---
title: "Cloud Outage Statistics 2026: AWS, Azure and Google Cloud Outage History"
date: 2026-11-17
author: "Brad Nickel"
featured_image: "https://tricitylocalnews.com/wp-content/uploads/2026/10/cloud-outage-statistics-aws-azure-google-cloud-outage-history.jpg"
categories:
  - name: "Cloud"
    url: "/cloud.md"
tags:
  - name: "Statistics"
    url: "/tag/statistics.md"
---

# Cloud Outage Statistics 2026: AWS, Azure and Google Cloud Outage History

Amazon Web Services (AWS) says the service disruption in its N. Virginia (us-east-1) Region started at 11:48 p.m. PDT on October 19 and ended at 2:20 p.m. PDT on October 20, 2025. That puts the AWS outage at about **14.5 hours** from start to end, against a total duration of **2 hours and 22 minutes** for multiple Google Cloud services in the us-west1 region on August 20, 2026. Parametrix, which monitors cloud provider availability, estimates that the financial losses for US companies resulting from the AWS event were between **$500 million** and **$650 million**.

On Azure’s status history page, Post Incident Reviews (PIRs) of previous service issues are each retained for **5 years**, and AWS keeps its public Post-Event Summaries available for a minimum of **5 years**. Every incident duration below comes from the start and end times a provider published, kept in the time zone that provider used, so the incidents can be compared on the providers’ own records. The Parametrix phase lengths are the exception, as they come from its own monitoring.

## Key Takeaways

- AWS’s October 2025 us-east-1 event ran about **14.5 hours** by AWS’s own start and end times, and there were three distinct periods of impact to customer applications.
- AWS traced the outage to its Domain Name System (DNS) automation, a latent race condition in the DynamoDB DNS management system; DynamoDB application programming interface (API) error rates rose between 11:48 p.m. on October 19 and 2:40 a.m. on October 20, a window of **2.9 hours**.
- **11** of the **18** Post-Event Summaries listed on AWS’s index concern the us-east-1 (Northern Virginia) region.
- For instances deployed across two or more Availability Zones (AZs), the Service Level Agreement (SLA) for Amazon Elastic Compute Cloud (EC2) sets a Monthly Uptime Percentage of at least **99.99%** per region, with a Service Credit when uptime is less than that, which leaves about **4.32 minutes** of downtime in a 30-day month.
- Roughly **30%** of all providers monitored by Parametrix experienced disruptions during the AWS event, almost double what it detected from the Google Cloud Platform (GCP) us-central1 outage in June 2025.
- Uptime Institute’s 2026 outage analysis reports that in Uptime’s 2025 Annual Survey, **57%** of respondents said their most recent major outage cost more than $100,000.

## Editor’s Choice

- Windows devices affected by CrowdStrike: Microsoft estimated that CrowdStrike’s update affected **8.5 million** Windows devices.
- Fortune 500 loss from the CrowdStrike outage: Parametrix estimates **$5.4 billion** in total direct financial loss for US Fortune 500 companies, excluding Microsoft.
- Delta’s direct revenue impact: approximately **$380 million**, alongside **7,000** flight cancellations over the five-day period.
- Google Cloud, June 12, 2025: a Duration of **3 hours**, Global, with recovery in most regions within **2 hours**.
- Azure Front Door, October 29 to 30, 2025: connection timeout errors and DNS resolution issues from 15:41 UTC to 00:05 UTC, about **8.4 hours**.
- EC2 service credit below 95.0% uptime: a **100%** Service Credit Percentage when the Monthly Uptime Percentage is less than 95.0%, against **30%** when it is less than 99.0% but equal to or greater than 95.0%.

## Major Cloud Outages by Date

- At 9:37 a.m. PST on February 28, 2017, an authorized Amazon Simple Storage Service (S3) team member executed a command; one of its inputs was entered incorrectly, and a larger set of servers was removed than intended; S3 was operating normally again about **4.3 hours** later.
- At 7:30 a.m. PST on December 7, 2021, an automated activity to scale capacity triggered a large surge of connection activity that overwhelmed the networking devices between AWS’s internal network and its main network in the Northern Virginia (us-east-1) Region, and the network devices took about **6.9 hours** to fully recover.
- Lambda function invocations in us-east-1 saw increased error rates and latencies from 11:49 a.m. PDT on June 13, 2023, and all affected services had fully recovered by 3:37 p.m. PDT.
- CrowdStrike released a content configuration update for the Windows sensor at 04:09 UTC on July 19, 2024, and the update resulted in a Windows system crash; Windows hosts on sensor 7.11 and above that were online during that roughly **1.3-hour** window, until 05:27 UTC, and received the update were in scope.
- Delta Air Lines’ direct revenue impact and non-fuel expense impact from the CrowdStrike-caused outage add up to about **$550 million**.
- Google’s incident report says the June 12, 2025 incident started at 10:49 US/Pacific, with increased 503 errors in external API requests across multiple Google Cloud and Google Workspace products.
- Multiple Google Cloud services in us-west1 had elevated latency, provisioning failures, and increased error rates from 08:00 to 10:22 US/Pacific on August 20, 2026, about **2.4 hours**.

| Date | Provider | Region or scope | Trigger (provider’s account) | Impact window (provider’s time zone) |
|---|---|---|---|---|
| February 28, 2017 | AWS (Amazon S3) | us-east-1 | Command input entered incorrectly | 9:37 AM to 1:54 PM PST |
| December 7, 2021 | AWS | us-east-1 | Automated capacity scaling triggered a connection surge | 7:30 AM to 2:22 PM PST (to network-device recovery) |
| June 13, 2023 | AWS Lambda | us-east-1 | Increased Lambda error rates and latencies | 11:49 AM to 3:37 PM PDT |
| July 19, 2024 | CrowdStrike Falcon | Windows hosts on sensor 7.11 and above | Content configuration update | 04:09 to 05:27 UTC (update window) |
| June 12, 2025 | Google Cloud | Global | Invalid automated quota update | 10:49 to 13:49 US/Pacific |
| October 19 to 20, 2025 | AWS (DynamoDB and dependent services) | us-east-1 | Latent race condition in DynamoDB DNS management | 11:48 PM PDT October 19 to 2:20 PM PDT October 20 |
| October 29 to 30, 2025 | Microsoft Azure | Azure Front Door and Azure CDN | Incompatible customer configuration metadata | 15:41 to 00:05 UTC |
| August 20, 2026 | Google Cloud | us-west1 | Scheduled fiber optic maintenance | 08:00 to 10:22 US/Pacific |

*Source: Amazon Web Services, Google Cloud, Microsoft Azure, CrowdStrike, 2017 to 2026*

### About This Data

- Compiled from 20 sources: 13 Amazon Web Services, Google Cloud and Microsoft Azure incident reports, status pages and SLA pages, a CrowdStrike incident review, a Microsoft blog post, a Delta Air Lines filing with the SEC, 3 Parametrix analyses and a Uptime Institute research release.
- The incidents run from February 2017 to September 2026, and every page was read in October 2026.
- Only the providers’ own reports, a company filing, and named research firms qualified; news coverage and outage-tracking sites were left out.
- Figures are reviewed on a rolling basis and updated when providers publish new reports.

## How Long Cloud Outages Last

- AWS’s October 2025 us-east-1 event lasted about **14.5 hours**, from 11:48 p.m. PDT to 2:20 p.m. PDT the next day.
- In December 2021, the us-east-1 event ran about **6.9 hours** to network-device recovery, from the 7:30 a.m. PST trigger until all network devices fully recovered at 2:22 p.m. PST; full recovery for Amazon Secure Token Service (STS) occurred at 4:28 p.m. PST.
- Google’s report gives its June 2025 incident a Duration of **3 hours**, while the incident header says it began at 10:51 and ended at 18:18 US/Pacific, a span of about **7.5 hours**.
- Google puts its August 2026 us-west1 incident at a total duration of **2 hours and 22 minutes**.

 LENGTH · Source: Amazon Web Services, Microsoft Azure, Google Cloud, 2017 to 2026   LENGTH    Amazon Web Services · 2017 to 2026 Cloud Outage Length by Incident (hours)         AWS us-east-1, October 2025  14.5 Azure Front Door, October 2025  8.4 AWS us-east-1, December 2021 (to network-device recovery)  6.9 AWS S3, February 2017  4.3 AWS Lambda, June 2023  3.8 Google Cloud global, June 2025  3 Google Cloud us-west1, August 2026  2.4  0 3 6 9 12 15   SOURCE Amazon Web Services, Microsoft Azure, Google Cloud, 2017 to 2026    View the data| Incident | Length (hours) |
|---|---|
| AWS us-east-1, October 2025 | 14.5 |
| Azure Front Door, October 2025 | 8.4 |
| AWS us-east-1, December 2021 (to network-device recovery) | 6.9 |
| AWS S3, February 2017 | 4.3 |
| AWS Lambda, June 2023 | 3.8 |
| Google Cloud global, June 2025 | 3 |
| Google Cloud us-west1, August 2026 | 2.4 |





In February 2017, the S3 disruption ran about **4.3 hours**, from 9:37 a.m. PST until the placement subsystem finished recovery at 1:54 p.m. PST. In June 2023, the Lambda event took **3.8 hours** to full recovery, and Google’s August 2026 incident works out to about **2.4 hours**. Google’s two figures for that June incident cover different windows: 10:49 to 13:49 US/Pacific in the incident report and 10:51 to 18:18 in the incident header.

### Which cloud outage lasted the longest?

Among the seven provider-documented incidents in the chart, the October 2025 AWS event in us-east-1 ran longest. It lasted about **14.5 hours** by AWS’s timestamps, against about **8.4 hours** for Azure Front Door’s October 2025 incident in UTC. Events outside the providers’ public reports fall outside this comparison.

## Recent Developments

- **September 30, 2026:** Between 20:30 UTC on September 30, 2026 and 02:15 UTC on October 1, 2026, a subset of customers using gateway services in multiple regions experienced degraded or interrupted network connectivity, including Azure ExpressRoute Gateway and Azure VPN Gateway.
- **September 29, 2026:** Azure OpenAI Service, Foundry Agent Service, Foundry Models, and Cognitive Services in the Sweden Central region had intermittent request failures and HTTP 5XX errors between 10:03 UTC and 15:58 UTC.
- **September 15, 2026 (status-feed update):** The AWS Health Dashboard feeds list a Service disruption item for Amazon Elastic Compute Cloud in me-central-1 and in me-south-1, each impacted by an event that affects multiple services; the two feeds are titled UAE and Bahrain, and the date is when each feed item was posted, not a start time for the event.
- **August 20, 2026:** Google Cloud services in us-west1 were degraded for a total duration of **2 hours and 22 minutes**, after scheduled fiber optic maintenance unexpectedly compromised network capacity between data centers.
- **July 23, 2026:** A subset of customers experienced connectivity failures, increased latency, or difficulty accessing Azure services in the West US region between 14:44 and 19:41 UTC.

## AWS Outage October 2025 (us-east-1)

- Network Load Balancer (NLB) connection errors ran from 5:30 a.m. to 2:09 p.m. on October 20, caused by health check failures in the NLB fleet, a window of about **8.7 hours**.
- New EC2 instance launches failed between 2:25 a.m. and 10:36 a.m. on October 20, and some newly launched instances had connectivity issues that were resolved by 1:50 p.m.
- The root cause was a latent race condition in the DynamoDB DNS management system that resulted in an incorrect empty DNS record for the service’s regional endpoint, dynamodb.us-east-1.amazonaws.com, that the automation failed to repair.
- Lambda customers saw API errors and latencies from 11:51 p.m. PDT on October 19 to 2:15 p.m. PDT on October 20, about **14.4 hours**.
- Amazon Elastic Container Service (ECS), Elastic Kubernetes Service (EKS), and Fargate had container launch failures and cluster scaling delays from 11:45 p.m. PDT on October 19 until 2:20 p.m. PDT on October 20, about **14.6 hours**. That window opened at 11:45 p.m., three minutes before the 11:48 p.m. start AWS gives for the whole event.

 IMPACT WINDOW · Source: Amazon Web Services, October 2025   IMPACT WINDOW    Amazon Web Services · 2025 AWS Impact Window by Service, October 2025 (hours)         ECS, EKS and Fargate  14.6 Lambda  14.4 Network Load Balancer  8.7 EC2 instance launches  8.2 DynamoDB API errors  2.9  0 3 6 9 12 15   SOURCE Amazon Web Services, October 2025    View the data| Service | Impact window (hours) |
|---|---|
| ECS, EKS and Fargate | 14.6 |
| Lambda | 14.4 |
| Network Load Balancer | 8.7 |
| EC2 instance launches | 8.2 |
| DynamoDB API errors | 2.9 |





Failed EC2 launches lasted **8.2 hours**, while the DynamoDB API error window itself lasted **2.9 hours**. Every window comes from AWS’s post-event summary. AWS ties Lambda’s early errors to DynamoDB endpoint issues, yet Lambda customers saw errors long after the DynamoDB error window closed. AWS attributes the load-balancer errors to health check failures in the NLB fleet.

## AWS Outage History

- The oldest entry is the Summary of the Amazon EC2 and Amazon RDS Service Disruption in the US East Region, dated **April 21, 2011**.
- The newest is the Summary of the Amazon DynamoDB Service Disruption in the Northern Virginia (US-EAST-1) Region, dated **October 19, 2025**.
- **11** summaries carry a US-EAST-1, US-East, or US East region in their titles.
- Outside the US, the index lists the Tokyo (AP-NORTHEAST-1) Region and the Asia Pacific (Seoul) Region (AP-NORTHEAST-2), plus the Sydney Region, the EU West Region, and the South America Region (SA-EAST-1).
- On December 7, 2021, Route 53 APIs were impaired from 7:30 a.m. PST until 2:30 p.m. PST, but existing DNS entries and answers to DNS queries were not impacted, a span of **7 hours**.
- Console access was fully restored by 2:22 p.m. PST that day.
- In the February 2017 S3 event, GET, LIST and DELETE APIs were functioning normally by 1:18 p.m. PST, about **3.7 hours** after the 9:37 a.m. PST command.

 POST-EVENT SUMMARIES · Source: Amazon Web Services Post-Event Summaries index, 2011 to 2025   POST-EVENT SUMMARIES    Amazon Web Services · 2011 to 2025 AWS Post-Event Summaries by Region (count)         US-East (Northern Virginia)  11 Tokyo  2 Seoul  1 Sydney  1 EU West  1 South America  1 No region in title  1  0 3 6 9 12 15   SOURCE Amazon Web Services Post-Event Summaries index, 2011 to 2025    View the data| Region named in the title | Post-Event Summaries |
|---|---|
| US-East (Northern Virginia) | 11 |
| Tokyo | 2 |
| Seoul | 1 |
| Sydney | 1 |
| EU West | 1 |
| South America | 1 |
| No region in title | 1 |





> **Key finding:** **11** of the **18** Post-Event Summaries AWS lists from April 2011 to October 2025 concern its us-east-1 region in Northern Virginia. AWS publishes one when an issue has broad and significant customer impact, so the index records only the events that clear that bar, not every regional fault.

## Where to Check an AWS Outage Today

- AWS commits to a public Post-Event Summary following the closure of an issue that causes the failure of a significant percentage of control plane API calls, or that results from total power failure or significant network failure.
- Each summary provides the issue’s scope of impact, the factors that contributed to it, and the actions taken to address the risks identified.
- Azure’s history page has carried PIRs for all issues communicated publicly from **November 20, 2019**, and PIRs for broad issues from **June 1, 2022**.

| Provider record | When the provider publishes it | How long it stays available |
|---|---|---|
| AWS Post-Event Summary | After an issue with broad and significant customer impact closes | Minimum of 5 years |
| Azure Post Incident Review | Broad issues, from June 1, 2022 | 5 years |

*Source: Amazon Web Services, Microsoft Azure, 2026*

### Is Amazon having server issues right now?

A statistics page cannot report live status, so the AWS Health Dashboard is the record to check during an incident. Its feed for Amazon Elastic Compute Cloud in me-central-1 carried a Service disruption item dated **September 15, 2026**. Post-Event Summaries arrive only after an issue closes, so they explain an outage after the fact. Azure’s history page shows how often those reviews appear.

## Azure Outage History

- Between 15:41 UTC on October 29 and 00:05 UTC on October 30, 2025, customers and Microsoft services using Azure Front Door (AFD) and Azure Content Delivery Network (CDN) experienced connection timeout errors and DNS resolution issues, a window of about **8.4 hours**.
- From 18:30 UTC on October 29, some customers started to see availability improve, with increased latency, until the system fully stabilized, about **2.8 hours** after impact began.
- Microsoft’s review says a specific sequence of customer configuration changes, performed across two different control plane build versions, resulted in incompatible customer configuration metadata that exposed a latent bug in the data plane.
- On July 23, 2026, connectivity failures in the West US region ran from 14:44 to 19:41 UTC, and traffic that remained entirely within the region was not affected, a window of **4.95 hours**.
- Azure’s review of that event says a defect in the blast radius analysis system incorrectly expanded the scope of a break-fix repair to include all optical devices egressing a specific datacenter.
- Azure OpenAI Service and the Foundry services in Sweden Central returned HTTP 5XX errors between 10:03 UTC and 15:58 UTC on September 29, 2026.
- Gateway services, including Azure ExpressRoute Gateway, Azure Firewall, and Azure VPN Gateway, were affected from 20:30 UTC on September 30, 2026, to 02:15 UTC on October 1, 2026.

 LENGTH · Source: Microsoft Azure status history, July to October 2026   LENGTH    Microsoft Azure · 2026 Azure Incident Length in 2026 Reviews (hours)         Sweden Central AI services, September 29  5.92 Gateway services, September 30  5.75 West US connectivity, July 23  4.95  0 1.2 2.4 3.6 4.8 6   SOURCE Microsoft Azure status history, July to October 2026    View the data| Incident | Length (hours) |
|---|---|
| Sweden Central AI services, September 29 | 5.92 |
| Gateway services, September 30 | 5.75 |
| West US connectivity, July 23 | 4.95 |





Sweden Central’s incident lasted **5.92 hours**, the gateway incident **5.75 hours**, and the West US incident **4.95 hours**, each counted from the UTC times in Azure’s reviews. The Azure Front Door review covers a longer window than any of these three. Google’s incident reports state their durations outright.

## Google Cloud Outage History

- On June 12, 2025 at ~10:45 am PDT, a policy change was inserted into the regional Spanner tables that Service Control uses for policies, and it was replicated globally within seconds.
- The policy data contained unintended blank fields, which hit a null pointer and sent the Service Control binaries into a crash loop in each regional deployment.
- Google mitigated all regions except us-central1 by 12:48, and the quota policy database in us-central1 became overloaded, resulting in much longer recovery in that region; that mitigation came about **2 hours** after the 10:49 start.
- Parametrix measured API failure rates for core services in us-central1 approaching **100%** for **2 hours and 35 minutes**.
- In the August 2026 incident, the impact covered Persistent Disk, Google Kubernetes Engine, Google Compute Engine, Cloud Run, Google Cloud Bigtable, Cloud Storage, and Identity and Access Management.
- Automated rerouting mechanisms failed to redistribute traffic to alternate capacity, causing network congestion as volumes exceeded available bandwidth.

| Date | Region or scope | Duration (Google’s report) | Trigger |
|---|---|---|---|
| June 12, 2025 | Global, us-central1 recovered last | 3 hours | Invalid automated quota update |
| August 20, 2026 | us-west1 | 2 hours and 22 minutes | Scheduled fiber optic maintenance |

*Source: Google Cloud Service Health incident reports, 2025 and 2026*

Google’s August 2026 incident report also gives the window in UTC, from 15:00 to 17:22 UTC. The next incident in the set came from a security vendor rather than a cloud provider.

## CrowdStrike Outage Statistics

- Systems in scope were Windows hosts running sensor version 7.11 and above that were online between 04:09 UTC and 05:27 UTC on July 19, 2024, and received the update, a window of about **1.3 hours**.
- Mac and Linux hosts were not impacted.
- CrowdStrike says the defect in the content update was reverted at 05:27 UTC, and systems coming online after this time were not impacted.
- Microsoft estimated that CrowdStrike’s update affected **8.5 million** Windows devices, and said the broad economic and societal impacts reflect the use of CrowdStrike by enterprises that run many critical services.
- Delta’s direct revenue impact from the CrowdStrike-caused outage was approximately **$380 million**, primarily from refunds for cancelled flights and customer compensation in cash and SkyMiles.
- Delta’s table of the outage’s impact on 3Q24 shows Operating Margin at (2.3) pts and Earnings Per Share at (45) cents, a **2.3-point** margin reduction and a **45-cent** reduction per share.

 AMOUNT · Source: Delta Air Lines, September quarter 2024 results   AMOUNT    Delta Air · 2024 Delta CrowdStrike Outage Impact, Q3 2024 ($ millions)          Direct revenue impact  $380M Non-fuel expense impact  $170M Fuel expense (lower)  -$50M  -100 0 100 200 300 400   SOURCE Delta Air Lines, September quarter 2024 results    View the data| Impact line | Amount ($ millions) |
|---|---|
| Direct revenue impact | 380 |
| Non-fuel expense impact | 170 |
| Fuel expense (lower) | (50) |





Delta’s [quarterly earnings release](https://www.sec.gov/Archives/edgar/data/0000027904/000168316824007033/delta_ex9901.htm) breaks out the rest: the non-fuel expense impact was **$170 million**, primarily due to customer expense reimbursements and crew-related costs, and fuel expense was **$50 million** lower because of the **7,000** flight cancellations over the five-day period.

### What caused the CrowdStrike outage?

CrowdStrike says the problematic Rapid Response Content configuration update resulted in a Windows system crash. It released the content configuration update for the Windows sensor at 04:09 UTC on July 19, 2024, as part of regular operations, to gather telemetry on possible novel threat techniques. Delta’s filing is one measure of cost; insurers model the rest.

## Cost of Cloud Outages

- Parametrix estimates the total direct financial loss facing the US Fortune 500 companies, excluding Microsoft, from the CrowdStrike outage on July 19, 2024, at **$5.4 billion**.
- Its estimate draws on more than **54 billion** data points and direct monitoring of the real-time service status of **6,000** leading technology businesses, a 2024 count that differs from the over **7,000** SaaS, PaaS and IaaS providers Parametrix cited for the October 2025 AWS event.
- Parametrix estimates that the financial losses for US companies resulting from the AWS us-east-1 event on October 20, 2025 were between **$500 million** and **$650 million**.
- Parametrix also says most cyber policies include waiting periods of **8 to 12 hours**, meaning that most of the affected companies will receive no compensation.
- **57%** of respondents to Uptime’s 2025 Annual Survey said their most recent major outage cost more than $100,000, and for the second consecutive year, **1 in 5** reported costs exceeding $1 million.

| Estimate | Measured by | Scope | Figure |
|---|---|---|---|
| CrowdStrike outage, direct revenue impact | Delta Air Lines, company filing | One airline, September quarter 2024 | Approximately $380 million |
| CrowdStrike outage, direct financial loss | Parametrix, modeled | US Fortune 500, excluding Microsoft | $5.4 billion |
| AWS us-east-1 event, October 2025 | Parametrix, modeled | US companies | $500 million to $650 million |
| Most recent major outage cost more than $100,000 | Uptime Institute, survey | 2025 Annual Survey respondents | 57% of respondents |
| Most recent major outage cost more than $1 million | Uptime Institute, survey | 2025 Annual Survey respondents | 1 in 5 respondents |

*Source: Delta Air Lines, Parametrix, Uptime Institute, 2024 to 2026*

> **Worth noting:** These loss estimates measure different populations, so they are never added together. Delta reported a direct revenue impact of approximately **$380 million**, while Parametrix estimates a **$5.4 billion** total direct financial loss for US Fortune 500 companies, excluding Microsoft. One is a company’s own filing, and the other is a model.

Parametrix calls us-east-1 one of the largest and most critical cloud regions globally, a regional view of the weight that [cloud provider market share](/cloud-market-share-statistics/) measures by company. The causes behind these costs repeat across providers.

## Cloud Outage Root Causes

- AWS uses an internal network to host foundational services including monitoring, internal DNS, authorization services, and parts of the EC2 control plane.
- The 2017 S3 command was executed by an authorized team member using an established playbook.
- Microsoft’s Front Door review says the customer configuration changes themselves were valid and non-malicious.
- Azure’s July 2026 break-fix repair was initiated on an optical device to address a network reliability risk.
- Google’s August 2026 disruption originated during scheduled fiber optic maintenance, which unexpectedly compromised network capacity between data centers within the us-west1 region.
- Uptime Institute’s 2026 analysis says external infrastructure failures are becoming more prominent in publicly reported outages, and outages linked to fiber and connectivity issues are rising.

| Incident | Trigger in the provider’s report | Report |
|---|---|---|
| AWS S3, February 2017 | One command input entered incorrectly, removing more servers than intended | AWS Post-Event Summary |
| AWS us-east-1, December 2021 | Automated capacity scaling triggered a connection surge on internal network devices | AWS Post-Event Summary |
| CrowdStrike Falcon, July 2024 | Problematic Rapid Response Content configuration update | CrowdStrike Preliminary Post Incident Review |
| Google Cloud, June 2025 | Policy data with unintended blank fields hit a null pointer in Service Control | Google Cloud incident report |
| AWS us-east-1, October 2025 | Latent race condition in DynamoDB DNS management left an empty DNS record | AWS Post-Event Summary |
| Azure Front Door, October 2025 | Incompatible customer configuration metadata exposed a latent data plane bug | Azure Post Incident Review |
| Azure West US, July 2026 | Blast radius analysis defect widened a break-fix repair | Azure Post Incident Review |
| Google Cloud us-west1, August 2026 | Scheduled fiber optic maintenance and failed automated rerouting | Google Cloud incident report |

*Source: Amazon Web Services, Google Cloud, Microsoft Azure, CrowdStrike, 2017 to 2026*

Most triggers in the table are changes, configuration data, or automation rather than failed hardware, power, or cooling in the [data centers](/data-center-statistics/) these services run in. Provider SLAs set what each of these incidents is supposed to cost the provider.

## Cloud SLA Uptime Guarantees

- For Amazon EC2 spread across Availability Zones (AZs), AWS will use commercially reasonable efforts to make Amazon EC2 available for each AWS region with a Monthly Uptime Percentage of at least **99.99%**, and a month less than that makes customers eligible for a Service Credit.
- Against that commitment of at least 99.99%, a month less than 99.99% but equal to or greater than 99.0% earns a **10%** Service Credit, less than 99.0% but equal to or greater than 95.0% earns **30%**, and less than 95.0% earns **100%**.
- For a Single EC2 Instance, AWS will use commercially reasonable efforts to make it available with an Instance-Level Uptime Percentage of at least **99.5%**.
- Google’s Compute Engine SLA for its Premium Tier in Cloud Regions other than Mexico and Stockholm sets a Monthly Uptime Percentage of **99.99%** for Instances in Multiple Zones, **99.95%** for a Single Instance of the Memory Optimized family, and **99.9%** for a Single Instance of all other families.
- At the Region-Level commitment, a 30-day month allows about **4.32 minutes** of downtime.

 DOWNTIME ALLOWED · Source: Amazon Web Services, Google Cloud SLA pages, 2026   DOWNTIME ALLOWED    Amazon Web Services · 2026 Downtime Allowed per 30-Day Month (minutes)         99.99% uptime  4.32 99.95% uptime  21.6 99.9% uptime  43.2 99.5% uptime  216  0 50 100 150 200 250   SOURCE Amazon Web Services, Google Cloud SLA pages, 2026    View the data| Uptime commitment | Downtime allowed (minutes) |
|---|---|
| 99.99% uptime | 4.32 |
| 99.95% uptime | 21.6 |
| 99.9% uptime | 43.2 |
| 99.5% uptime | 216 |





A **99.95%** level allows about **21.6 minutes** a month, **99.9%** about **43.2 minutes,** and the EC2 Instance-Level SLA of at least **99.5%** about **216 minutes**. Each figure applies the levels in the Amazon Compute Service Level Agreement and Google’s Compute Engine SLA to the same month length.

> **By the numbers:** AWS commits to a Monthly Uptime Percentage of at least **99.99%** for EC2 deployed across two or more AZs, with credits when it is less than that, and Google sets **99.99%** for Instances in Multiple Zones on Compute Engine’s Premium Tier outside Mexico and Stockholm. That level leaves about **4.32 minutes** of downtime in a 30-day month, while each cloud provider incident above with published start and end times ran for hours.

SLA uptime is measured per region or instance in each monthly billing cycle, and the credits follow each provider’s own terms. The knock-on reach of an outage runs wider than any one provider’s customers.

## How Many Companies a Cloud Outage Affects

- Parametrix monitors the real-time service availability of over **7,000** SaaS, PaaS and IaaS providers, meaning [cloud computing](/cloud-computing-statistics/) software, platforms and infrastructure sold as a service.
- Roughly **30%** of all providers monitored experienced disruptions of varying levels of severity during the AWS us-east-1 event.
- Roughly **16%** of all providers were experiencing disruptions during the June 12, 2025 GCP outage, so the AWS share was about **1.9 times** as large.
- The first and most severe disruption of the AWS event lasted approximately **2 hours and 36 minutes**, impacting the majority of customers in the region.
- The second critical phase lasted **7.5 hours** and primarily affected new instance creation and autoscaling.
- For **3 hours and 15 minutes**, 3 out of 6 AZs experienced EC2 provisioning issues.
- Uptime Institute says third-party IT and data center service providers, including cloud and internet giants, have accounted for about two-thirds of the publicly reported outages it has tracked over nine years.

 LENGTH · Source: Parametrix, October 2025   LENGTH    Parametrix · 2025 AWS October 2025 Outage Phases (hours)         Phase 1, core service outage  2.6 EC2 provisioning issues in 3 of 6 AZs  3.25 Phase 2, provisioning and autoscaling  7.5  0 1.6 3.2 4.8 6.4 8   SOURCE Parametrix, October 2025    View the data| Phase (Parametrix monitoring) | Length (hours) |
|---|---|
| Phase 1, core service outage | 2.6 |
| EC2 provisioning issues in 3 of 6 AZs | 3.25 |
| Phase 2, provisioning and autoscaling | 7.5 |





Phase 1 works out to about **2.6 hours** and the AZ provisioning window to **3.25 hours**, both shorter than the second phase.

> **The takeaway:** Parametrix saw roughly **30%** of the over **7,000** providers it monitors disrupted during the AWS us-east-1 event, against roughly **16%** during the June 2025 GCP outage. The base is the providers Parametrix monitors, not every company that runs on a cloud.

## What is the reason for the AWS outage?

AWS says the root cause was a latent race condition in the DynamoDB DNS management system that resulted in an incorrect empty DNS record for the service’s regional endpoint, which the automation failed to repair. Its summary also lists failed EC2 instance launches between 2:25 a.m. and 10:36 a.m. on October 20, and NLB connection errors caused by health check failures in the NLB fleet; the failed launches alone lasted **8.2 hours**.

## How long did the AWS outage last?

AWS dates the event from 11:48 p.m. PDT on October 19 to 2:20 p.m. PDT on October 20, 2025, about **14.5 hours** in total. Parametrix says the first and most severe disruption lasted approximately **2 hours and 36 minutes**, and a second critical phase lasted **7.5 hours**.

## Conclusion

AWS’s October us-east-1 event ran from 11:48 p.m. PDT to 2:20 p.m. PDT the next day, about **14.5 hours**, and **11** of the **18** summaries in AWS’s public index concern the same us-east-1 region. Parametrix estimates US company losses from that event at between **$500 million** and **$650 million**, a modeled figure that sits beside Delta’s filed CrowdStrike impact rather than on the same scale.

Uptime Institute’s latest outage analysis says outages linked to fiber and connectivity issues are rising and more likely to result in extended disruptions, the same kind of trigger behind Google’s us-west1 incident. For IT buyers and engineering leads, each provider’s own post-incident report remains the record for an outage’s cause and length.