But on 15 July, a Google Cloud data centre supporting its europe-west4-a zone suffered a power-related incident that eventually affected cooling and interrupted several services for nearly 15 hours.
The interesting detail is how small the original event sounds: Google says the root cause involved a 3 millisecond voltage drop in the utility power feed. Protective systems reacted, cooling was affected, temperatures increased and infrastructure had to be shut down to protect the hardware. A few milliseconds in the physical world eventually became hours of disruption in the digital one.
It is a useful reminder that abstraction doesn’t remove dependencies; it hides them. Behind every API, database, virtual machine and beautifully designed cloud dashboard there are still cables, switches, cooling systems, power supplies and buildings. Redundancy helps enormously, but redundancy itself has dependencies — and sometimes several supposedly independent systems can still share the same failure path.
This is also why architecture matters beyond simply choosing a reputable cloud provider. If an application genuinely cannot tolerate a regional or zonal failure, that requirement needs to exist in its design. “It’s in the cloud” isn’t a disaster recovery strategy.
My take: I actually like incidents like this from an engineering perspective because they expose the layers we normally forget about. We keep making software more abstract, automated and intelligent, but eventually everything reaches something physical. Sometimes the most sophisticated system in the world can still have a very simple problem: no power, too much heat, or a cable that isn’t doing what it should.
