Every big outage sets off the same three questions: what does downtime cost, how often does it happen, and how long does it take to fix? This page collects more than 100 sourced answers for journalists, engineers, and anyone making the case for reliability work.

The figures come from original surveys, monitoring datasets, official postmortems, and research reports. Each one links to the organization that published it and is dated by its original release. Hyperping, which publishes this page, sells uptime monitoring, status pages, and on-call scheduling.

Key takeaways

  • Unplanned downtime cost Global 2000 companies an average of $15,000 per minute in a survey of 2,000 executives at those companies. Splunk sells observability software and Oxford Economics ran the survey, so treat the figure as a large-enterprise benchmark rather than a typical one. (Splunk and Oxford Economics, May 2026)
  • 47% of technology leaders at Global 2000 companies said customers are often or very often the first to detect service degradation or outages. Splunk does not publish how many technology leaders answered, so pair this with New Relic's similar finding when you cite it. (Splunk and Oxford Economics, May 2026)
  • High-impact outages took an average of 41 minutes to detect, in New Relic's survey of 2,575 IT and engineering leaders. The report gives a median of 35 minutes, so a few slow detections pull the average up. (New Relic, September 2026)
  • Third-party IT and data center service providers, including cloud and internet giants, telecoms, and colocation companies, accounted for about two-thirds of the publicly reported outages Uptime Institute tracked over 9 years. Public outages skew toward large providers that make the news, so the share overstates third parties' role in the outages a typical company has. (Uptime Institute, May 2025)
  • 50% of data center operators reported an impactful or serious outage in the past 3 years, down from 74% in 2020. Uptime Institute's series has fallen every year since 2020, the strongest evidence that outages are getting rarer at the facility level. (Uptime Institute, May 2026)
  • Downdetector recorded 17M+ user reports during the October 20, 2025 AWS outage, a 970% increase on its average daily baseline, with disruptions at over 3,500 companies in more than 60 countries. Downdetector counts user reports, not unique users, so the figure shows how visible the outage was rather than how many people it hit. (Ookla, October 2025)
  • Authorities shut down the Internet 313 times across 52 countries in 2025, the most the #KeepItOn coalition has documented. Access Now counts every deliberate shutdown, including regional and platform blocks, which is why its total is higher than Top10VPN's. (Access Now, March 2026)

Cost of downtime statistics

Cost figures are the most quoted numbers in reliability writing and the most often misused. The studies below measure different things (per-minute averages, hourly ranges, annual totals), so each one says who was asked and what the number includes.

Cost per minute and per hour

Unplanned downtime cost Global 2000 companies an average of $15,000 per minute in a survey of 2,000 executives at those companies. Splunk sells observability software and Oxford Economics ran the survey, so treat the figure as a large-enterprise benchmark rather than a typical one. (Splunk and Oxford Economics, May 2026)

A single hour of downtime cost more than $300,000 for over 90% of mid-size and large enterprises in ITIC's survey of more than 1,000 firms. ITIC's estimates are self-reported and exclude litigation and civil or criminal penalties, so the full cost for affected firms is likely higher. (ITIC, September 2024)

High-impact IT outages had a mean cost of $1.85 million per hour, or $30,833 per minute, among 2,575 IT and engineering leaders surveyed. New Relic reported a median the year before, so its two editions should not be read as a trend. (New Relic, September 2026)

Unplanned IT downtime cost an average of $14,056 per minute across 415 IT professionals at companies with 1,000 or more employees. The survey was commissioned by BigPanda, which sells AIOps tools, and covers only companies of that size. (EMA for BigPanda, April 2024)

41% of enterprises said 1 hour of downtime costs them $1 million to over $5 million. This comes from the same ITIC survey as a range of answers, so cite it as a share of firms, not an average cost. (ITIC, September 2024)

IT leaders estimated the cost of downtime at $4,537 per minute, with the average incident taking 175 minutes to resolve, in a survey of 500 IT leaders at companies with 1,000+ employees. PagerDuty, which sells incident response tools, ran the survey with Censuswide in the US, UK, and Australia. (PagerDuty, June 2024)

Cost by company size and industry

IT outages cost companies with more than 10,000 employees an average of $23,750 per minute in EMA's 2024 survey. Per-minute costs climb with company size in EMA's data, so use the band that matches the business you are writing about. (EMA for BigPanda, April 2024)

37% of small and medium-sized businesses said a single hour of downtime costs between $1,000 and $5,000, in a survey of 715 SMB organizations. Calyptix, a firewall vendor for small businesses and MSPs, drew part of the sample from its own partners, so the result leans toward managed small businesses. (Calyptix and ITIC, February 2025)

An hour of downtime in a large automotive plant cost $2.3 million, more than $600 a second. Siemens, which sells predictive maintenance, built this from interviews at industrial firms, so it describes factory lines, not software services. (Siemens, 2024)

Unplanned downtime cost manufacturers an average of $1.7M per hour in a survey of 600 manufacturing decision-makers and maintenance professionals in the US, UK, and Germany. Fluke sells condition monitoring tools, and the figure covers production downtime in heavy industry rather than IT. (Fluke, October 2025)

Annual losses and what they include

The aggregate cost of unplanned downtime for Global 2000 companies reached $600 billion a year, a 50% increase in two years. The total is modeled from survey answers across the Global 2000, so it shows scale and direction, not an audited loss. (Splunk and Oxford Economics, May 2026)

Businesses lost an annualized $74 million to high-impact outages, among 2,575 IT and engineering leaders surveyed in 24 countries. The report PDF gives $73.5 million and the press release rounds it, so quote the version you link to. (New Relic, September 2026)

Siemens estimated that the world's 500 biggest companies lose almost $1.4 trillion a year to unplanned downtime, equivalent to 11% of their revenues. The total is extrapolated from 181 interviews at large industrial firms, so it is a rough upper bound rather than a measured loss. (Siemens, 2024)

57% of data center operators said their most recent major outage cost more than $100,000, in Uptime Institute's 2025 annual survey. The base is the operators who had a major outage, a group of about 94 respondents, so the share is less precise than it looks. (Uptime Institute, May 2026)

All 1,000 surveyed senior technology executives reported outage-related revenue loss in the past year, and 84% lost at least $10,000. Cockroach Labs sells a distributed database marketed on resilience, and Wakefield Research ran the survey. (Cockroach Labs, October 2024)

1 in 5 data center operators reported that their most recent major outage cost more than $1 million, for the second year in a row. Uptime Institute counts direct, opportunity, and reputation costs, which makes the figure broader than lost revenue alone. (Uptime Institute, May 2026)

Regulatory fines tied to downtime averaged $51 million per Global 2000 organization, and 57% of technology executives called these penalties very or prohibitively disruptive. The release does not say how many respondents paid fines, so treat the average as a sign of exposure, not a typical bill. (Splunk and Oxford Economics, May 2026)

51% of organizations said internet outages and disruptions cost them more than $1M in the last 30 days, in a survey of 475 managers and executives. Catchpoint sells internet performance monitoring, and two-thirds of respondents worked at technology and as-a-service providers, so the sample skews toward businesses that run online services. (Catchpoint, June 2025)

Outage frequency and duration statistics

Surveys, facility data, and monitoring tools all count outages differently. Surveys capture how often leaders feel hit, while monitoring datasets count every confirmed failure on the endpoints they watch, so expect their numbers to disagree.

How often organizations have outages

Organizations averaged 86 outages a year, according to 1,000 senior technology executives surveyed by Cockroach Labs. Respondents count outages in their own way, so this measures how often leaders feel hit, not a logged incident rate. (Cockroach Labs, October 2024)

36% of IT and engineering leaders said their companies have a high-impact outage weekly or more, in a survey of 2,575 respondents. A weekly high-impact outage at a third of companies is the number to quote when arguing that incident response needs a standing rotation. (New Relic, September 2026)

50% of data center operators reported an impactful or serious outage in the past 3 years, down from 74% in 2020. Uptime Institute's series has fallen every year since 2020, the strongest evidence that outages are getting rarer at the facility level. (Uptime Institute, May 2026)

55% of 1,000 senior technology executives said their organizations experience outages or disruptions at least once a week. Cockroach Labs' report PDF gives a different weekly share than its blog post, so cite this figure with the blog link. (Cockroach Labs, October 2024)

26% of 415 IT professionals said significant outages happen yearly or almost never, up from 19% in 2022. EMA ran the survey for BigPanda, and the rise suggests serious outages became rarer between its two editions. (EMA for BigPanda, April 2024)

Cisco ThousandEyes observed 1,142 global network outages in February 2026, down 2% from January. ThousandEyes counts outages across ISPs, cloud providers, and collaboration apps from its own vantage points, so the count works best for month-to-month trends. (Cisco ThousandEyes, April 2026)

The total duration of critical cloud downtime events fell about 28% to 175.3 hours in 2025, from a peak of 244.8 hours in 2024. Parametrix sells downtime insurance and tracks only complete or significant disruptions at major cloud providers. (Parametrix, March 2026)

Outages seen by monitoring tools

54.1% of endpoints monitored by Pulsetic had at least one confirmed outage in the 12 months to November 1, 2026. Pulsetic's dataset covers its own customers' endpoints, so it reflects sites that already pay for monitoring. (Pulsetic, November 2026)

61.4% of sites monitored by Oh Dear went down at least once between March and December 2025. Oh Dear confirms each failure from a second location, which filters out most false alarms. (Oh Dear, 2025)

90.7% of outages confirmed by Pulsetic in the 12 months to November 1, 2026 were resolved within an hour. Most outages are short, which is why the check interval decides whether you see them at all. (Pulsetic, November 2026)

The average site monitored by Oh Dear had 47 outages in 10 months. Oh Dear also found that most of those outages started outside business hours, when nobody is watching a dashboard. (Oh Dear, 2025)

Availability targets and recovery trends

90% of organizations require a minimum of 99.99% availability, in ITIC's most recent reliability poll. Four nines leaves a budget of about 52 minutes of downtime a year, which a single cloud outage can use up. (ITIC, September 2024)

82% of monitors in Pulsetic's network reached 99.9% uptime or better over 12 months, while the mean endpoint was up 98.90%. The gap between the two numbers comes from a tail of endpoints with long outages that drag the mean down. (Pulsetic, November 2026)

The median time to resolve an unplanned incident on public status pages was 3.9 hours in 2025, across 17,214 incidents at a fixed group of 234 companies. StackGen measures from the timestamps companies publish, so it tracks how long incidents stay open in public, not internal repair time. (StackGen, August 2026)

Outage cause statistics

No single cause dominates every dataset. Facility surveys point to power and human error, cloud trackers point to changes and configuration, and network monitors point to ISPs and cloud providers, so the right cause figure depends on which layer you are writing about.

Third parties and cloud providers

Third-party IT and data center service providers, including cloud and internet giants, telecoms, and colocation companies, accounted for about two-thirds of the publicly reported outages Uptime Institute tracked over 9 years. Public outages skew toward large providers that make the news, so the share overstates third parties' role in the outages a typical company has. (Uptime Institute, May 2025)

Cloud service providers caused 27% of the ISP and cloud outages Cisco ThousandEyes observed in 2024, up from 17% in 2023 and 11% in 2022. The ratio comes from ThousandEyes' own monitoring, and ISPs still caused most of the outages it saw. (Cisco ThousandEyes, January 2025)

56% of security leaders at Global 2000 companies said they experience cybersecurity-related downtime caused by SaaS and third-party application issues often or very often, nearly triple the 2024 level. The release does not say how many security leaders answered, so cite it as a direction of travel rather than a precise share. (Splunk and Oxford Economics, May 2026)

Third-party cascades, where an upstream provider's failure takes a service down, made up 26% of assessed status-page incidents in 2026 through June, after 35% in 2025. StackGen calls this a floor, since many status-page posts never name the upstream cause. (StackGen, August 2026)

Service provider outages such as ISP disruptions accounted for 19% of disruptions reported by 3,051 IT professionals surveyed by Unitrends. Unitrends sells backup and recovery, and its sample is mostly midsized businesses in the Americas. (Unitrends, 2025)

Human error, changes, and configuration

92% of data center operators said human error was at least a minor contributor to their most recent impactful outage. Uptime Institute counts human error as a contributing factor, so this share sits alongside other causes rather than competing with them. (Uptime Institute, May 2026)

Of 62 major AWS, Microsoft Azure, Google Cloud, and Cloudflare outages documented since 2020, 34 (55%) started with a configuration or deployment change, a software bug, or automation acting on bad input. Control Plane sells a multi-cloud platform and picked the outages itself, so the sample is small and hand-selected. (Control Plane, October 2026)

26% of 2025 DORA survey respondents said 8% to 16% of their changes to production degrade service, the most common answer among 4,867 technology professionals. Change failure rate is self-reported in DORA's survey, which makes it a peer benchmark rather than a measured rate. (Google DORA, September 2025)

87% of data center operators who had an impactful outage in the past 3 years said better management, processes, or configuration could have prevented it, up 7 percentage points from 2024. The base is 98 operators who had such an outage, so read it as a strong majority view, not a precise share. (Uptime Institute, July 2025)

79% of production issues in Komodor's Kubernetes customer data started with a recent system change. Komodor sells Kubernetes troubleshooting tools, and the data covers only its customers' clusters. (Komodor, September 2025)

4 out of 5 Kubernetes issues were related to configuration or change management, in Red Hat's survey of 785 Kubernetes decision-makers and practitioners. Read with Komodor's telemetry, it is strong evidence that change control matters more than hardware on container platforms. (Red Hat, June 2026)

Power, networks, security, and certificates

Power was the top cause of impactful data center outages in 2025, accounting for 45% of incidents. Power failures hit facilities rather than code, so they matter most to teams that run their own data centers or colocation. (Uptime Institute, July 2025)

Security incidents such as ransomware and viruses were the leading cause of outages for 22% of organizations in 451 Research's 2026 study. The public post does not give a sample size, so cite the ranking with care. (451 Research, August 2026)

86% of companies had at least one outage caused by expired or mismanaged digital certificates in the past year, and 31% had one at least once a quarter. Keyfactor sells certificate management and surveyed 450 PKI practitioners at large companies, a group closer to the problem than most. (Keyfactor, September 2025)

Network failures were the leading cause of outages for 21% of organizations in 451 Research's 2026 study. Network and security causes ran within a point of each other, so neither should be dismissed when planning redundancy. (451 Research, August 2026)

More than one-third of organizations had a service outage caused by an expired certificate in the past year, in DigiCert's survey of 1,001 IT and cybersecurity decision makers. DigiCert, a certificate authority, surveyed a broader group of IT roles than Keyfactor, which may explain the lower share. (DigiCert, September 2026)

Detection and recovery statistics

How fast a team finds out about an outage decides how long customers sit with it. These figures cover who notices first, how long detection and repair take, and what customers hear while it happens.

Who finds outages first

47% of technology leaders at Global 2000 companies said customers are often or very often the first to detect service degradation or outages. Splunk does not publish how many technology leaders answered, so pair this with New Relic's similar finding when you cite it. (Splunk and Oxford Economics, May 2026)

42% of organizations still learn about disruptions through manual checks and customer complaints, in New Relic's survey of 2,575 IT and engineering leaders. Learning from complaints means the outage has already reached customers before anyone starts working on it. (New Relic, September 2026)

78% of organizations had at least one incident where no alert fired at all, in NeuBird's survey of more than 1,000 SRE, DevOps, and IT professionals. NeuBird sells AI incident tools, and the question counts any single silent incident, so it measures coverage gaps rather than how often alerts fail. (NeuBird, April 2026)

Failures detected by UptimeRobot's external monitoring dropped just 7.1% on weekends in the first half of 2026, while incidents providers published on their status pages dropped 62.4%. UptimeRobot notes the gap does not prove providers hide incidents, but it suggests many weekend outages go unannounced. (UptimeRobot, October 2026)

73% of 1,855 ITOps and engineering professionals reported outages caused by ignored or suppressed alerts. Splunk sells observability tools, and a 2026 NeuBird survey found a much lower share for a similar question, so cite this as an upper estimate. (Splunk, October 2025)

30% of 1,363 observability practitioners named alert fatigue as the biggest obstacle to faster incident response, nearly double the next answer. Grafana Labs' sample comes from its own community and events, so it leans toward engineers who already run monitoring. (Grafana Labs, March 2026)

Time to detect and resolve

High-impact outages took an average of 41 minutes to detect, in New Relic's survey of 2,575 IT and engineering leaders. The report gives a median of 35 minutes, so a few slow detections pull the average up. (New Relic, September 2026)

Across 350,000+ incidents, the median team stopped the impact of a SEV0 or SEV1 incident in 88 minutes, and the best quarter of teams in 44. Rootly measures from when an incident is declared to when impact stops, so the time spent noticing the problem is not included. (Rootly, October 2026)

High-impact outages took an average of 54 minutes to resolve, in New Relic's survey of 2,575 IT and engineering leaders. The report's median is 51 minutes, close to the mean, so resolution times are less skewed than detection times. (New Relic, September 2026)

The average outage lasted 196 minutes, according to 1,000 senior technology executives surveyed by Cockroach Labs. The report PDF gives a shorter average recovery time of 151 minutes, so outage length depends on where respondents start and stop the clock. (Cockroach Labs, October 2024)

21.3% of 2025 DORA survey respondents said they generally restore service within an hour after a change to production causes degraded service. DORA's answers are self-reported by 4,867 technology professionals, which makes them a peer benchmark for teams setting recovery goals. (Google DORA, September 2025)

65% of 415 IT professionals said a significant outage usually lasts between 30 minutes and 2 hours. EMA asked about a typical significant outage, so the long tail of multi-day incidents falls outside this range. (EMA for BigPanda, April 2024)

56.5% of 2025 DORA survey respondents said they recover from a failed deployment in less than a day. That leaves more than 4 in 10 teams needing a day or longer to recover from a bad change. (Google DORA, September 2025)

55% of the 122 publicly reported outages Uptime Institute tracked in 2025 were resolved within 12 hours, and the share lasting more than 48 hours rose for the second year in a row. Uptime Institute calls its public-outage data less reliable than its surveys, so use it for trends, not precise shares. (Uptime Institute, May 2026)

Communicating with customers during outages

When a SEV0 or SEV1 incident got a status-page update, the median team posted its first one 9 minutes after declaring the incident. Rootly counts only incidents that got a status-page update, so incidents that were never posted publicly are left out. (Rootly, October 2026)

30% of 1,000 US adults said they got no useful information during their last service disruption. Xurrent sells service management software, and Dynata fielded the survey with adults aged 25 and over. (Xurrent, August 2026)

Across 2,544 incidents on 77 software providers' status pages, the typical gap between updates was 46 minutes, but at the 90th percentile customers waited 9 hours 52 minutes for the next update. NoDisrupt sells status page monitoring and read the public incident histories directly, so the gaps are what customers actually saw. (NoDisrupt, 2026)

34% of 1,000 US adults said they had to go hunting for updates themselves during their last service disruption. Customers who have to search for updates end up in support queues and on social media, which is where an outage turns into a reputation problem. (Xurrent, August 2026)

Major cloud and platform outage statistics

A handful of incidents at AWS, CrowdStrike, Cloudflare, Google Cloud, and Microsoft account for most of the outage coverage since 2024. The figures below come from the providers' own postmortems where possible, and from independent monitors and insurers for reach and cost.

AWS us-east-1, October 2025

AWS's us-east-1 event ran from 11:48 PM PDT on October 19 to 2:20 PM PDT on October 20, 2025, with 3 distinct periods of impact to customer applications. The postmortem uses Pacific time while much of the coverage uses UTC, which is why reported start times differ. (AWS, October 2025)

Downdetector recorded 17M+ user reports during the October 20, 2025 AWS outage, a 970% increase on its average daily baseline, with disruptions at over 3,500 companies in more than 60 countries. Downdetector counts user reports, not unique users, so the figure shows how visible the outage was rather than how many people it hit. (Ookla, October 2025)

Parametrix estimated that the October 2025 AWS us-east-1 outage cost US companies between $500 million and $650 million. Parametrix sells outage insurance and built the estimate from its own monitoring of 7,000+ cloud providers. (Parametrix, October 2025)

PagerDuty created 2,314,163 incidents in the 12 hours of peak impact from the October 20, 2025 AWS outage. PagerDuty gives no baseline for a normal 12 hours, so the count shows the load on incident tooling, not how much worse the day was. (PagerDuty, October 2025)

incident.io saw a 4x traffic increase to the status pages it hosts during the October 20, 2025 AWS outage, peaking at around 4.5 million requests per hour. Status page traffic spikes when customers can't reach the product, which is why a status page needs to run on separate infrastructure. (incident.io, October 2025)

CrowdStrike, July 2024

Microsoft estimated that the faulty CrowdStrike update of July 2024 affected 8.5 million Windows devices, less than one percent of all Windows machines. Microsoft gave no method for the estimate, which it published the day after the outage. (Microsoft, July 2024)

Parametrix estimated that the July 19, 2024 CrowdStrike outage cost US Fortune 500 companies, excluding Microsoft, $5.4 billion in direct losses. Parametrix expected insurance to cover only a small share of that loss. (Parametrix, July 2024)

Downdetector received nearly 5 million reports during the July 19, 2024 CrowdStrike outage, more than three times the number of users who visit the site on an average day. Ookla corrected the figure from users to reports after publication, so cite it as reports. (Ookla, July 2024)

Cloudflare, November 2025

The November 18, 2025 Cloudflare outage started when a Bot Management feature file grew past a hard limit of 200 features, against about 60 normally in use. The postmortem traces the oversized file to a database permissions change that doubled its size. (Cloudflare, November 2025)

The November 18, 2025 Cloudflare outage drew over 3.3 million Downdetector reports across all impacted services and lasted nearly 5 hours, making it the third-largest outage of 2025. The count spans every site and app that failed behind Cloudflare, not Cloudflare's own services alone. (Ookla, December 2025)

About 18% of UptimeRobot users had at least one monitor go down during the November 18, 2025 Cloudflare outage, compared with about 1.2% on a normal day. UptimeRobot counts user accounts, not endpoints, so one affected monitor marks the whole account. (UptimeRobot, November 2025)

Google Cloud and Microsoft Azure, 2025

The June 12, 2025 Google Cloud outage drew over 1.4 million Downdetector reports across more than 50 online services. The count includes the Cloudflare services that failed the same day because they depended on Google Cloud. (Ookla, June 2025)

Google posted its first incident report about 1 hour after the June 12, 2025 crashes began, because its Cloud Service Health status page was down in the same outage. A status page that shares infrastructure with the product goes down exactly when customers need it. (Google Cloud, June 2025)

StatusGator's early warning signals flagged the June 12, 2025 Google Cloud outage 53 minutes before Google acknowledged it. StatusGator measures from spikes in user reports on its own platform, so the lead time depends on how quickly users complain. (StatusGator, December 2025)

Microsoft's Azure Front Door outage ran from 15:41 UTC on October 29 to 00:05 UTC on October 30, 2025, with connection timeouts and DNS resolution failures. Microsoft's own services that run on Azure Front Door were hit along with customer sites. (Microsoft, October 2025)

Microsoft posted its first public status page update at 16:18 UTC on October 29, 2025, after customer impact began at 15:41 UTC. Customers who relied on that page heard nothing official for more than half an hour, the gap independent monitoring fills. (Microsoft, October 2025)

Outages in 2026

Cloudflare's February 20, 2026 outage unintentionally withdrew 25% of its 4,306 Bring Your Own IP prefixes from the Internet. Customers who bring their own IP ranges to Cloudflare saw those services become unreachable for about 50 minutes. (Cloudflare, February 2026)

A power failure in Google Cloud's europe-west2-b zone in London on October 20, 2026 disrupted services in the europe-west2 region, with the main impact resolved within 6 hours and 8 minutes. Google's report shows a single-zone power failure spreading to regional services, the failure multi-zone setups are meant to contain. (Google Cloud, November 2026)

The Microsoft 365 outage that began November 12, 2026 drew more than 7 million Downdetector reports by 16:00 UTC on November 13, about two and a half times as many as YouTube's February 2026 outage, the next largest. Ookla published while the outage was still developing, so the final count is likely higher. (Ookla, November 2026)

Internet disruption statistics

Some outages have nothing to do with a provider's code. Governments order shutdowns, power grids fail, and ships drag anchors across submarine cables, and the datasets below count those events country by country.

Disruptions worldwide

Almost half of the 174 major Internet outages Cloudflare observed worldwide in 2025 were government-directed regional and national shutdowns. The count covers January 1 to December 2, 2025, and Cloudflare's later Q4 summary puts the full year above 180. (Cloudflare, December 2025)

Authorities shut down the Internet 313 times across 52 countries in 2025, the most the #KeepItOn coalition has documented. Access Now counts every deliberate shutdown, including regional and platform blocks, which is why its total is higher than Top10VPN's. (Access Now, March 2026)

Cloudflare Radar recorded 49 Internet disruptions in Q3 2026, 19 of them government-directed shutdowns, including 18 exam shutdowns in Iraq. Exam shutdowns recur on a schedule, so a single country can dominate a quarter's government-directed count. (Cloudflare, October 2026)

Internet Society Pulse recorded 916 Internet shutdown events across 61 countries between 2019 and 2025. Pulse lists only government-mandated shutdowns it has verified, so it excludes outages from accidents or attacks. (Internet Society, August 2026)

14 of the Internet disruptions Cloudflare Radar recorded in Q3 2026 were power-related, across 10 countries. Power was the second most common cause that quarter, after government-directed shutdowns. (Cloudflare, October 2026)

70.8% of the shutdowns Internet Society Pulse recorded from 2019 to 2025 lasted 1 day or less, and 95.4% ended within 30 days. Pulse counts each shutdown once regardless of length, so this share describes events, not hours of lost access. (Internet Society, August 2026)

Conflict was the leading trigger for Internet shutdowns in 2025 for a third year running, with 125 conflict-related shutdowns documented across 14 countries. Access Now says perpetrators used these shutdowns to conceal abuses and terrorize populations during fighting. (Access Now, March 2026)

Government shutdowns and their cost

Government-imposed Internet outages in 28 countries, lasting more than 120,000 hours, cost the global economy $19.7 billion in 2025. Top10VPN builds the estimate with the NetBlocks cost tool, which models lost economic output rather than measuring it. (Top10VPN, January 2026)

Myanmar had a record 95 Internet shutdowns in 2025, 76 of them carried out by the military regime. One country supplying almost a third of the global count is a reason to break shutdown totals down by country before comparing years. (Access Now, March 2026)

Top10VPN counted 212 major government-imposed Internet outages across 28 countries in 2025, the highest number it has recorded in a single year. Top10VPN counts only major national or regional outages, so its total is lower than Access Now's for the same year. (Top10VPN, January 2026)

Iran's nationwide Internet shutdown that began on February 28, 2026 lasted 88 days before restoration started on May 26, 2026. Cloudflare Radar saw the country almost entirely offline for the whole period. (Cloudflare, July 2026)

Major government-imposed Internet disruptions lasted 120,095 hours in total in 2025, a 70% rise on the previous year. The total mixes full blackouts, social media blocks, and throttling, which Top10VPN reports separately. (Top10VPN, January 2026)

Deliberate Internet restrictions affected 798.12 million people in 2025, according to Top10VPN's count of major shutdowns. Top10VPN estimates the affected population from the regions covered, so the precision of the figure overstates its accuracy. (Top10VPN, January 2026)

Submarine cables and power blackouts

About 150 to 200 submarine telecommunications cable faults occur worldwide each year. The ICPC bases this on repair data from cable maintenance providers, so it counts the faults that needed a repair. (ICPC, 2026)

Internet traffic fell 68% in Spain and 74% in Portugal during the April 28, 2025 Iberian blackout. Kentik measured traffic volume in bits per second across the networks it sees, not the share of people offline. (Kentik, May 2025)

70% to 80% of submarine cable faults are caused by accidental human activity, mainly commercial fishing and ships' anchors, rather than sabotage. An ITU advisory report puts the share above 80%, so cite one source and say which. (ICPC, 2026)

Median mobile download speeds on April 28, 2025 were 73% lower in Spain and 75% lower in Portugal than the day before. Speedtest data comes from people who could still run a test, so it understates how many had no connection at all. (Ookla, April 2025)

Deliberate damage accounts for less than 1% of submarine cable damage worldwide each year. Sabotage draws the headlines, but fishing gear and anchors cause almost all real faults. (ICPC, 2026)

What these downtime statistics mean in practice

The numbers point to four habits that shorten outages and keep customers informed while they last.

Monitor from outside your own stack

In the Splunk and New Relic surveys, more than 4 in 10 organizations find out about outages from customers or manual checks, and most outages Pulsetic confirmed were over within an hour. A check that runs every few minutes misses many of them. Check public endpoints from several regions at an interval of a minute or less, confirm failures from a second location, and page a person, not a dashboard. Hyperping's uptime monitoring works this way.

Map the providers you depend on

Third parties caused about two-thirds of the public outages Uptime Institute tracked, and the October 2025 AWS outage disrupted more than 3,500 companies. List the providers behind your login, payments, DNS, and email, monitor them like your own services, and decide in advance what you will tell customers when one of them fails.

Treat changes and certificates as outage risks

Changes started most of the cloud outages Control Plane documented and 79% of the Kubernetes issues in Komodor's data, and 86% of companies in Keyfactor's survey had a certificate-related outage in a year. Watch error rates for an hour after every deploy, and alert on certificate expiry for every public hostname well before the renewal date. Hyperping's SSL monitoring covers the second part.

Post the first update within minutes

Rootly's median team posts its first status update 9 minutes after declaring an incident, while 30% of US adults in Xurrent's survey got no useful information during their last disruption. Google's own status page went down in the June 2025 outage it needed to report. Host your status page apart from your own infrastructure and write the first update before you need it. A hosted status page solves the first half.

Methodology and sources

190 candidate figures were checked against their original publications, and 103 distinct statistics were kept. Duplicate or weakly defined claims were removed, no single study supplies more than 6 figures, and figures from unlike samples were never averaged. Each citation date is the original publication or data release. Last checked November 15, 2026.

Most of these studies come from vendors that sell a fix for what they measured, including some of Hyperping's competitors, so each statistic names who ran the study and what it covered. Figures that only appear on a vendor's gated report were left out unless the vendor published them on a public page. The widely repeated "$5,600 per minute" cost of downtime was left out on purpose: it comes from a July 2014 Gartner analyst blog post, not a study.

To cite a figure, link to the original study and to this page: "100+ downtime and outage statistics for 2026", Hyperping, updated November 2026, [page URL].

Sources

Splunk Hidden Costs of Downtime (May 2026) · New Relic Observability Forecast (September 2026) · Uptime Institute Annual Outage Analysis press release (May 2025) · Uptime Institute Annual Outage Analysis report (May 2026)

Downdetector AWS outage analysis (October 2025) · Access Now shutdowns in 2025 press release (March 2026) · ITIC hourly cost of downtime, part 1 (September 2024) · EMA IT outages: costs and containment (April 2024)

ITIC hourly cost of downtime, part 2 (September 2024) · PagerDuty cost of incidents study (June 2024) · Calyptix and ITIC SMB survey (February 2025) · Siemens True Cost of Downtime (2024)

Fluke unplanned downtime survey (October 2025) · Uptime Institute Annual Outage Analysis press release (May 2026) · Cockroach Labs State of Resilience (October 2024) · Catchpoint Internet Resilience Report (June 2025)

ThousandEyes outage counts (April 2026) · Parametrix Cloud Outage Risk Report (March 2026) · Pulsetic uptime statistics (November 2026) · Oh Dear uptime statistics (2025)

StackGen State of Reliability (August 2026) · ThousandEyes ISP and cloud outage ratio (January 2025) · StackGen status-page incident study (August 2026) · Unitrends State of Backup and Recovery (2025)

Control Plane Cloud Outage Report (October 2026) · DORA State of AI-assisted Software Development (September 2025) · Uptime Institute Global Data Center Survey (July 2025) · Komodor Enterprise Kubernetes Report (September 2025)

Red Hat State of Kubernetes Fleet Management (June 2026) · 451 Research outage causes (August 2026) · Keyfactor Digital Trust Digest (September 2025) · DigiCert certificate failure research (September 2026)

NeuBird State of Production Reliability (April 2026) · UptimeRobot status page reporting (October 2026) · Splunk State of Observability (October 2025) · Grafana Labs Observability Survey (March 2026)

Rootly incident management benchmarks (October 2026) · Xurrent service disruption survey (August 2026) · NoDisrupt State of Incident Response (2026) · AWS us-east-1 postmortem (October 2025)

Parametrix AWS us-east-1 analysis (October 2025) · PagerDuty on the October 20 outage (October 2025) · incident.io on the October 20 outage (October 2025) · Microsoft on the CrowdStrike outage (July 2024)

Parametrix CrowdStrike loss estimate (July 2024) · Downdetector CrowdStrike outage analysis (July 2024) · Cloudflare November 18 postmortem (November 2025) · Downdetector largest outages of 2025 (December 2025)

UptimeRobot Cloudflare outage report (November 2025) · Downdetector Google Cloud outage analysis (June 2025) · Google Cloud June 12 incident report (June 2025) · StatusGator 2025 cloud outages (December 2025)

Azure Front Door post-incident review (October 2025) · Cloudflare February 20 postmortem (February 2026) · Google Cloud europe-west2 incident report (November 2026) · Downdetector Microsoft 365 outage analysis (November 2026)

Cloudflare Radar Year in Review (December 2025) · Cloudflare Radar Q3 disruption summary (October 2026) · Internet Society Pulse shutdown tracker (August 2026) · Access Now KeepItOn report (March 2026)

Top10VPN cost of Internet shutdowns (January 2026) · Cloudflare Radar Q2 disruption summary (July 2026) · ICPC submarine cable faults (2026) · Kentik Iberian blackout analysis (May 2025)

Ookla Iberian blackout analysis (April 2025)