserving the failed plant and its telemetry before anything is replaced, and getting a forensic engineer appointed early — because whether a failure was sudden and unforeseen or a maintenance omission is what decides the machinery breakdown claim.
- Make the facility safe and stabilise the load. Life safety first, then protect what is still running. Resist the instinct to restore the failed path immediately if doing so risks the redundant one — a second failure during recovery is a materially worse outcome than a longer single-path period.
- Notify your broker within the hour, not at the end of the day. Data centre claims move fast and the adjuster and forensic engineer appointments made in the first day shape everything afterwards. Your broker should be arranging those while you are still restoring service.
- Preserve the failed plant and all telemetry. Do not scrap damaged UPS modules, switchgear or cooling components, and export BMS, DCIM and alarm logs immediately before retention windows roll. This evidence determines the cause finding, and the cause finding determines whether the policy responds.
- Appoint the forensic engineer jointly where possible. A cause investigation commissioned only by the insurer puts you on the back foot. Agreeing the expert and the scope early tends to produce a faster and less contested outcome.
- Handle regulatory notification separately and deliberately. Where the incident is reportable — to Ofcom under the incoming regime, or to the ICO where personal data is affected — those timescales run independently of the insurance claim, and the accounts given must be consistent with each other.
- Communicate with customers on a controlled basis. Statements made during an outage are read later by claimants and insurers. Confirm facts, avoid speculating on cause, and route anything touching liability through your broker.
- Separate the SLA credits from the insurance claim from the outset. Credits are a contractual cost and are generally not recoverable. Quantify them separately so they are neither double-counted in the BI claim nor overlooked in the financial impact assessment.
- Order long-lead replacement plant immediately. Do not wait for coverage confirmation on items with multi-month lead times — discuss an early payment on account with the adjuster. Every week of delay in ordering extends the business interruption directly, and the indemnity period is finite.
Glossary of data centre insurance terms
- Black-building test
- A full simulated loss of utility power, testing whether generators, UPS and cooling carry the load as designed. The most persuasive single piece of resilience evidence at underwriting.
- CDU (coolant distribution unit)
- The plant managing coolant supply to liquid-cooled racks. A single point of failure worth examining, and a machinery breakdown exposure in its own right.
- Concurrent maintainability
- The ability to take any capacity component out of service for maintenance without affecting IT load. Broadly the Tier III standard and a key underwriting distinction.
- Contingent business interruption
- Cover responding where the loss occurs at a supplier's or customer's premises rather than your own. Relevant both to your grid and fibre dependencies and to your customers' dependency on you.
- Critical National Infrastructure (CNI)
- The designation applied to UK data centres in September 2024, treating disruption of a major facility as a national security matter.
- DCIM
- Data Centre Infrastructure Management — the software layer monitoring power, cooling and capacity. Part of the OT estate and therefore part of the cyber-physical exposure.
- Delay in start-up (DSU)
- Cover for lost revenue where a construction or expansion project is delayed by insured damage. Usually the largest exposure on a data centre build.
- Denial of access
- Business interruption extension responding where the facility is undamaged but unusable or unreachable — a cordon, a fibre cut or a grid failure outside the boundary.
- Escape of water sub-limit
- The cap applied to water damage claims. Frequently set before liquid cooling existed and inadequate once a pressurised circuit runs inside the rack.
- Indemnity period
- The maximum period over which business interruption is payable. In this class it must be modelled from plant lead times, not building reinstatement.
- N+1 redundancy
- One more component than required to carry the load. Widely asserted; underwriters increasingly want evidence it has been tested under real transfer conditions.
- NIS Regulations
- The Network and Information Systems Regulations 2018, being amended by the Cyber Security and Resilience Bill to bring in-scope data centres under Ofcom regulation.
- Off-gassing
- Release of flammable and toxic electrolyte vapour from a lithium cell before thermal runaway establishes. Detecting it provides the only meaningful intervention window.
- Operational technology (OT)
- The control systems running the physical building — BMS, chiller controllers, UPS management, access control. Where a cyber compromise produces a physical loss.
- SLA credit
- Contractual service credit payable to a customer when an uptime commitment is missed. A contractual cost rather than an insured loss, and frequently mistaken for one.
- Thermal runaway
- A self-accelerating reaction in a lithium cell producing its own heat and oxygen, propagating between cells and largely unresponsive to oxygen-displacing suppression.
- VRLA
- Valve-regulated lead acid — the traditional UPS battery chemistry now widely displaced by lithium-ion, with a materially different fire profile.
Frequently asked questions about data centre insurance
Rarely. Three things typically fail. The business interruption is scoped as a secondary consideration when it is usually the dominant loss. The indemnity period is set at a default twelve months against electrical plant lead times that can exceed two years. And customers' equipment sits in a care, custody and control exclusion that nobody has addressed. Beyond that, the severity in this class is driven by plant failure and lithium-ion fire rather than conventional perils, which requires an underwriter who prices the engineering rather than the square footage.
In September 2024 UK data centres were designated CNI — the first such designation in nearly a decade — which signalled that disruption of a major facility is treated as a national security event and gave the sector prioritised engagement with government and the NCSC. The Cyber Security and Resilience Bill, introduced in November 2025 with Royal Assent expected later in 2026, converts that into statute by amending the NIS Regulations 2018 to make data centre services an essential service, with Ofcom as regulator and duties covering security measures, incident reporting and regulator engagement.
The indicative thresholds are 1MW of capacity at a single site, or 10MW across multiple locations, which brings most colocation, hyperscale and large enterprise facilities into scope while leaving smaller server rooms and individual edge nodes outside. Detail follows secondary legislation and DSIT Codes of Practice, with a lead-in period before commencement. If your facility sits near the threshold, or you operate several sites that aggregate above it, the position is worth reviewing properly rather than assuming.
Because thermal runaway does not behave like the fires most halls were designed for. A lithium cell in runaway generates its own heat and oxygen and propagates to adjacent cells, so the gaseous and clean-agent suppression common in data centres — which works by displacing oxygen or absorbing heat — cannot reliably arrest it. It also produces flammable off-gas that creates an explosion risk in an enclosed room. Underwriters look hardest at compartmentation, suppression appropriate to the chemistry, off-gas detection and cell provenance.
Yes, and you should notify now rather than at renewal. Replacing VRLA with lithium-ion is a material change of risk under the Insurance Act 2015 because it alters the fire profile of the building substantially. Since the Act the remedy is proportionate rather than automatic avoidance, but a fire in an undisclosed lithium installation is still a serious coverage problem and will at minimum cause a reservation of rights and significant delay. Notifying voluntarily is straightforward; being found out during a claim investigation is not.
Normally the customer, and most colocation agreements say so explicitly — placing the risk of physical loss of customer equipment on the customer and requiring them to insure it. That is the correct allocation. What operators need to check is that their own policy matches what the contract says, because customers' equipment is third-party property in your care, custody and control and is excluded by standard public liability. Note also that a contractual allocation does not prevent a subrogated recovery against you if negligence is alleged.
Generally not. Service credits are a contractual price adjustment you owe your customer, not a loss of your own revenue, and standard business interruption responds to the latter. On a large facility with a multi-hour outage the aggregate credit exposure across the customer base can be substantial, so it should be modelled as a financial exposure in its own right rather than assumed to fall within the BI claim. Quantify it separately at claim stage too, so it is neither double-counted nor overlooked.
Model it from plant lead times, not building reinstatement. Twenty-four months should be the starting assumption for any facility whose critical path runs through bespoke electrical plant, and thirty-six is defensible where equipment is long-lead or non-standard. Two things push it further: transformers and switchgear now carry lead times measured in many months and sometimes years, and colocation customers who migrate during a long outage may not return, so revenue recovery extends past physical reinstatement.
Significantly. Direct-to-chip liquid cooling puts a pressurised water circuit inside the rack, directly above extremely high-value compute — a different exposure from a CRAC unit in an aisle. Check whether the escape of water sub-limit in your property section contemplates water in the rack, because many wordings were drafted when the only water in a hall was overhead. Underwriters will want rack and manifold level leak detection with automatic isolation, CDU redundancy, and evidence that any retrofit was properly designed rather than added on.
This is the most contested gap in the class. Property policies contain cyber exclusions; cyber policies are often built around data and network interruption rather than physical damage to plant. An attacker who manipulates a chiller controller causes a physical loss from a cyber cause, and both insurers can plausibly disclaim. The fix is to read the two wordings side by side at placement — establishing whether the property cyber exclusion carves back resulting physical damage, and whether the cyber policy extends to OT systems and physical loss — rather than discovering the boundary during a claim.
These are bespoke programmes rather than packaged products, so ranges depend heavily on total insured values, indemnity period and limits. Broadly, a small enterprise facility or edge estate may sit within a wider commercial programme at modest incremental cost, while a single well-engineered colocation facility typically runs into the tens of thousands annually across property, machinery breakdown, business interruption and liability. Hyperscale and multi-site portfolios are placed as structured schedules, often layered and with a Lloyd's component. Quality of engineering evidence moves the number more than size does.
Look for a broker who asks about UPS chemistry, suppression design and your failover test history before discussing premium, who understands that the indemnity period is a plant lead time question, and who has direct access to Lloyd's syndicates and specialist MGAs rather than a general property panel. Miller & Partner approaches every placement through the Insurability Framework — underwriter intelligence, difficult-risk expertise, risk assessment and claims advocacy — which is what allows lithium retrofits, liquid-cooled AI halls and operators recovering from an outage to be placed on their engineering rather than their trade code. Start with our AI and tech hub or our cyber insurance page.
Important information
Regulatory status. Miller & Partner Limited is an Appointed Representative of Gauntlet Risk Management Ltd, which is authorised and regulated by the Financial Conduct Authority. Miller & Partner Limited is registered in England and Wales and trades from Vivian House, Roman Bridge Close, Mumbles, Swansea SA3 5BG.
This is general information, not advice. This guide is published for general information about how insurance for data centres works. It is not personalised advice, a personal recommendation, or an offer of cover, and it does not take account of the circumstances of any particular business. No liability is accepted for any action taken, or not taken, on the basis of this guide alone.
Cover, wordings and figures. Cover is subject to insurer acceptance, underwriting criteria, and the full terms, conditions, limits and exclusions of the policy issued. Wordings differ substantially between insurers, so always read your own policy documents. Any premium ranges, limits or claim figures shown are illustrative examples only and are not quotations or indications of price.
Case studies. The case studies in this guide are illustrative composites written for the purpose of explanation. They do not describe real clients or actual claims.
Regulation and legislation. References to legislation, regulators and official guidance reflect our understanding as at the date of publication and may change. Always work from the current published source rather than this guide.
Miller & Partner Limited, Vivian House, Roman Bridge Close, Mumbles, Swansea SA3 5BG. Telephone 01792 001350. FS Register FRN 1029698 — you can check this on the Financial Services Register at register.fca.org.uk.
Instagram
LinkedIn