What the 2023 Google Cloud Paris outage teaches about data center water risk
On April 25, 2023, a cooling-water pipe started leaking at a data center in Paris. Within hours, water had reached a battery room, a fire broke out and Google Cloud’s europe-west9 region went offline for more than a day.
The timeline
Google’s public incident report lays out how quickly the situation escalated. Times are US/Pacific:
- 4:46 pm — a cooling-system water pipe begins to leak.
- 7:14 pm — smoke is detected.
- 8:08 pm — the battery room is flooded and on fire.
- 11:04 pm — the building is shut down.
About six hours passed between the start of the leak and the shutdown. According to Google’s report, 58% of the Compute Engine instances in zone europe-west9-a went down, and the region stayed offline for more than a day.
Lesson 1: a leak is an uptime problem, not a facilities footnote
It is tempting to treat water as a housekeeping issue: mop it up, replace a fitting, move on. Paris shows how a single leak can travel to the one room where water does the most harm, turn into a fire and take down a cloud region. In liquid-cooled facilities, the cooling system and the power system share the same building, and sometimes the same paths.
Lesson 2: detection starts the clock too late
Most facilities rely on tools that alarm once liquid has escaped. By that point, the team is in emergency mode: find the source, isolate the loop, protect the electrical rooms. Every hour of warning before the leak starts is time to act calmly instead.
Lesson 3: the risk is growing with AI
AI racks increasingly rely on liquid cooling, which brings more pipes, CDUs, manifolds and quick-disconnects into the white space. Insurers have noticed: Allianz Commercial reports that water damage is the most frequent data center insurance claim, and FM notes that liquid damage is more likely than fire to hit a business.
What operators can do now
- Map where water can travel. Trace every liquid line and note which electrical, battery and UPS rooms sit below or beside it.
- Separate water from batteries. Review drainage, curbs and barriers around battery rooms in particular.
- Watch the signals that come before a leak. Pressure, flow and vibration often change hours or days before liquid escapes.
- Rehearse the response. Know who isolates which loop, and how fast.
We are not claiming any tool would have prevented this specific event. The point is simpler: early warning creates time to act, and time is what the Paris team didn’t have.
MayimFlow predicts liquid failures in AI data centers and infrastructure with 24–72 hours of warning. Learn how it works.