Cloudflare publishes a quarterly review of Internet disruptions based on its Radar traffic data. The Q3 2026 edition counts 49 disruptions between July and September: 19 government-directed shutdowns (18 of them exam-related shutdowns in Iraq), 14 power outages across 10 countries, and 5 natural-disaster events. The rest were cable cuts, a suspected attack and other network-specific failures. Cloudflare says most of these were brief, with 14 recovering in under an hour. Its conclusion is that Internet resilience depends on electricity, physical infrastructure, policy, and whether networks can route around failures.
This matters for builders because these are the failure modes that redundancy plans usually ignore. A cloud region's own SLA doesn't help when a national grid, a domestic interconnection point or a government order removes your users' access. A single colocation facility losing cooling can also take a whole provider's control plane offline. The newsroom asked for the specific incidents and the evidence behind them. This note pulls them from the primary post and checks the biggest ones against independent reporting.
What we checked
- Which specific Q3 disruptions did Cloudflare measure, and what traffic, duration, geography, and recovery evidence supports its conclusions? partly confirmed. Cloudflare names specific incidents with dates, ASNs and traffic-drop percentages, and independent reporting confirms the underlying events for Honduras, Colombia, Tigray, Iraq, Cuba and Namecheap. The traffic figures come only from Radar and cannot be checked from outside. Several incidents (Eswatini, Botswana, Kenya, Digicel Haiti, TurkNet, Fanap) are listed with little or no measurement detail. [1][2][5][4]
- Is Cloudflare's label of Tigray as a 'government-directed' shutdown uncontested? refuted. Reuters, citing four sources, reported that the Tigray People's Liberation Front ordered Ethio telecom and Safaricom to cut service. A source close to the TPLF denied this and blamed the federal government. Cloudflare's label does not say which authority ordered the shutdown, and who did is disputed. [5][1]
- Was the Honduras outage a local facility failure or part of a wider grid event? partly confirmed. Cloudflare describes a power failure 'at an interconnection facility' that hit at least 11 providers. Infobae reports a failure of the Central American Regional Interconnection System around 11:47 local time on September 10, which also blacked out Guatemala, El Salvador and Nicaragua. The exact origin was still undetermined. [1][2]
What we found
- Honduras is the clearest case of concentration risk. Cloudflare says a September 10 power failure at an interconnection facility affected at least 11 major providers for about three hours, and national HTTP traffic fell 63%. Telecomunicaciones de Guatemala (AS14754) and Celtel (AS52262) were unaffected, which Cloudflare suggests is because they rely on infrastructure outside Honduras. Infobae places the trigger in the Central American Regional Interconnection System at 11:47 local time, with blackouts in Guatemala, El Salvador and Nicaragua as well. Investigators had not yet determined where it started. [1][2]
- Iraq ran 18 exam shutdowns in Q3, each lasting roughly 15 to 90 minutes in the early-morning UTC window. Three fell between July 1 and 8, and most came between August 26 and September 28. Iraq's Council of Ministers approved a 06:00–07:30 local suspension on May 19, 2026. Communications Minister Mustafa Sanad (also spelled Sand) proposed a randomized digital question-distribution system to end the practice. Access Now counts at least 34 Iraqi disruptions in 2026, fewer and shorter than in 2025, and warns that similar pledges have been broken before. It also notes that Syria imposed no exam shutdowns in 2026. [1][3][4]
- In Tigray, Cloudflare measured a 10-day shutdown in which regional traffic fell 100% and Ethiopia's national HTTP traffic fell up to 30%. Reuters reported that networks went down on September 25 as fighting resumed. Ethio telecom calls failed, and Safaricom customers had only limited connectivity. Reuters sources attributed the order to the TPLF, a claim a source close to the group denied, and NetBlocks independently confirmed the disruption. [1][5]
- A single facility took a hosting provider down. Cloudflare saw a near-total traffic drop for Namecheap (AS22612) on August 13 after a cooling failure at a data center in Phoenix. Domain Name Wire reports that a storm knocked out cooling at the RadiusDC facility. Hosting, EasyWP and email went down, and DNS resolution kept working but could not be managed. Services were back on the afternoon of August 14, and Namecheap said it would consider adding redundancy. [1][6]
- Cuba's unstable grid caused four disruptions in the quarter. On August 3 traffic fell 59%, and Cloudflare describes a three-day blackout from September 18 to 20. France 24 reports that the September 18 collapse began with a high-voltage transmission failure in central Cuba around 2pm. It was at least the sixth partial or total collapse since a US oil blockade began in January, and that evening only scattered Havana neighbourhoods around hospitals had power back. [1][7]
- Physical damage from natural disasters lasted longer than shutdowns. Cloudflare reports a 4.5-day disruption at Colombia Telecomunicaciones (AS3816) after the August M7.4 earthquake damaged fiber. Major outlets confirm a magnitude 7.4 quake in western Colombia on August 10, 2026, with at least 111 dead in early reports. Cloudflare also reports outages in Sevastopol: one lasted eight hours with traffic down 98% (July 5), and drone strikes on energy infrastructure caused another of ten hours (August 12). It notes outages in Guam after Super Typhoon Bavi, and a Japanese earthquake that hit desktop traffic but largely spared mobile. This note did not independently verify the Sevastopol, Guam or Japan figures. [1][8]
What it means
For anyone running services with users in specific countries, the main lesson is that the last mile and the national grid are part of your availability budget, and you can't buy them away. The Honduras case shows that one domestic interconnection point or one regional power grid can take out most providers at once. The providers that stayed up were the ones with connectivity outside the country. If you depend on a market like that, ask your carriers and CDNs where their traffic actually leaves the country. In synthetic monitoring and incident triage, treat a sudden drop across all ISPs at once as a possible infrastructure or government event, not your own bug. The scheduled Iraqi shutdowns are predictable enough to put on an operations calendar: avoid running time-sensitive jobs, OTP-based logins or payment flows in that early-morning UTC window on exam days.
The Namecheap case is a warning about vendor concentration closer to home. If your DNS management, email and hosting sit with one provider in one facility, losing that facility's cooling can lock you out of changing DNS, even while resolution keeps working. Keep DNS at a provider separate from hosting, keep registrar access and secondary DNS independent, and don't put a single data center in the critical path of your status page or incident tooling. Finally, treat Radar percentages as one vendor's view based on its own traffic. Who ordered a shutdown, as in Tigray, and where a failure started, as in Honduras, need separate sourcing.
Still open
- What exactly is the Honduran 'interconnection facility' Cloudflare refers to, an Internet exchange or the regional electric interconnection, and what was the root cause per Central America's regional grid operator (EOR)?
- Who ordered the Tigray shutdown, and has connectivity been fully restored across the region?
- Cause, duration and traffic impact for the Digicel Haiti, TurkNet, Fanap Telecom and MEO ('mass attack') outages, which the post gives little detail on.
- Will Iraq's proposed digital exam system actually be deployed before the 2027 exam season?