Cloud Provider Outage Statistics: How Often Do AWS, Azure & Cloudflare Go Down?
A compiled history of major AWS, Microsoft Azure, and Cloudflare outages from 2020 through 2026 — built from public status pages, official post-incident reviews, and independent network-monitoring analyses. Every incident below links to its primary source.
📋 What’s in this page
Key Outage Statistics at a Glance
Every figure below is tied to a named incident report or industry study in the sources list. Durations are as reported by the provider or the monitoring firm cited — see the methodology note for why different trackers sometimes report slightly different numbers for the same incident.
AWS Outage History
AWS’s most consequential outages have repeatedly centered on us-east-1 (Northern Virginia), its oldest and most heavily used region. Below are the major publicly documented incidents from 2020 through 2025.
Duration of major AWS outages, 2020–2025
Hours from incident start to full service recovery, as reported in each incident’s post-incident review or independent analysis.
Sources: StatusGator AWS outage history; ThousandEyes AWS Outage Analysis, October 20, 2025.Free to reuse with attribution and a link to paidhosting.com
A latent race condition in DynamoDB’s automated DNS management system let two internal “DNS Enactor” processes overwrite each other’s updates. When cleanup routines deleted what looked like a stale record, the regional DynamoDB endpoint’s DNS entry went empty — and the system couldn’t self-heal. EC2, STS, Redshift and network load balancers cascaded into failure behind it. Slack, Snapchat, and Atlassian were among the highest-profile services knocked offline.
| Date | Incident | Duration | Root Cause |
|---|---|---|---|
| Oct 20, 2025 | us-east-1 DynamoDB / EC2 | 15+ hrs | DNS management race condition |
| Feb 13–14, 2025 | eu-north-1 networking | ~4.5 hrs | Internal networking disruption |
| Jul 30, 2024 | us-east-1 Kinesis | ~7 hrs | New Kinesis architecture flaw |
| Jun 13, 2023 | us-east-1 (Connect, Lambda) | Several hours | Not disclosed |
| Jul 28, 2022 | us-east-2 power event | ~3 hrs | Power outage in one AZ |
| Dec 7–10, 2021 | us-east-1 (multi-phase) | ~7 hrs + ~1 hr | API Gateway error rates/latency |
| Sep 26, 2021 | us-east-1 EBS | ~8 hrs | Stuck I/O in EBS |
| Nov 25, 2020 | us-east-1 Kinesis cascade | ~10 hrs | Kinesis API error rates/latency |
Durations reflect time to full recovery where reported; some incidents had partial service restored earlier.
Azure Outage History
Azure’s worst incidents have been more evenly spread across regions than AWS’s, with networking, DNS, and power events all recurring causes. One of the most disruptive “Azure” days in its history wasn’t even Azure’s fault — it compounded with an unrelated third-party vendor failure.
Duration of major Azure outages, 2020–2025
Hours from incident start to full service recovery, as reported by Microsoft or independent monitoring.
Source: StatusGator Azure outage history, compiled from Microsoft’s Azure status history and incident post-mortems.Free to reuse with attribution and a link to paidhosting.com
A faulty configuration update pushed by security vendor CrowdStrike to its Falcon Sensor caused an out-of-bounds memory read on Windows systems, forcing roughly 8.5 million Windows devices worldwide into boot loops. It wasn’t an Azure infrastructure failure — but an unrelated Azure outage the day before had already strained central US regions, and once the CrowdStrike update landed, Azure virtual machines began rebooting and crashing alongside every other affected Windows system. Total global damage has been estimated in the tens of billions of dollars; Delta Air Lines alone reported roughly $500 million in costs. It remains the most disruptive single IT event of the decade, even though the root cause sat outside any cloud provider’s own infrastructure.
| Date | Incident | Duration | Root Cause |
|---|---|---|---|
| Apr 9, 2025 | AI services disruption | ~9 hrs | Not officially disclosed |
| Jan 9–11, 2025 | East US 2 networking | Multi-day | Network infrastructure issue |
| Dec 26, 2024 | South Central US power | ~2 hrs+ (partial recovery into next day) | Power incident in one AZ |
| Nov 13, 2024 | DNS resolution failure (10+ regions) | ~8.5 hrs | DNS resolution failures |
| Jul 19, 2024 | CrowdStrike-compounded VM crashes | Days (residual) | Third-party security update (not Azure-caused) |
| Jan 25, 2023 | Global networking (25+ regions) | ~3.5 hrs | Network infrastructure degradation |
| Jan 18, 2023 | West Europe storage | ~5.5 hrs | Power event in a storage scale unit |
| Jun 29–30, 2022 | Firewall / Data Explorer (global) | ~24 hrs | Not fully disclosed |
| Jun 1, 2022 | Azure AD sign-in logs | ~9.5 hrs | Telemetry infrastructure failure |
Cloudflare Outage History
Because Cloudflare sits in front of such a large share of the web, even short outages tend to be highly visible. Its most talked-about recent incident lasted only a few hours — but reached an unusually large slice of the internet at once.
Duration of major Cloudflare outages, 2020–2025
Hours from incident start to full service recovery, as reported by Cloudflare or independent monitoring.
Sources: StatusGator Cloudflare outage history; ThousandEyes Cloudflare Outage Analysis, November 18, 2025.Free to reuse with attribution and a link to paidhosting.com
Cloudflare’s November 18, 2025 outage lasted roughly three hours — shorter than several AWS and Azure incidents above — but because Cloudflare sits directly in front of a huge share of global web traffic, it knocked out access to X, OpenAI, and Anthropic’s services simultaneously. The cause was almost comically small in scale: a database permissions change caused an internal “feature file” used by Cloudflare’s bot-management system to balloon from around 60 rows to over 200, which exceeded a hard-coded size limit and crashed the proxy software each time it tried to reload the file.
Between August 7 and August 14, 2026, Cloudflare logged 13 separate incidents — spanning R2 object storage write failures, an email-delivery disruption caused by a Spamhaus blocklist entry, regional connectivity problems in Istanbul and London, and a cluster of minor regional errors across the Middle East, Southeast Asia, the US, and Mexico. None matched the scale of the November 2025 global outage individually, but the clustering itself is notable — and a good example of why outage frequency, not just headline duration, matters when evaluating how dependent your site should be on any single edge network.
| Date | Incident | Duration | Root Cause |
|---|---|---|---|
| Aug 7–14, 2026 | 13-incident cluster (R2, email, regional) | 8-day span | Multiple unrelated causes |
| Nov 18, 2025 | Global bot-management crash | ~3 hrs | Oversized config file crashed proxy |
| Apr 20–24, 2025 | Multi-day cascading issues | Several days (intermittent) | Not fully disclosed |
| Mar 18, 2024 | China performance degradation | ~2 hrs | Unconfirmed (billing system upgrade nearby) |
| Apr 28, 2022 | Zaraz / Spectrum connectivity | ~3.25 hrs | Not disclosed |
| Mar 29, 2021 | South America (Brazil, ANZ) | ~8 hrs (rolling) | Regional network instability |
| Apr 14–15, 2020 | Dashboard, API, Analytics, Stream | 24+ hrs (multiple waves) | Not disclosed |
How the Three Providers Compare
No provider is immune, and each tends to fail in a slightly different way. Here’s the pattern that emerges across the incidents documented above.
AWS
- Failures concentrate heavily in us-east-1, its oldest and busiest region
- Most common root cause: internal service dependencies (DynamoDB, Kinesis) cascading into EC2 and dependent services
- Longest documented outage in this dataset: 15+ hours (October 2025)
Azure
- Incidents spread more evenly across regions and services than AWS’s
- Networking, DNS, and power events recur most often
- Longest documented outage in this dataset: ~24 hours (June 2022, Firewall/Data Explorer)
Cloudflare
- Outages tend to run shorter than AWS or Azure’s biggest incidents, but reach a disproportionately large share of the web at once because of Cloudflare’s position in front of so much traffic
- Small configuration or software bugs (an oversized file, a permissions change) have caused some of its most visible failures
- Recent incident frequency has been rising — 13 separate incidents in just 8 days in August 2026
Why Cloud Outages Keep Happening
Cloud infrastructure has gotten more reliable over time, not less — but the industry-wide data shows the improvement is slowing, and the nature of failures is shifting.
What This Means If Your Business Depends on One Provider
None of this is a reason to avoid AWS, Azure, or Cloudflare — all three remain far more reliable than running equivalent infrastructure yourself. But the pattern above is consistent: even the largest providers have multi-hour, sometimes multi-day, failures that are completely outside a customer’s control, and a growing share of the internet’s outages now originate at exactly this layer rather than at any individual site or host.
The practical takeaway is about blast radius, not brand loyalty: knowing which of your own services depend on a single region, a single CDN, or a single DNS provider — and having a plan for what happens when that dependency goes down — matters more than picking the “most reliable” provider, since on any given year, that title can change hands.
Sources & Methodology
Incident dates, durations, and root causes on this page were compiled from public status pages, providers’ own post-incident reviews, and independent network-monitoring analyses. Where a duration is reported as approximate, it reflects the time between the first public status-page update and confirmation of full recovery — different trackers sometimes measure this window slightly differently (detection time vs. full resolution vs. residual/downstream effects), which is why you may see a marginally different number for the same incident elsewhere. This page is updated as new major incidents occur and older figures are re-verified periodically.
- StatusGator — AWS Outage History, Azure Outage History, and Cloudflare Outage History (compiled incident timelines)
- ThousandEyes — AWS Outage Analysis, October 20, 2025
- ThousandEyes — Cloudflare Outage Analysis, November 18, 2025
- Wikipedia — 2024 CrowdStrike-related IT Outages
- Uptime Institute — Annual Outage Analysis 2026
- Uptime Institute — 2025 Outage Survey (cost and severity data)
- shattered.io — Cloudflare incident tracking, August 2026
- Provider status pages and official post-incident review pages (AWS, Microsoft Azure, Cloudflare)
Cite this analysis
https://www.paidhosting.com/cloud-outage-statistics/
Charts and figures on this page may be reproduced with attribution and a link back to this page. Press and data enquiries: [email protected]