Update — 24 July 2026: incident closed end to end
On the Equinix side, the corrective maintenance on PA4's chillers was completed between 16 and 24 July 2026 (fan replacement, condenser cleaning). The datacenter's cooling capacity, and its N+3 redundancy, are now fully restored; no pending repair remains: the incident is closed on both the RDEM Systems side and the datacenter side.
On 24 June 2026 at 15:06, a cooling incident occurred at the Equinix PA4 datacenter (Pantin), one of the three sites where we operate our infrastructure. The incident originated with our datacenter, not our own equipment. We are publishing here, transparently, how we handled it.
Our priority was simple: protect hardware and data rather than keep services running at all costs under degraded environmental conditions. Our services returned to nominal operation on 25 June at 08:30, redundancy included. On the datacenter side, the last cooling unit has since been brought back to full capacity: following Equinix's corrective maintenance, completed on 24 July 2026, PA4's cooling capacity is fully restored, N+3 redundancy included.
In short
A water-supply failure on Equinix PA4's adiabatic cooling system caused the protective shutdown of several chillers and a rise in room temperatures. In line with our incident management plan, and after coordinating with the affected customers, we failed over redundant services to their continuity site, migrated single-site machines between clusters where possible, then performed a controlled shutdown of the vast majority of our PA4 equipment. No data loss or uncontrolled outage occurred.
The incident is resolved on the RDEM Systems side and for our customers: all of our PA4 services returned to nominal operation on 25 June at 08:30, redundancy included. The last repair on the Equinix side (one chiller to be brought back to full capacity) was completed on 24 July 2026: PA4's cooling capacity — N+3 redundancy included — is now fully restored, with no impact on our services.
Timeline
| Timestamp | Event |
|---|---|
| 24 Jun · 15:06 | Cooling incident detected, handled and communicated by Equinix at PA4. The cause is not yet known at this stage. |
| 24 Jun · 15:46 | Equinix communicates a suspected initial cause: loss of water supply to the adiabatic cooling system, which triggered a protective shutdown of the compressors. |
| 24 Jun · ~16:00 | Our temperature graphs begin to rise in the rooms, after the incident was detected. |
| 24 Jun · 20:30 | Our thermal alarm thresholds reached. Incident management plan activated. |
| 24 Jun · 20:45 – 22:00 | Coordination with the affected customers to decide, together, on the precautionary measures to apply. |
| 24 Jun · 22:00 – 23:00 | Precautionary measures carried out: critical services (which must not be interrupted) relocated to the other datacenters we operate, redundant services failed over to their continuity site, then pre-production and non-critical environments shut down. |
| 24 Jun · 23:00 | Controlled shutdown of the vast majority of our PA4 equipment to protect hardware. |
| Night · 23:00 – 07:00 | Temperature monitoring. With the precautionary measures in place, no immediate risk to hardware remained: no further intervention was required. |
| 25 Jun · 02:00 – 04:00 | Datahalls back to normal temperature conditions — confirmed in Equinix's 25 Jun 04:00 communication. |
| 25 Jun · 07:30 | Temperatures deemed stable enough: start of the restart and gradual restoration of our services. |
| 25 Jun · 08:30 | All our services back to nominal operation — incident closed on the RDEM Systems side. |
| Day after the incident | Redundancy back to normal on the datacenter side; five chillers at full capacity, one at 50% (awaiting parts) before returning to full cooling capacity. |
| 16 – 24 Jul 2026 | Equinix corrective maintenance on the affected chillers (fan replacement, condenser cleaning), completed on 24 July 2026: PA4's cooling capacity fully restored, N+3 redundancy included. |
Network impact: one of our transit providers present at PA4 had its equipment shut down due to the temperature. Thanks to our multi-site, multi-transit connectivity via our AS206014, traffic kept flowing through our other points of presence, with no loss of connectivity.
Our response
Our architecture across three interconnected Equinix datacenters gave us real continuity options. Here is what we activated:
Failover
Services with a continuity plan were failed over to their backup site, away from PA4.
Inter-cluster migration
For single-datacenter machines, migration to other clusters where the situation allowed, to keep services running.
Controlled shutdown
Clean, orderly power-off of the remaining PA4 equipment to avoid any heat-related hardware damage.
What this incident demonstrates
Thermal monitoring with alarm thresholds defined in advance, not endured.
A proactive decision: protect hardware rather than risk an uncontrolled outage.
Communication with customers before acting, not after.
Real continuity options thanks to a multi-datacenter architecture.
Precautionary measures considered well enough to bring risk down to an acceptable level, without depending on a permanent human presence.
After the precautionary measures were carried out in the evening, no further action was required during the night: the risk to hardware had been brought down to an acceptable level. Services were simply brought back online the next morning, once thermal conditions returned to satisfactory levels.
Cause of the incident
According to our datacenter's communications, the suspected initial cause is a loss of water supply to the adiabatic cooling system, which triggered high-pressure conditions and the protective shutdown of several chillers' compressors. Recovery and repair work on the Equinix side was completed on 24 July 2026.
We do not reproduce the datacenter's internal operational detail here. The consolidated root-cause analysis (RCA) will be published once available and confirmed.
Our transparency commitment
- We only state "no impact" when it is objectively true for the affected customers.
- On 25 June, our PA4 services returned to nominal operation at 08:30, redundancy included; the datacenter's cooling capacity (N+3 redundancy) was fully restored on 24 July 2026.
- This page is updated as the situation evolves.
Acknowledgements
We thank the Equinix PA4 teams in Pantin for their regular communication throughout the incident. The regular updates and the technical information shared throughout the incident let us track the situation precisely and adapt our operational decisions.
We also commend the work of the on-site technicians who, despite particularly difficult conditions caused by the extreme heat, gradually restored the situation.
A well-managed incident is not an incident that never happened. It is an incident that is documented, communicated, and whose lessons serve to improve operations.
Infrastructure built for continuity
Three interconnected Equinix datacenters, high-availability clusters and proven incident procedures. Let's talk about the resilience of your infrastructure.