Cloud Outage Analysis

AWS Outage Cost - History, Real Losses & Your Exposure

Updated July 2026 · 9 major incidents documented

AWS Outage History, In Brief

AWS has had 9 major documented outages since 2012, and 8 of the 9 struck the us-east-1 (Northern Virginia) region. The most recent was July 16, 2026, a roughly 3.5-hour global CloudFront failure in the VPC Origins control plane - a capacity limit in a Frankfurt Availability Zone (euc1-az2) stopped edge processors loading routing config and served worldwide 5xx errors, taking Canvas, Blackboard, Hugging Face, and the UK National Lottery offline. The longest single stretch of AWS downtime remains May 2026, a roughly 28-hour single-availability-zone (use1-az4) failure caused by a data-center cooling/thermal event that knocked EC2 and EBS offline and took Coinbase (about 7 hours), FanDuel, and CME Group down. The largest region-wide event was October 2025, a roughly 15-hour failure triggered by a DynamoDB DNS race condition that cascaded across EC2, Lambda, and STS and affected around 70,000 organizations. The two costliest were the December 2021 7-hour us-east-1 outage and the February 2017 S3 outage, each with aggregate customer losses estimated at $150M+. Significant AWS incidents have occurred at a rate of roughly one per year over the recent 2017-2026 window. Full chronology and SLA credit math below.

AWS Outage & Downtime Statistics

9

Major outages since 2012

~1/yr

Incident frequency (2017-2026)

8 / 9

In us-east-1 (region concentration)

~28h

Longest single event (May 2026)

$150M+

Costliest (Dec 2021, Feb 2017)

Frequency reflects the 9 major incidents documented in the chronology below, not every status-page event. Region concentration counts documented major outages: 8 of 9 struck us-east-1; the exception is the July 2026 CloudFront VPC Origins event, which originated in a Frankfurt AZ (euc1-az2) but degraded customers globally. AWS has had smaller incidents in other regions not listed here.

AWS Outage History

DateRegionServicesDurationImpactEst. Cost
Jul 2026Global (euc1-az2 origin)CloudFront (VPC Origins control plane)~3.5 hoursGlobal 5xx; Canvas, Blackboard, Hugging Face, UK National Lottery offline$15M+ est.
May 2026us-east-1 (use1-az4)EC2, EBS (cooling/thermal event); ELB, EKS, Redshift, MSK~28 hoursSingle AZ; Coinbase ~7h, FanDuel, CME Group offline$25M+ est.
Oct 2025us-east-1DynamoDB DNS, cascaded EC2/Lambda/STS~15 hours~70K orgs, 17M+ reports$38-581M insured est.
Jun 2023us-east-1Lambda (capacity subsystem); cascaded STS, Console, EKS, Connect, EventBridge~3 hours104 services degraded; ~2h core Lambda, ~4h full recovery$25M+ est.
Dec 2021us-east-1EC2, ECS, Lambda, SNS~7 hoursMajor global disruption$150M+ est.
Nov 2020us-east-1Kinesis (cascading)~17 hoursCascaded to CloudWatch, Cognito, Lambda, EventBridge$50M+ est.
Aug 2019us-east-1EC2, EBS (AZ power failure)~8 hoursSingle-AZ power/generator failure; some EBS data permanently lost$25M+ est.
Feb 2017us-east-1S3~4 hoursSignificant internet-wide disruption$150M+ est.
Jun 2012us-east-1EC2, RDS, EBS~8 hoursNetflix, Instagram, Pinterest affected$100M+ est.

Cost estimates are aggregate customer losses except where noted. AWS does not disclose customer impact figures. The October 2025 figure is CyberCube's preliminary insured-loss range ($38M-$581M, clustering toward the lower end); other estimates are based on analyst reports and revenue-at-risk modeling. The July 2026 CloudFront event (07:45-11:18 UTC, ~3.5h) affected only the VPC Origins feature; no analyst loss figure has been published, so its cost is a conservative revenue-at-risk estimate reflecting the ~3.5-hour global outage at Canvas, Blackboard, Hugging Face, and the UK National Lottery (per AWS's status updates and The Register). The May 2026 event was confined to a single Availability Zone (use1-az4) after a data-center cooling/thermal event; likewise no firm analyst figure was published, so its cost is a conservative revenue-at-risk estimate reflecting the multi-hour outages at Coinbase, FanDuel, and CME Group. Updated July 2026.

AWS SLA Credit Rules

ServiceSLA %Credit (99% to SLA)Credit (<99%)Claim Window
EC299.99%10% of monthly fee30% of monthly fee30 days
S3 Standard99.9%10%25%30 days
RDS Multi-AZ99.95%10%25%30 days
Lambda99.95%10%25%30 days

Credits applied to future invoices. Business or Enterprise support required. Full SLA credit asymmetry analysis.

Frequently Asked

When was the most recent AWS outage?
The most recent major AWS outage was July 16, 2026: a ~3.5-hour (07:45-11:18 UTC) global CloudFront VPC Origins control-plane failure, triggered by a capacity limit in a Frankfurt AZ (euc1-az2) that stopped edge processors loading routing config and served worldwide 5xx errors, taking Canvas, Blackboard, Hugging Face, and the UK National Lottery offline. Before that, May 2026 was the longest AWS outage on record (~28h single-AZ us-east-1 thermal event), and October 2025 was the largest region-wide event (~15h us-east-1 DynamoDB, ~70,000 orgs).
How many major AWS outages have there been?
At least 9 major documented outages since 2012, at roughly one significant incident per year over the recent 2017-2026 window. Almost all struck us-east-1 (Northern Virginia): the May 2026 thermal event, October 2025 DynamoDB outage, June 2023 Lambda event, December 2021 (~7h), November 2020 Kinesis (~17h), and February 2017 S3 (~4h). The exception is the July 2026 CloudFront VPC Origins outage, which originated in a Frankfurt AZ but degraded customers worldwide.
How often does AWS have a major outage?
Across the 9 major outages documented here since 2012, that is roughly one significant incident every 18-19 months overall, and closer to one per year over the recent 2017-2026 window. This counts only major, widely-felt events - not the smaller service degradations on the AWS status page, which are more frequent.
Which AWS region has the most outages?
us-east-1 (Northern Virginia) by a wide margin - 8 of the 9 major outages in this history struck it. It is AWS's oldest and largest region, and many global control-plane services depend on us-east-1 even when your workload runs elsewhere. The one exception, the July 2026 CloudFront VPC Origins outage, originated in a Frankfurt AZ (euc1-az2) yet still degraded customers worldwide. AWS has had smaller incidents in other regions, but the region-wide, customer-visible events cluster in us-east-1.
What was the worst AWS outage?
By aggregate customer loss, the December 2021 us-east-1 outage (~7h) and the February 2017 S3 outage (~4h) were the costliest, each estimated at $150M+ in aggregate customer losses. By duration, the May 2026 thermal event was the longest at ~28 hours, though confined to a single Availability Zone.
How much does an AWS outage cost businesses?
AWS outage costs depend on your revenue and architecture. The December 2021 us-east-1 outage lasted ~7 hours and caused aggregate customer losses estimated at $150M+. A business running $100K/day on AWS-dependent revenue would lose $29,167/hr during a total outage. Multi-AZ architecture reduces exposure by 60-80% for most failure modes.
Why does us-east-1 have so many outages?
us-east-1 is AWS's original and largest region with the most complex dependencies. It affects more customers when it fails. AWS's control plane and many managed services have us-east-1 dependencies even when deployed in other regions.
What is the AWS SLA credit for an outage?
AWS SLA credits are 10% of your monthly bill for the affected service when monthly uptime falls below the SLA target, rising below 99% to 30% for EC2 (Compute) and 25% for S3, RDS, and Lambda. A $1,000/month EC2 spend during a month with a 7-hour outage receives $100-$300 in SLA credit - not related to your revenue loss.

Related

AWS July 2026 Deep Dive

CloudFront VPC Origins global 5xx event

AWS May 2026 Deep Dive

use1-az4 thermal event, hour by hour

AWS Oct 2025 Deep Dive

DynamoDB DNS outage, hour by hour

AWS SLA Credit Math

What you actually get reimbursed

Azure Outage History

Compare Azure reliability

GCP Outage History

Compare GCP reliability

Updated 2026-04-27