Iraqi authorities switched off internet access 18 times between July and September 2026 to stop cheating in school exams, making one government's exam timetable the largest single source of disruption in Cloudflare's count for the third quarter. The figures come from the company's quarterly review of internet disruptions, published on October 27 by Lai Yi Ohlsen, which records 49 disruptions observed on Cloudflare's network between July 1 and September 30. Government-directed shutdowns accounted for 19 of them, power outages for 14 and natural disasters for five.

The rest of the quarter was shaped less by policy than by physics. Electricity grids failed in Cuba, Honduras, Ukraine and seven other countries. Earthquakes struck Japan and Colombia, flash floods hit Nepal, and Guam took its second super typhoon in less than three months. Renewed conflict in Ethiopia's Tigray region brought a 10-day shutdown during which traffic from the region fell by 100%. In Phoenix, a failed cooling system at a data center took a domain company's network almost entirely offline.

In Short

Cloudflare, a company whose network carries a large share of the world's web traffic, counted 49 times between July and September 2026 when the internet broke down somewhere. Most of those breaks came from governments switching the internet off, almost all of them in Iraq during school exams, or from electricity failing, because routers and phone masts stop working when the power goes. For anyone who runs a website or buys ads, these breaks show up as sudden holes in traffic and sales from the affected places, and Cloudflare's public data is one way to see why a country went quiet.

What Cloudflare counted

The review draws on Cloudflare Radar, the company's public data platform. According to Cloudflare, "Because Cloudflare's global network handles more than 20% of the web's traffic, it offers a unique perspective of how Internet traffic is behaving." Radar observes the effect of failures "by monitoring traffic to our customers," according to the post, and each quarterly review covers the disruptions "detected and annotated on Radar." The percentages that follow therefore describe changes in requests reaching Cloudflare from a given country or network. They are a proxy for connectivity, not a census of every user.

That 20% figure has appeared in different forms. When Cloudflare redesigned the Radar landing page on October 8, it described sitting in front of roughly a fifth of all web traffic, and PPC Land noted at the time that the company's statements do not always use the same denominator.

The Q3 2026 review does something its predecessors did not: it gives a total. Cloudflare's Q3 2025 summary and its Q2 2026 summary both described themselves as overviews of notable, confirmed disruptions rather than complete lists, and directed readers to the Radar Outage Center for more. This quarter's post states 49.

The breakdown it provides covers 38 of them. Government-directed shutdowns were 19, including 18 exam shutdowns in Iraq. Power outages were 14. Natural disasters were five. That leaves 11 disruptions, by PPC Land's arithmetic, spread across the remaining categories on Cloudflare's map legend: cable cuts, cyberattacks, network problems, technical problems, mechanical failures and unknown causes. The post does not give counts for those categories. It names four natural disasters, in Guam, Japan, Colombia and Nepal, rather than five.

By PPC Land's calculation, Iraq's exam shutdowns alone represent about 37% of the quarter's 49 disruptions. Government-directed shutdowns as a group represent about 39%, and power outages about 29%.

For scale, Cloudflare counted 174 outages across the whole of 2025, of which 83 were government-directed shutdowns, 25 were power outages and 19 were cable cuts, as PPC Land reported in December 2025. The 14 power-related disruptions in this single quarter equal 56% of that full-year power figure, again by PPC Land's calculation. The annual and quarterly reviews are compiled separately, so the comparison indicates direction rather than a like-for-like trend.

Iraq's exam timetable

Iraq accounted for 18 of the quarter's 19 government-directed shutdowns. According to Cloudflare, they "took place on a strict schedule" between 03:30 and 04:00 UTC and lasted "about 15-90 min each." Three fell between July 1 and July 8. Most of the rest came between August 26 and September 28.

The two figures do not quite fit together: a 30-minute window cannot contain a 90-minute shutdown, and the post does not say whether the window refers to start times. Internet Society Pulse, which Cloudflare cites, logged the September 28 shutdown from 06:20 to 07:10 local time, or 03:20 to 04:10 UTC, a 50-minute nationwide outage. According to Pulse, that shutdown belonged to a series of expected shutdowns during the second round of Iraq's school exam schedule, which +964Media expected to run until October 10. That would carry the series into the fourth quarter, which began on October 1. Pulse's shutdown history for Iraq lists 29 shutdowns in the previous 12 months and 186 since 2019.

The pattern repeats from year to year. In the third quarter of 2025, Cloudflare recorded high school exam shutdowns in Iraq from August 26 through September 13, between 03:00 and 05:00 UTC, according to its Q3 2025 summary. The 2026 series began on the same calendar date. Earlier in 2025, Cloudflare's Q2 2025 review recorded daily restrictions from May 20 to July 3, also between 03:00 and 05:00 UTC. In the second quarter of 2026, Cloudflare logged three Iraqi shutdowns, on June 2, June 11 and June 28, each lasting about 90 minutes.

Is there an end in sight? Cloudflare quotes Internet Society Pulse on the government's stated plan: "In a video message, Minister Mustafa Sanad, Iraq's Minister of Communication, confirmed the shutdown and vowed to develop a digital system that would eliminate the need for centralized exam distribution and, by extension, the need to shut down Internet access on exam days." The post gives no timetable for that system, and Cloudflare says only that Radar will keep monitoring Iraqi networks as it develops.

Syria and Sudan, which have also used exam-related shutdowns in recent years, recorded none this quarter. Cloudflare's reading is that "this likely reflects differences in exam calendars rather than a clear shift in policy." The company's own records add some texture to that. Sudan's shutdowns came in the previous quarter: Cloudflare's Q2 2026 summary lists 10 between April 13 and April 23, each lasting about three and a half hours. Syria's 2025 exam disruptions, on the other hand, fell squarely in the third quarter, with six cellular shutdowns between July 12 and August 3, 2025. The Q3 2026 post does not give Syria's 2026 exam dates.

Power: the dependency underneath

Power outages were the second-most common cause by incident count, behind the Iraqi shutdowns. Cloudflare recorded 14 power-related disruptions across 10 countries and summed up the pattern plainly: "when power systems become unstable, Internet connectivity often becomes unstable with them."

Cuba

Cuba's connectivity followed the failures of its grid. Radar observed four power-related outages during the quarter, on July 6, July 14 and August 3, and a three-day blackout from September 18 to September 20. Traffic declines ranged from 18% on July 14 to 59% on August 3, according to Cloudflare. That is at least one grid-driven disruption every month of the quarter.

Honduras

The Honduran case is the one with the clearest lesson about network design. On September 10, a power failure at a single interconnection facility affected at least 11 major providers. HTTP traffic across the country fell by 63% for approximately three hours. Cloudflare described the incident as "showing how concentrated traffic exchange points can become single points of failure."

Two networks were unaffected: the cross-border provider Telecomunicaciones de Guatemala (AS14754) and the mobile carrier Celtel (AS52262). Cloudflare suggests their Honduran connectivity "may rely more heavily on infrastructure outside the country." A country-wide drop of nearly two-thirds, from one building losing power, is a measure of how much traffic depends on a handful of exchange points.

Sevastopol

On July 5, a power outage in Sevastopol caused an eight-hour blackout, with HTTP traffic falling by 98%. On August 12, drone strikes on energy infrastructure caused another outage, a ten-hour disruption with traffic down by up to 95%. Cloudflare lists both under Ukraine. Sevastopol is in Crimea, which Russia annexed in 2014. The post does not say who carried out the August strikes. Cloudflare's conclusion was that the incidents "show how conflict, energy infrastructure, and Internet connectivity are fundamentally intertwined."

Seven more countries

Power-related disruptions also appeared in Eswatini, Botswana, Tajikistan, Kenya, Saint Lucia, Georgia and Gabon. Some were sudden failures. Others were deliberate: Eswatini's outages reflected planned load-shedding, in which a utility cuts supply to parts of the grid on a schedule to manage a shortage. Cloudflare does not give dates or traffic figures for these seven.

A cooling failure in Phoenix

One of the quarter's outages happened inside a data center in the United States, which Cloudflare groups with Japan and Portugal as countries with a relatively strong history of internet resilience that were still not immune to disruption. On August 13, a cooling system failure at a facility in Phoenix, Arizona, caused a near-total drop in traffic from Namecheap's network (AS22612), according to Cloudflare. Namecheap is a domain registrar that also sells hosting; at the end of October 2025 it held 12.3 million .com names, second to GoDaddy's 52.5 million.

The post does not say how long the Namecheap outage lasted or which customers were affected. Its comment on the episode looks forward: "Outages such as these will be important to watch as the role of cooling systems becomes increasingly critical." Servers convert almost all the electricity they draw into heat, and that heat has to leave the building. As hyperscalers and colocation operators pack more power-hungry hardware into each rack, the cooling plant is as much a single point of failure as the grid connection.

Storms, earthquakes and floods

Guam

Guam was hit twice in one year. In the second quarter, Super Typhoon Sinlaku knocked out power on the island, and traffic fell as much as 80% below expected levels on April 13 and 14, according to Cloudflare's Q2 2026 summary. HTTP traffic took four days to return to typical levels.

Super Typhoon Bavi followed less than three months later. KUAM reported on July 6 that the Guam Power Authority had begun recovery work after widespread outages. Officials took earlier preparedness steps this time, according to Cloudflare, including a state of emergency declared before peak storm conditions. Yet the outcome was familiar. Strong winds brought down key transmission lines, and at one point only about 25% of distribution feeders were online, according to the Guam Power Authority figures Cloudflare cites. The result, once again, was a multi-day recovery for internet traffic.

Japan

A magnitude 7.1 earthquake in Kumamoto prompted evacuation orders affecting roughly 150,000 people, according to Cloudflare. The quake struck on July 28, and Kyushu Electric Power said about 40,000 homes lost power, according to a report published that day. The internet impact was unusual in its shape: it appeared mostly in desktop traffic, while mobile traffic remained largely unaffected. The post does not explain the split.

Colombia

Colombia's earthquake was the most damaging to connectivity among the quarter's natural disasters. A magnitude 7.4 quake struck on August 10 at 12:30 UTC near San José del Palmar, with damage reported in Quibdó, Pereira, Manizales and Cali, according to the Copernicus Emergency Management Service. Cloudflare recorded building collapses, injuries and mass evacuations, and a disruption lasting four and a half days. Damage to fiber-optic infrastructure drove a significant drop in traffic from Colombia Telecomunicaciones (AS3816). The post gives no percentage for that drop.

Nepal

On August 26, flash floods swept through the Bhote Koshi and Trishuli river systems, affecting communities across the Rasuwa, Nuwakot and Dhading districts. The floods damaged homes, infrastructure and power supplies. The resulting outage on Radar lasted two hours, during which HTTP traffic dropped by up to 18%.

Conflict, elections and contested labels

Tigray

In Ethiopia, renewed conflict in the Tigray region came with an internet shutdown that lasted 10 days, during which traffic fell by 100%, according to Cloudflare. Although the outage appears to have stayed within Tigray, it pulled country-level HTTP traffic down by up to 30%. Cloudflare recalled the 2020 to 2022 civil war, when Tigray was largely cut off from telecommunications, electricity, banking, fuel and humanitarian aid, a situation the World Health Organization described as a "de facto blockade."

Cloudflare classifies the Tigray outage as government-directed but does not name which authority ordered it. Reporting at the time pointed in two directions. On September 25, Reuters reported that four people with knowledge of the matter said the Tigray People's Liberation Front had ordered Ethio telecom and Safaricom to shut their networks, while a source close to the TPLF denied this and blamed the federal government. Neither operator responded to Reuters.

Russia's Far East

On September 19, during Russia's nationwide elections, fiber-optic cable damage in the Far East disrupted connectivity in Chukotka, Magadan Oblast, Kamchatka Krai and Sakhalin Oblast. Cloudflare says Rostelecom described the incidents as deliberate damage to backbone fiber lines and restored connectivity through backup channels.

The sourcing behind that sentence needs a caveat. The article Cloudflare links for it, from The Record, is dated April 6, 2026, and describes a different event: a DDoS attack on Rostelecom that disrupted service in about 30 cities. A Russian news report from September 19 described two breaks, on the Kholmsk-Tomari line in Sakhalin and the Selikhino-Yagodny section in Khabarovsk Krai. That report attributed the description of deliberate damage to the Ministry of Digital Development and said communications were restored within about an hour.

Cloudflare was careful about the timing. "Radar data alone cannot determine whether the cable damage was related to the elections, but the timing is notable," the post says.

Failures inside single networks

Beyond the country-level events, Radar recorded outages confined to individual networks. Cloudflare gives causes for two of them.

One Communications (AS19863) in Guyana lost service after a cable cut. On September 23, Demerara Waves reported that a fiber cable connected to the subsea cable had been cut during drainage work, with services restored gradually that evening.

MEO (AS3243) in Portugal recorded an outage following what Cloudflare calls a reported "mass attack." The Portugal News reported on September 15 that MEO described a cyberattack causing congestion and degradation of its international network, and said no customer data had been accessed.

The other three received no explanation in the post: Digicel Haiti (AS27653), TurkNet (AS12735) in Turkey and Fanap Telecom (AS206065) in Iran. The Radar views Cloudflare links for them cover July 30 to 31 for TurkNet, September 22 to 27 for Digicel Haiti and September 27 to 28 for Fanap Telecom. Digicel Haiti is a recurring entry: Cloudflare's Q2 2025 review recorded fiber damage on its network, and its Q3 2025 summary recorded two fiber cuts on August 26, 2025 that cut traffic by about 80%.

How fast traffic came back

Many disruptions were short. According to Cloudflare, 14 saw traffic return to typical levels in under an hour, and 16 recovered in less than three hours. The post does not make clear whether the 16 includes the 14. Read as separate groups, 30 of the 49 disruptions ended within three hours; read cumulatively, only 16 did.

Brevity is not the same as harmlessness, Cloudflare argued, especially during storms, earthquakes, conflict or political unrest. "Access to the Internet is often access to information, coordination, safety, and help," the post says. Its closing assessment widened the frame further: internet resilience "depends on electricity, physical infrastructure, policy decisions, and the ability of networks to route around failure."

Why the numbers matter for advertising and publishing

Every disruption in the review is also a gap in someone's data. When traffic in Honduras fell 63% for three hours, the impressions, sessions and conversions that would have come from Honduran users during that window simply did not happen. Exam shutdowns in Iraq recur at the same hour on known dates, which makes them appear as regular dips in any dataset that includes Iraqi traffic. And a near-total drop from Namecheap's network meant sites and services running on that network were unreachable for the duration, whatever their owners were spending to bring visitors in.

The concentration theme runs through the industry's recent history too. The Honduran exchange point is a national-scale version of a problem the advertising business met when AWS's US-EAST-1 region failed on October 19 and 20, 2025, an outage lasting more than 15 hours that also degraded Amazon's own retail media tools. Cloudflare itself is part of that picture. On November 18, 2025, an oversized Bot Management configuration file crashed the proxy at the core of its network, and sites behind its CDN returned HTTP 5xx errors for about three hours before the main impact was resolved, according to Cloudflare's incident report, which called it the company's worst outage since 2019.

Physical conflict reaches the infrastructure underneath the internet as well. Drone strikes on energy infrastructure produced Sevastopol's August outage. In October, after the quarter closed, YandexBot traffic on Cloudflare's network collapsed after drone strikes forced Yandex to shut or partly disable three data centers, and Cloudflare later found that bots displaced from Yandex.Cloud reappeared on other Russian hosts. Those events fall into the fourth quarter and will presumably feature in the next review.

Radar data has become a common reference point for explaining such events. PPC Land relied on it when Iran's nationwide shutdown on January 8, 2026 cut traffic by nearly 90% within 30 minutes. The data has limits, though. It measures traffic to Cloudflare customers, it attributes causes using public reports, and its quarterly reviews are written by a vendor that sells resilience and security services. The Russian example above shows why the cited sources deserve a second look.

What the Q3 2026 review establishes is narrower and still useful: 49 counted disruptions, a government that cuts access on an exam timetable, and a long list of places where the internet failed because the electricity did.

Timeline

Summary

Who: Cloudflare, through its Radar data platform and a quarterly review written by Lai Yi Ohlsen. The events involve Iraq's government and its communications minister, grid operators in Cuba, Honduras, Guam and seven other countries, Namecheap, Colombia Telecomunicaciones, Rostelecom, MEO, One Communications, Digicel Haiti, TurkNet and Fanap Telecom.

What: Cloudflare recorded 49 internet disruptions in the third quarter of 2026. Government-directed shutdowns accounted for 19, including 18 exam shutdowns in Iraq; power outages for 14 across 10 countries; and natural disasters for five. Notable events included a 63% national traffic drop in Honduras from one interconnection facility, a 10-day shutdown in Tigray, a cooling failure that took Namecheap's network almost entirely offline, and a 7.4 earthquake in Colombia that disrupted traffic for four and a half days.

When: The disruptions took place between July 1 and September 30, 2026. Cloudflare published the review on October 27, 2026.

Where: Iraq, Cuba, Honduras, Sevastopol (listed by Cloudflare under Ukraine), Eswatini, Botswana, Tajikistan, Kenya, Saint Lucia, Georgia, Gabon, Guam, Japan, Colombia, Nepal, Ethiopia, Russia, the United States, Guyana, Portugal, Haiti, Turkey and Iran.

Why: The review shows how much internet availability depends on electricity, physical infrastructure and government decisions. For advertisers and publishers, each disruption removes traffic, impressions and conversions from the affected market, and concentrated points such as a single exchange facility or data center can take a whole country or network offline at once.