← All incidents
Incident dossier · Rank #19
NorthC Almere Data Center Fire: A Compartment Blaze That Took Out the Facility Power Plant and Rippled Across Dutch Public Services
NorthC Datacenters
2026-05-07
144h 0m core impact
FirePower
On 7 May 2026 (around 08:45 local time) a fire broke out in a rear technical compartment of NorthC's Almere data center (Rondebeltweg) that houses power infrastructure. No servers or data carriers caught fire, but the blaze burned part of the facility's power supply and the site lost power entirely — the compute survived, yet everything went dark, and NorthC ultimately had to rebuild the power plant with new transformers and UPS. (Published reporting says only that 'part of the power supply burnt down'; that BOTH the normal and standby feeds were lost is an engineering inference from the total, multi-day power loss and the full rebuild, not an official finding.) Because multiple Dutch organizations host critical systems there, the outage cascaded outward: Utrecht University, Statistics Netherlands (CBS), public-transport operator Transdev, a regional water board, research-ICT provider Surf, IBM Cloud's Amsterdam 03, and numerous GP practices all lost service. The fire was scaled down to GRIP 1 at 20:50 that evening and to GRIP 0 with the building partially released to NorthC teams at 00:35 on 8 May, kicking off a phased restoration estimated at up to 72 hours. Temporary redundant power and cooling were activated on 13 May, roughly a week after the fire. Grid power came back building-by-building over the following weeks, with Building 1's full utility-plus-standby topology targeted for 17 July 2026. The root cause remains officially undetermined; a draft cause report was only expected at end-September 2026 because the fire-damaged structure had to be secured first.
Failure cascade
Trigger → primary fault → downstream blast radius, derived from the sourced root cause and affected-services record.
Facility & location
- Operator
- NorthC Datacenters
- Data center
- NorthC Almere (Rondebeltweg)
- Location
- Almere, Netherlands
- Date
- 2026-05-07
Impact & scale
- Users affected
- Undisclosed headcount; multi-sector impact documented by named victims (Utrecht University, Statistics Netherlands/CBS, Transdev bus & tram, water board De Stichtse Rijnlanden, Surf, IBM Cloud AMS03, and GP/healthcare practices)
- Financial
- Not published; facility owner Keppel DC REIT stated the incident was not expected to have a material impact on FY2026 distribution per unit; no damages, insurance, or customer-loss figure disclosed
- Scope
- Sev-1 facility-level power loss (multi-day)
Services / systems down
- Utrecht University network/applications/websites and access passes
- Statistics Netherlands (CBS) platforms
- Transdev bus and tram services
- Water board Hoogheemraadschap De Stichtse Rijnlanden network
- Surf member research/education infrastructure
- IBM Cloud Amsterdam 03
- GP practices' patient-record, scheduling and contact systems
Impact data & metrics
| Affected system class | Power infrastructure only — transformers + UPS destroyed, 0 IT devices fire-damaged |
| Facility electrical capacity | 11 MW |
| Facility footprint | ~26,000 sqm across two buildings (Almere 1 + Almere 2) |
| Fire outbreak time | ~08:45 (NorthC) / ~08:30 (Omroep Flevoland, TNW), Thu 7 May 2026 |
| Time to 'brand meester' (under control) | ~20:30 — approximately 11.5–12 hours of active firefighting |
| Casualties / injuries | 0 — all occupants evacuated in time; no dangerous-substance concentrations detected |
| Temporary power train | Activated 13 May 2026 (~6 days post-fire): generators + UPS + distribution units + >1 km of cable |
| Almere 2 grid reconnection | 29 May 2026 |
| Full site restoration target | Almere 1 phased — initial connection end June, full grid operation end July 2026 |
| Cause report timeline | DRAFT only by end September 2026 (~4.5 months after the fire) |
| Detection / suppression forensics | NOT DISCLOSED — no source reveals detection type/time or whether any active suppression discharged |
Magnitude profile
Magnitude 7.8 = blast 9×0.35 + users 8×0.25 + financial 5×0.20 + duration 8×0.20 (sub-scores 0–10 · weighted composite)
Blast radius is high because the failure reached across sectors of Dutch public life (higher education, national statistics, public transport, water management, healthcare, cloud) from a single site. Users and duration are high (multi-day facility power loss, weeks to full grid). Financial score is a mid-band estimate held down by the absence of any published damages figure and the owner's 'no material FY2026 DPU impact' statement; the true operational and tenant cost is undisclosed.
Sequence of events (SOE)
- TRIGGER Fire breaks out at the rear technical room of the 'Almere 1' compartment on Rondebeltweg, Almere, where the technical facilities (emergency-power / technische ruimte) are located. NorthC dates it to 'around 08:45'; Omroep Flevoland and TNW report ~08:30.
- DETECTION Occupants alerted and the building evacuated in time — 'All people present were evacuated from the building in time' (CEO Alexandra Schless). NO source discloses the detection technology, zone or exact first-alarm time; only the human outcome (safe evacuation, no injuries) is documented.
- MITIGATION Fire brigade responds and orders a COMPLETE datacenter shutdown; the fire is confirmed in the technical room of 'Almere 1'.
- CASCADE In the undamaged half, backup generators start automatically — 'De noodstroomgeneratoren in het andere deel (Almere 2) sloegen automatisch aan.'
- MITIGATION Passive fire compartmentation confines the blaze to the rear compartment: 'The fire compartmentation prevented the smoke from spreading further' and 'smoke in the halls remained limited.' No active-suppression discharge is disclosed in any source.
- IMPACT Almere 1's power plant is lost: transformers and UPS destroyed; the emergency electrical installation is 'classified as lost.' No IT gear fire-damaged — damage concentrated around the site's power infrastructure.
- MITIGATION Veiligheidsregio Flevoland (at GRIP 2) issues an NL-Alert: avoid smoke, close windows/doors, switch off mechanical ventilation. Smoke drifts south toward 't Gooi; emergency measurements detect no dangerous-substance concentrations.
- MITIGATION Firefighting hampered — the compartment is too unstable and too damaged to send crews inside. Facade sections are demolished with cranes for access; a drone from Brabant provides interior imaging; a fire truck from Lelystad Airport assists.
- CASCADE Transdev loses contact with its regional control centre hosted at Almere; its servers 'had not been migrated to a backup location,' so bus-driver comms fail and the in-vehicle emergency button stops working.
- CASCADE Further downstream impact: Utrecht University research/education disrupted, Infomedics affected, local water board affected; SURF warns member institutions.
- MITIGATION Mayor Hein van der Loo characterises the scene as a 'vreselijke brand' / 'vuurzee' and warns the fire is not yet under control and may take a while.
- MITIGATION 'Brand meester' (fire under control) reported — roughly 11–12 hours after outbreak.
- RECOVERY Incident downscaled from GRIP 2 to GRIP 1; NL-Alert withdrawn. (Precise clock time not independently re-verified this pass.)
- RECOVERY Downscaled to GRIP 0; the building is partially released back to NorthC. (Precise clock time not independently re-verified this pass.)
- RECOVERY Temporary redundant power train activated for Almere 1 — generators, UPS, distribution units and more than 1 km of cable (~6 days after the fire; Computable reports 09:30). NorthC confirms 'temporary power May 13.'
- RESTORED Almere 2 reconnected to the electricity grid.
- RESTORED Almere 1 restored to grid in phases as transformers and UPS are reinstalled — initial grid connection by end June, full grid operation targeted end July 2026.
- DETECTION Independent expert cause investigation proceeds only after the unstable structure is secured (demolition/investigation window 27 Jul–7 Aug); NorthC expects a DRAFT cause report by end September 2026 — ignition source and chemistry remain publicly unknown.
Root cause
SPECIFIC IGNITION SOURCE — NOT PUBLICLY DISCLOSED (investigation open). No primary or press source names the equipment that ignited: no make, model, age, or chemistry, and no failure mechanism. On day one NorthC stated plainly that nothing was yet known about the cause. NorthC commissioned an independent expert investigation, but forensic work could only begin after the severely damaged, unstable structure was secured (a demolition/investigation window of 27 July–7 August 2026), and NorthC expects only a DRAFT cause report by end of September 2026. As of the reporting window the ignition source and chemistry are genuinely unknown; any specific attribution would be fabrication.
WHAT IS FIRMLY ESTABLISHED (confirmed on the official NorthC status page) is the location and the affected system class. The fire started at the rear of the building, where the technical facilities are located, inside the emergency-power / technical room of the "Almere 1" compartment, and centred on the site's power infrastructure. The consequence set narrows the field: the transformers and UPS systems in that compartment were destroyed and the emergency electrical installation was "classified as lost" (NorthC), while no IT gear was fire-damaged (DataCenterDynamics, crawl-cited). That the destroyed plant was UPS + transformers makes two failure families physically plausible — (a) a UPS energy-storage thermal event (a VRLA or Li-ion battery string entering thermal runaway) or (b) an electrical fault in a transformer or switchgear/distribution assembly (arc flash, insulation breakdown, connection overheating). BOTH are UNCONFIRMED; no source attributes ignition to either, and this dossier asserts neither.
LATENT / SYSTEMIC ROOT (what the record supports). At the operator level the design worked as intended where it mattered for life-safety and containment: the two-compartment topology and PASSIVE fire compartmentation confined the fire to the rear power room and kept the data halls and their IT largely intact — "The fire compartmentation prevented the smoke from spreading further" and "smoke in the halls remained limited" (NorthC, verified). Where the design failed was in continuity of the affected compartment's own power train: the fire took out transformers and UPS together, i.e. primary and backup power for Almere 1 shared enough physical proximity in one technical room that a single fire destroyed the entire emergency electrical installation of that half of the site. That is a redundancy / physical-separation latent root inside Almere 1's power topology, even though inter-compartment separation (Almere 1 vs Almere 2) held.
Second — the more consequential systemic root for the outside world — the severe downstream impact was NOT caused by any NorthC fire-safety failing but by SINGLE-SITE customer architectures with no disaster-recovery failover. Transdev's regional control-centre servers had not been migrated to a backup location, so bus drivers lost communications and the in-vehicle emergency button stopped working; Utrecht University, Infomedics and the local water board carried single-site dependencies on this one facility (TNW / Computable, crawl-cited).
MAINTENANCE / INSPECTION / TESTING LATENT ROOT — NONE EVIDENCED, NOT EXONERATED. No source implicates any hot-work permit, deferred maintenance, failed test, or procedural breach at NorthC — consistent with the cause investigation still being open. Any maintenance-cause theory must be treated as unproven pending the expert report (draft due end September 2026).
Contributing factors
- Co-location of primary and backup power in one room (Almere 1): a single fire in the rear technical room destroyed transformers AND UPS together, so the entire emergency electrical installation of that compartment was 'classified as lost' (NorthC, verified) — indicating insufficient physical separation between the power train's redundant elements within Almere 1.
- Absence (or non-disclosure) of effective automatic suppression in the affected power room: no source evidences that any gaseous / water-mist / sprinkler system discharged or held the fire, which burned ~11–12 hours and required manual firefighting until 'brand meester' at ~20:30 (Omroep Flevoland, crawl-cited). Whether a suppression system existed and failed, or was never installed, is an open gap.
- Structural fragility of the burning compartment slowed both firefighting and forensics: the room was too unstable and too damaged to send crews inside, forcing crane demolition of the facade and drone imaging (Omroep Flevoland, crawl-cited) — extending outage duration and pushing the cause investigation to a 27 July–7 August 2026 demolition/investigation window (NorthC, verified).
- Customer-side single-site architecture with no DR failover (a downstream contributor to IMPACT, not to ignition): Transdev's regional control-centre servers 'had not been migrated to a backup location' (TNW, crawl-cited), collapsing bus-driver communications and the in-vehicle emergency button; Utrecht University, Infomedics and the local water board carried single-facility dependencies.
- Maintenance / inspection / testing status is UNEVIDENCED, not exonerated: with the cause investigation open and only a draft report due end September 2026 (NorthC, verified), no maintenance, hot-work, or testing lapse has been disclosed or ruled out — the factor cannot be assessed from the public record.
Correction of errors (COE)
- Independent expert cause investigation (ignition source, chemistry, mechanism)
- Activate temporary redundant power train for Almere 1 (generators, UPS, distribution, >1 km cable)
- Reconnect Almere 2 to the electricity grid
- Rebuild Almere 1 emergency electrical installation (reinstall transformers + UPS) and restore to grid
- Publish detection-to-alarm timeline and automatic-suppression status for the affected power room
- Review and implement DR failover for single-site critical dependencies
Lessons learnt
- Passive compartmentation is the hero of this incident — it confined an intense power-room fire and kept the data halls and IT largely intact with no injuries; sustained investment in passive fire separation pays off exactly when active systems are absent or unknown.
- Redundancy that shares a room is not redundancy — co-locating transformers and UPS in one technical space meant a single fire took out an entire compartment's power train; physical separation, not just N+1 counts, defines real resilience.
- Facility resilience does not equal customer resilience — the worst real-world harm (bus comms and the in-vehicle emergency button failing) fell on a tenant that had not replicated to a backup site; operators should actively push single-site critical customers toward DR.
- Cause timelines stretch to months when the structure is unsafe — forensics were gated by a demolition window, so a draft cause report only ~4.5 months out is a realistic expectation, not negligence; set stakeholder expectations accordingly.
- Silence on detection and suppression erodes trust — the absence of any public detail on fire detection or automatic suppression is itself a finding; transparent post-incident disclosure (once safe) is part of operational credibility.
Improvements & remediation
- Safety (disclosure): once the expert report permits, publish the detection-to-alarm timeline and whether any automatic suppression existed in the power room, discharged, and held — the ~11–12 h manual firefight and total silence on active suppression are the single largest unexplained safety gaps in the public record.
- Design (separation): physically separate the redundant elements of each compartment's power train (transformers, UPS/battery, switchgear) into distinct fire compartments so a single fire cannot destroy both primary and backup power for one hall, as happened in Almere 1.
- Maintenance (assurance): once the cause report lands, publish the technical room's maintenance, inspection, hot-work and testing history; institute an independent thermographic/IR inspection cadence on transformers, switchgear and UPS, plus continuous battery-string health/temperature monitoring for early thermal-runaway detection.
- Customer resilience: advise/require single-site critical tenants (mobility, health, utilities such as Transdev, Utrecht University, Infomedics, the water board) to implement DR failover to a second site, since facility resilience did not prevent tenant outages.
- Emergency response: pre-plan structural-collapse firefighting access for power rooms (external attack points, pre-engineered crane/demolition access) so a compartment that becomes too unstable to enter does not extend the outage by ~11–12 hours.
- Battery chemistry: confirm and disclose the UPS battery chemistry (VRLA vs Li-ion) and, where Li-ion, deploy off-gassing/aspirating detection and battery-room-rated suppression sized for thermal runaway.
Comprehensive analysis
What happened
At ~08:30–08:45 on 7 May 2026 a fire broke out in the rear technical room of the 'Almere 1' compartment of NorthC's Almere data center, centred on the site's power infrastructure. Everyone was evacuated in time with no injuries. Transformers and UPS in that room were destroyed and the emergency electrical installation was 'classified as lost' (NorthC, verified). The fire brigade ordered a full shutdown and fought the blaze into the evening, with 'brand meester' reported around 20:30. The ignition source and chemistry remain publicly unknown; NorthC expects only a draft cause report by end September 2026.
Why the data halls survived: passive containment
The load-bearing safety story is passive fire compartmentation, not active suppression. Per NorthC, 'The fire compartmentation prevented the smoke from spreading further' and 'smoke in the halls remained limited.' The two-compartment topology held between Almere 1 and Almere 2, so no IT gear was fire-damaged even as the Almere 1 power room was destroyed. No source discloses whether any automatic suppression existed in that room or discharged — a genuine gap given an ~11–12 hour manual firefight.
The real single point of failure: a co-located power train
Inter-compartment separation worked, but within Almere 1 the primary and backup power shared one technical room closely enough that a single fire destroyed transformers and UPS together — the entire emergency electrical installation of that half of the site. This is the central latent root: redundancy defined by device counts (N+1) rather than physical fire separation collapses to a single point of failure when both elements sit in one space. Rebuild and future design should place redundant power-train elements in distinct fire compartments.
Downstream impact was a customer-architecture failure, not a NorthC fire-safety failure
The most visible real-world harm came from single-site tenant architectures. Transdev's regional control-centre servers 'had not been migrated to a backup location' (TNW, crawl-cited), so bus-driver communications and the in-vehicle emergency button failed. Utrecht University, Infomedics and the local water board carried single-facility dependencies. Utrecht University's reported damages claim targets its telecom provider KPN, not a NorthC maintenance breach — underscoring that DR failover is a tenant responsibility the operator should actively encourage.
What remains unknown — and why that is defensible
Three things are genuinely undisclosed: (1) the ignition source/equipment/chemistry, (2) the detection technology and first-alarm time, and (3) whether any automatic suppression existed or discharged. The cause gap is defensible: forensics could only start after an unstable structure was secured (27 Jul–7 Aug window), so a draft report ~4.5 months out is realistic. The detection/suppression gaps are less defensible as disclosure and should be closed once safe. No maintenance, hot-work, or testing lapse is evidenced — but with the investigation open, none is ruled out either.
Technical deep-dive
The NorthC Almere site is an 11 MW facility of roughly 26,000 sqm across two buildings, split into two functionally separate power/fire compartments referred to as "Almere 1" and "Almere 2" (11 MW / 26,000 sqm figures from DataCenterDynamics, crawl-cited and not independently re-verified this pass). The fire, at approximately 08:30–08:45 on Thursday 7 May 2026, originated in the rear technical room housing Almere 1's emergency-power plant (NorthC dates the outbreak to "around 08:45"; Omroep Flevoland/TNW report ~08:30).
Two forensic facts dominate the physics. FIRST, this was an ELECTRICAL / POWER-PLANT fire, not an IT-white-space fire: damage was concentrated around the site's power infrastructure, the transformers and UPS in the Almere 1 room were destroyed, and "the emergency electrical installation was classified as lost" (NorthC, verified), while no IT gear was fire-damaged (DCD, crawl-cited). SECOND, PASSIVE containment — not active suppression — is what saved the halls: "The fire compartmentation prevented the smoke from spreading further" and "smoke in the halls remained limited" (NorthC, verified).
Two disclosure gaps must be stated honestly. DETECTION: no source discloses the detection technology, zone, or first-alarm time. What is documented is the human outcome — the building was evacuated in time with no injuries; per CEO Alexandra Schless, "All people present were evacuated from the building in time" (NorthC, verified). ACTIVE SUPPRESSION: no source states whether any automatic suppression (gaseous/clean-agent, water-mist, sprinkler) was installed in the affected power room, whether it discharged, or why it failed to hold. This is a real gap: the fire burned into the evening and was fought manually, with "brand meester" (fire under control) reported only around 20:30 — roughly 11–12 hours after outbreak (Omroep Flevoland, crawl-cited). Firefighting was severely hampered because the compartment was too unstable and too damaged to send crews inside; facade sections were demolished with cranes for access, a drone from Brabant was flown for interior imaging, and a fire truck from Lelystad Airport assisted (Omroep Flevoland / Computable, crawl-cited).
The electrical response and recovery are the best-documented portion and are confirmed against the official NorthC status page. The fire department ordered a complete datacenter shutdown. In the undamaged half, the backup generators started automatically ("De noodstroomgeneratoren in het andere deel (Almere 2) sloegen automatisch aan," Computable, crawl-cited). Almere 1, having lost its transformers and UPS, required a full temporary power train — generators, UPS, distribution units and more than 1 km of cable — activated on 13 May (about six days post-fire; NorthC confirms "temporary power May 13"). Almere 2 was reconnected to the grid on 29 May; Almere 1 was reinstated in phases (transformers/UPS reinstalled) with initial grid connection by end June and full grid operation targeted end July 2026 (all confirmed on the NorthC page). Veiligheidsregio Flevoland ran the incident at GRIP 2 with an NL-Alert advising residents to avoid smoke, close windows/doors and disable mechanical ventilation as smoke drifted south toward 't Gooi; emergency measurements found no dangerous-substance concentrations. Reported GRIP de-escalation (GRIP 2 → GRIP 1 ~20:50 on 7 May, NL-Alert withdrawn; → GRIP 0 ~00:35 on 8 May) originates with Veiligheidsregio Flevoland / Omroep Flevoland and was not independently re-verified this pass.
References & provenance
- official-postmortem Fire in our data center in Almere (running update log)“On 17 July, the data center will be fully reconnected to the regular grid, with the emergency generators as back-up.”https://www.northcdatacenters.com/en/news/fire-in-our-data-center-in-almere/
- press DataCenterDynamics — NorthC datacenter catches fire in Almere, Netherlands“Damage was concentrated around the site's power infrastructure; no IT gear was damaged by fire (11 MW / ~26,000 sqm, two buildings).”https://www.datacenterdynamics.com/en/news/
- press The Next Web (TNW) — Fire at Dutch data center NorthC Almere“Transdev's servers had not been migrated to a backup location, so bus-driver communications and the in-vehicle emergency button stopped working.”https://thenextweb.com/
- press Computable — NorthC Almere datacenter fire and recovery“De noodstroomgeneratoren in het andere deel (Almere 2) sloegen automatisch aan.”https://www.computable.nl/
- press Omroep Flevoland — Grote brand datacenter Almere“'Brand meester' was reported around 20:30, roughly 11–12 hours after outbreak; the compartment was too unstable to enter, requiring crane demolition and drone imaging.”https://www.omroepflevoland.nl/
- press Major fire takes down NorthC data centre in the Netherlands for six days — Structure Research“NorthC Datacenters had a fire at its facility in Almere near Amsterdam on May 7 which took down both primary and backup power systems.”https://structureresearch.net/2026/05/20/major-fire-takes-down-northc-data-centre-in-the-netherlands-for-six-days/
- news NorthC data center outside Amsterdam suffers fire“Transdev said its bus and tram services were also impacted ... IBM Cloud has also reported issues at the data center hosting its Amsterdam 03 facility.”https://www.datacenterdynamics.com/en/news/northc-data-center-outside-amsterdam-suffers-fire/
- news Almere data center fire under control; caused outages throughout Netherlands“No servers or data carriers caught fire, but part of the power supply burnt down.”https://nltimes.nl/2026/05/08/almere-data-center-fire-control-caused-outages-throughout-netherlands
- news Major fire takes down NorthC data centre for six days“Major fire takes down NorthC data centre in the Netherlands for six days.”https://www.structureresearch.net/2026/05/20/major-fire-takes-down-northc-data-centre-in-the-netherlands-for-six-days/
- news Fire at NorthC data center: all personnel evacuated in time“At the time of writing, it was unclear how many companies experienced technical issues as a result of the fire.”https://www.techzine.eu/news/infrastructure/141131/fire-at-northc-data-center-all-personnel-evacuated-in-time/
Sourced from public post-incident reports. Quotes are short attributed excerpts for provenance only; the analysis above is original and substantially shorter than its sources. Last verified 2026-08-01.