AWS DNS failure took down Snapchat, banks and 1,000+ sites for most of a day

On October 20, 2025, a massive AWS outage knocked some of the world's largest websites offline for much of the day. More than 1,000 apps and websites were affected — Snapchat, Fortnite, Duolingo, Reddit, and banks including Lloyds and Halifax among them. Downdetector saw more than four million user reports within hours, more than double a normal full weekday, and reports eventually soared past 11 million globally.

Amazon pinned the root cause on 'DNS resolution of the DynamoDB API endpoint in US-EAST-1' — the phone-book layer that translates service names into machine-readable addresses. Disruptions there ripple: services couldn't find the databases they were built on.

Recovery was messy. Even after the underlying problem was addressed, Amazon had to throttle parts of its own system, and Notre Dame professor Mike Chapple warned of 'cascading failures' — like a power grid coming back that 'might flicker a few times' — if only symptoms were fixed first. AWS said all services 'returned to normal operations' around 23:00 BST, roughly sixteen hours after the first reports.

The postmortem conversation was about concentration, not just code. 'What this episode has highlighted is just how interdependent our infrastructure is,' said University of Surrey professor Alan Woodward; Cloudflare CEO Matthew Prince put it plainly: 'Everyone has a bad day, today Amazon had a bad day.'

More than 1,000 apps and websites affected for much of the day; Downdetector reports peaked above 11 million; Amazon had to throttle parts of its own system

A single region's DNS resolution failure propagated to over 1,000 customer services with no geographic containment.

The recovery required throttling Amazon's own systems — capacity planning had no margin for its own incident.

Cascading failures after the initial fix showed recovery tooling, not just the root cause, was under stress.

The 16-hour window for a control-plane fault demonstrated how much of the consumer internet sits on one provider.

The internet's default cloud has a default blast radius: when one region's control plane stumbles, a thousand unrelated products share the fall — and recovery means throttling yourself.

As of the BBC's October 21, 2025 report: AWS services had returned to normal; Amazon had not yet fully detailed the failure mechanism or issued an official postmortem; experts renewed warnings about single-provider dependence for critical services including banking.