Back to the archive

The encyclopedia · Software & IT · Technical decision · 2026

Google Cloud's Delhi data-centre fire left India's traffic degraded for three weeks

A fire at a third-party Delhi colo forced Google to cut its network there, and customers saw latency spikes for weeks.

Google Cloud · 2026-06-09

What happened

On 9 June 2026, a fire at a third-party data centre in Delhi forced Google Cloud to shut down network equipment and reduce capacity in the region. Google isolated the Delhi point of presence and rerouted traffic, but demand exceeded the available capacity, and customers in Delhi, Chennai, Mumbai and surrounding areas saw latency spikes and possible packet loss.

The outage hit Google's Hybrid Connectivity and VPC services, which underpin how enterprises connect on-premises networks to Google Cloud. Google's status page tracked the incident from 5 June to 26 June, meaning customers carried degraded network performance for roughly three weeks while the company rebuilt capacity.

The failure was a resilience decision: Google had concentrated a point of presence in a single third-party facility, so a fire there cut the backbone links that carried India's traffic. Rerouting was only a partial answer because the surviving network could not absorb the demand.

Google said it would optimise backbone capacity, expand the Delhi point of presence and migrate peering to reduce the risk of a repeat. The episode showed how a single physical facility failure in one market could degrade a cloud region for weeks, even with traffic rerouting in place.

Why it happened

  • A single third-party facility hosted a Google point of presence, so a fire there cut the backbone links for an entire region
  • Rerouting assumed the surviving network could absorb demand, but capacity was already near its limit
  • Recovery took weeks because rebuilding backbone capacity and migrating peering is slow work, not a failover toggle
  • The incident fed fears of data loss even though the fire was contained, widening the reputational cost
What it costthree weeks of degraded network across India's cloud regioncostly

The lesson

A single point of presence in one shared facility is a whole region's single point of failure — reroute only works if the surviving network can carry the load.

Sources

spotted an error? The club wants to know.

Comments · 0

    Sign in to join the comments.

    More like this

    Somewhere, someone solved the problem this company failed at. 2nd Opinion →