Cloud Provider Outage Statistics: How Often Do AWS, Azure & Cloudflare Go Down?

Cloud Infrastructure Research — PaidHosting.com

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.

📊 2020–2026 incident history ☁️ AWS · Azure · Cloudflare 🔄 Updated September 2026

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.

2 of 3publicly reported outages over the last nine years now trace back to a third-party provider — cloud, telecom, or colocationUptime Institute · 2026
~3 hrsduration of Cloudflare’s November 18, 2025 global outage — short, but it took down access to X, OpenAI and AnthropicThousandEyes · Nov 2025
13separate incidents Cloudflare logged in just eight days in August 2026Incident tracking · Aug 2026
8.5MWindows devices crashed worldwide on July 19, 2024, when a third-party security update compounded an active Azure outageMicrosoft · July 2024
20%of outages cost more than $1 million — the second year running this has been trueUptime Institute · 2025 survey
5th yearrunning that outage frequency per site has declined industry-wide — though the pace of improvement is slowingUptime Institute · 2026

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.

Nov 2020 — Kinesis cascade~10 hrs
Sep 2021 — EBS stuck I/O~8 hrs
Dec 2021 — API Gateway errors~7 hrs
Jul 2022 — AZ power event~3 hrs
Jul 2024 — Kinesis architecture flaw~7 hrs
Feb 2025 — eu-north-1 networking~4.5 hrs
Oct 2025 — DynamoDB DNS race condition15+ hrs

Sources: StatusGator AWS outage history; ThousandEyes AWS Outage Analysis, October 20, 2025.Free to reuse with attribution and a link to paidhosting.com

⚠️
The October 20, 2025 outage, explained

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.

DateIncidentDurationRoot Cause
Oct 20, 2025us-east-1 DynamoDB / EC215+ hrsDNS management race condition
Feb 13–14, 2025eu-north-1 networking~4.5 hrsInternal networking disruption
Jul 30, 2024us-east-1 Kinesis~7 hrsNew Kinesis architecture flaw
Jun 13, 2023us-east-1 (Connect, Lambda)Several hoursNot disclosed
Jul 28, 2022us-east-2 power event~3 hrsPower outage in one AZ
Dec 7–10, 2021us-east-1 (multi-phase)~7 hrs + ~1 hrAPI Gateway error rates/latency
Sep 26, 2021us-east-1 EBS~8 hrsStuck I/O in EBS
Nov 25, 2020us-east-1 Kinesis cascade~10 hrsKinesis 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.

Jan 2020 — Global SQL dependency~6 hrs
Jun 2022 — AD sign-in/telemetry~9.5 hrs
Jun 2022 — Firewall/Data Explorer~24 hrs
Jan 2023 — West Europe storage~5.5 hrs
Jan 2023 — Global networking~3.5 hrs
Nov 2024 — DNS resolution failure~8.5 hrs
Apr 2025 — AI services disruption~9 hrs

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

🌀
July 19, 2024: when a security update took down the internet, not Azure itself

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.

DateIncidentDurationRoot Cause
Apr 9, 2025AI services disruption~9 hrsNot officially disclosed
Jan 9–11, 2025East US 2 networkingMulti-dayNetwork infrastructure issue
Dec 26, 2024South Central US power~2 hrs+ (partial recovery into next day)Power incident in one AZ
Nov 13, 2024DNS resolution failure (10+ regions)~8.5 hrsDNS resolution failures
Jul 19, 2024CrowdStrike-compounded VM crashesDays (residual)Third-party security update (not Azure-caused)
Jan 25, 2023Global networking (25+ regions)~3.5 hrsNetwork infrastructure degradation
Jan 18, 2023West Europe storage~5.5 hrsPower event in a storage scale unit
Jun 29–30, 2022Firewall / Data Explorer (global)~24 hrsNot fully disclosed
Jun 1, 2022Azure AD sign-in logs~9.5 hrsTelemetry 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.

Apr 2020 — Dashboard/API/Analytics24+ hrs
Mar 2021 — South America instability~8 hrs
Apr 2022 — Zaraz / Spectrum~3.25 hrs
Mar 2024 — China performance~2 hrs
Nov 2025 — Global bot-management crash~3 hrs

Sources: StatusGator Cloudflare outage history; ThousandEyes Cloudflare Outage Analysis, November 18, 2025.Free to reuse with attribution and a link to paidhosting.com

📌
Duration isn’t the same as impact

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.

🔁
August 2026: 13 incidents in 8 days

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.

DateIncidentDurationRoot Cause
Aug 7–14, 202613-incident cluster (R2, email, regional)8-day spanMultiple unrelated causes
Nov 18, 2025Global bot-management crash~3 hrsOversized config file crashed proxy
Apr 20–24, 2025Multi-day cascading issuesSeveral days (intermittent)Not fully disclosed
Mar 18, 2024China performance degradation~2 hrsUnconfirmed (billing system upgrade nearby)
Apr 28, 2022Zaraz / Spectrum connectivity~3.25 hrsNot disclosed
Mar 29, 2021South America (Brazil, ANZ)~8 hrs (rolling)Regional network instability
Apr 14–15, 2020Dashboard, API, Analytics, Stream24+ 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.

  • Outage frequency per site has declined for five consecutive years industry-wide, though the pace of improvement has slowed compared to previous years.Source: Uptime Institute, Annual Outage Analysis 2026
  • Third-party providers — cloud, telecommunications, and colocation companies — now account for roughly two-thirds of all publicly reported outages tracked over the last nine years, reflecting how much of the internet’s infrastructure has consolidated onto a small number of platforms.Source: Uptime Institute, Annual Outage Analysis 2026
  • Power infrastructure — UPS systems, transfer switches, and generators — remains the single leading technical cause of data center outages; human error tied to procedural non-compliance is the leading human factor.Source: Uptime Institute, Annual Outage Analysis 2026
  • About 1 in 10 organizations describe their most recent outage as having a “serious” or “severe” impact — but 57% of reported outages in the 2025 survey cost more than $100,000, and 20% cost more than $1 million, for the second year running.Source: Uptime Institute, 2025 Outage Survey
  • 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

    Paid Hosting. Cloud Provider Outage Statistics: How Often Do AWS, Azure & Cloudflare Go Down? September 2026. Compiled from public status pages, provider post-incident reviews, and independent network-monitoring analyses, 2020–2026.
    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]