The Day AWS Went Dark: Inside the US-East-1 Outage
A DNS failure inside Amazon's Virginia data hub knocked Snapchat, Fortnite, Ring and banks offline for 15 hours. Here is what actually broke.
Just before 3 a.m. on the American east coast, the internet started to wobble. Snapchat wouldn't load. Fortnite lobbies froze mid-match. Ring doorbells stopped answering their owners. By sunrise on October 20, the wobble had a name: Amazon Web Services, and specifically its oldest and busiest region, US-East-1, tucked into the data center corridor of Northern Virginia.
What followed was one of the longest disruptions in AWS's history, a cascading failure that rippled far past Amazon's own apps and into the plumbing of services most people never think of as "the cloud" at all: banking apps, airline check-ins, game servers, even parts of Amazon's own retail operation. According to a report from NBC News, the disruption left leading games, publishers and streaming platforms unusable for millions of people before it was resolved by Monday evening.
The scale of it is what stood out. This wasn't a niche service hiccup. It was a reminder, delivered at 3 a.m., of how much of the visible internet quietly depends on one company's servers in one American state.
What actually broke inside AWS
The trigger, according to Amazon's own account, was not a hack or an external attack. It was internal: an error inside the automated DNS management system that AWS uses to keep its DynamoDB database service reachable. DynamoDB is the kind of unglamorous, foundational product that half the internet leans on without knowing it, storing shopping carts, session logins, game states, and app data for thousands of companies.
When the DNS records for DynamoDB's regional endpoint in US-East-1 went bad, client requests simply stopped resolving. Services that depended on DynamoDB began failing, and because AWS's own internal systems, from EC2 to Lambda to load balancers, also depend on each other, the failure spread sideways before anyone could contain it. Amazon published its own incident writeup at aws.amazon.com/message/101925, describing the chain reaction in engineering terms that amounted to one broken link pulling down a dozen others.
A detailed technical breakdown from the network monitoring firm ThousandEyes, published as AWS Outage Analysis: October 20, 2025, traced the failure to a race condition, essentially two automated processes updating the same DNS record at almost the same moment, with an older version overwriting a newer one. It sounds small. It was not.
Who felt it, and for how long
The outage began around midnight Pacific time and stretched for roughly 15 hours before AWS declared full recovery. A live-updating account from Tom's Guide tracked the casualty list as it grew through the morning: Snapchat, Fortnite, Ring, Alexa, Coinbase, Robinhood, Canva, Perplexity, Roblox, and Crunchyroll all reported disruptions, alongside banking and airline services that quietly route through AWS infrastructure without ever putting the Amazon name on their own front door.
Engineers could not simply flip a switch to fix it. Automated rollback systems, designed to catch bad configuration changes, instead reintroduced the same stale DNS state they were supposed to prevent, according to the ThousandEyes analysis. Human operators eventually had to step in and manually correct the records, a slower and more delicate process than most users waiting on a frozen loading screen would have guessed.
Why one region can take down so much
US-East-1 is not just another AWS region. It was Amazon's first cloud region, launched in 2006, and it has grown into the default home for an outsized share of AWS customers, including many of AWS's own internal control-plane services. That history means a bad day in Northern Virginia doesn't stay in Northern Virginia. Services hosted in other regions can still depend on US-East-1 for authentication, billing, or DNS resolution, so the blast radius extends well beyond the map.
It is worth being precise about what this was and was not. This was a software failure inside Amazon's own automation, not an intrusion, not a breach, and not evidence of anyone attacking AWS's servers. The distinction matters, because the instinct to describe a bad outage as an "attack" understates how fragile the underlying engineering can be even without an adversary in the picture.
What comes next
Amazon has promised a fuller post-incident review, and regulators, insurers and enterprise customers are already asking harder questions about how one region's DNS system could hold so much of the internet hostage for half a day. Coverage from outlets tracking the fallout, including a rundown at EM360Tech, suggests this outage will be studied for years the way the 2021 Fastly outage and the 2024 CrowdStrike incident already are.
For now, the practical lesson lands closer to home than most readers expect. If your banking app, your favorite game, or your smart doorbell went dark that Monday, it wasn't a coincidence of bad luck. It was a map of who depends on whom, drawn in real time by an outage nobody saw coming.
Published in The Outspoken Digest
Read Next
More Technology →
