Does a VPN Help During an Internet Outage? Every Major Outage Since 2022, Checked
Every big outage produces the same question in search logs: will a VPN get me back online? The honest answer, across the fourteen major outages ledgered below, is usually no, sometimes partly, and in a few specific failure modes genuinely yes. Most pages answering this question sell the yes. This one maps where each answer actually applies, because knowing that a VPN cannot fix a dead line is as useful as knowing it can fix a broken resolver.
A definition to anchor the whole page: a VPN is a different path for your traffic, not a different connection; it can replace the route, the resolver and the visible endpoint of a working line, and it can replace nothing about a line, a power grid or a destination server that is down.
What actually took the internet down since 2022?
The ledger below covers the major, well-documented outages from 2022 through 2026, each with its root cause as stated by the operator or the investigating regulator, and a verdict on whether a VPN changed, or could have changed, anything for an affected user.
| Date | Event | Cause | Scale | Would a VPN have helped? | Source |
|---|---|---|---|---|---|
| Jul 8, 2022 | Rogers nationwide outage, Canada (15+ hours) | Maintenance update broke internal routing; Rogers withdrew its BGP routes | Roughly 12 million customers; Canada's internet traffic fell by about a quarter | No. The access line itself was dead; even Canada's Interac debit network went down with it. Only a different carrier helped | Cloudflare |
| Nov 8, 2023 | Optus outage, Australia (12-13 hours) | Routing-information flood from a peer's upgrade; Optus routers self-isolated | About 10 million customers | No. Subscribers had no path out to tunnel through | Kentik |
| Feb 22, 2024 | AT&T US wireless outage (about 12 hours) | Misconfigured network element put the network in protect mode | Over 125 million devices; more than 92 million calls blocked; over 25,000 failed 911 attempts | No. Phones could not register on the network at all. Wi-Fi calling helped; a VPN did not | FCC report |
| Feb 2024 | Red Sea submarine cable cuts (Seacom, TGN, AAE-1, EIG) | Cable damage, attributed to the anchor of the sinking Rubymar | About 25 percent of Asia-Europe-Middle East traffic diverted, per HGC | Partly. Carriers rerouted automatically; where the surviving paths congested, a VPN exit in a better-connected region could pick a cleaner route to specific destinations | HGC/DE-CIX |
| Mar 14, 2024 | West Africa four-cable cut (WACS, MainOne, SAT-3, ACE) | Subsea cable damage, likely seabed movement | At least eight countries degraded; the four cable systems lost 90 to 100 percent of capacity, per Ghana's telecom regulator; repairs ran into May | Partly. Where any upstream capacity survived, tunneling to a European server could stabilize sessions; where a country's transit was gone, nothing helped | Data Center Knowledge |
| Jul 19, 2024 | CrowdStrike Falcon update crash | Faulty endpoint-security update; Windows machines blue-screened | About 8.5 million Windows devices, per Microsoft | No. The internet was fine; the computers were not. Included because "the internet is down" headlines often describe exactly this | Microsoft |
| Nov 17-18, 2024 | Baltic cable cuts (C-Lion1, BCS East-West) | Cable damage, sabotage suspected | One cable carried about a third of Lithuania's capacity; Cloudflare saw no national traffic drop | Not needed. Redundancy absorbed it; users never noticed. The row that shows what good peering buys | Cloudflare |
| Apr 28, 2025 | Iberian power grid collapse | National electricity failure in Spain and Portugal | Portugal's traffic about 90 percent below normal within five hours; Spain about 80 percent; decaying as backup power drained | No. No power means no modem and no cell tower. A VPN is software on hardware that is off | Cloudflare |
| Jun 12, 2025 | Google Cloud global outage (about 3 hours) | Malformed quota-policy data replicated globally, crashing API validation | All GCP regions; dragged down Cloudflare Workers KV and many SaaS apps | No. The destination's backend was down; changing your own path cannot fix the server's side | Cloudflare |
| Oct 19-20, 2025 | AWS us-east-1 outage (about 14 hours to full recovery) | Race condition in DynamoDB's DNS automation left an empty DNS record, cascading across services | Thousands of dependent apps, from banks to smart devices | Mostly no. Apps hosted there were down for everyone. Partly, for the few multi-region apps that route users by location: an exit elsewhere could land on a healthy region | The Register on AWS's postmortem |
| Oct 29, 2025 | Azure Front Door global disruption (about 8 hours) | Bad configuration deployed globally past a defect in deployment safety checks | Azure, Microsoft 365, Entra and Intune served errors worldwide | Mostly no. Partly where only some edge nodes failed: a different egress sometimes reached a healthy node | Azure status history |
| Nov 18, 2025 | Cloudflare global outage (about 6 hours to stability) | An oversized Bot Management feature file hit a hard limit and panicked the core proxy | Millions of Cloudflare-fronted sites served errors | No. The CDN in front of the sites was broken on every path; your route to it was irrelevant | Cloudflare postmortem |
| Q1-Q2 2026 | Iran near-total national shutdown, 88 days | Government-ordered shutdown | Traffic stabilized at about 59 percent of pre-shutdown levels after restoration | No during the blackout itself: the line is cut by decree. Blocking short of shutdown is censorship, a different problem with different answers | Cloudflare Q2 2026 report |
| Q2 2026 | Sudan and Iraq exam shutdowns | Scheduled government shutdowns during national exams | 13 shutdowns in one quarter: 10 in Sudan, 3 in Iraq | No. Full network-level shutdowns sever the line itself | Same Cloudflare report |
Count the verdicts and the pattern is stark: of fourteen major outages, a VPN was flatly useless in nine, partly useful in four, and unnecessary in one. That is the honest baseline this page is built on. The partly rows share a shape: some path, somewhere, still worked, and the question was whether you could reach it.
When does a VPN actually help, by cause?
The ledger sorts into eight failure classes, and the VPN verdict is a property of the class, not the headline.
| Cause class | What breaks | Can a VPN route around it? |
|---|---|---|
| Last-mile ISP failure (Rogers, Optus, AT&T) | Your own access line or mobile registration, and everything behind it | No. A VPN rides on the connection you have. The workaround is a second carrier, not a tunnel |
| Submarine cable cut (Red Sea, West Africa, Baltic) | International transit capacity; latency and loss spike on surviving routes | Partly. Carriers reroute on their own, but a VPN exit in a well-connected region can give you a cleaner, differently-routed path while congestion lasts |
| Routing or BGP error affecting specific destinations | Reachability of particular sites; traffic blackholed or misdirected | Partly, and this is the VPN's best case: if your ISP's route to a site is broken but the VPN server's route is clean, a different egress restores access |
| Cloud or CDN provider outage (AWS, Azure, Cloudflare, GCP) | The destination's servers, or the CDN in front of them, for everyone | No, with a narrow exception for geo-routed multi-region apps and partial edge failures, where a different exit point sometimes lands on healthy infrastructure |
| DNS resolver failure at your ISP | Name lookups fail, so sites look down although their servers are up | Yes. A VPN tunnel carries your DNS queries to the VPN provider's own resolvers, bypassing the broken one entirely |
| Government shutdown (Iran, Sudan, Iraq) | The national network is ordered off at the line or routing level | No against a true shutdown. Filtering short of a shutdown is censorship, not an outage, and has its own toolbox |
| Power grid failure (Iberia 2025) | Modems, routers, towers and data centers as batteries drain | No. Connectivity dies with the electricity |
| Endpoint software failure (CrowdStrike) | The computers themselves; networks are untouched | No. Nothing network-side was wrong |
The second table earns one paragraph of reading. The two yes-shaped rows, resolver failure and destination-specific routing breakage, have something in common: the outage is partial and positional. Your line works, most of the internet works, and what failed is the particular path or lookup chain between you and what you want. A VPN substitutes its own path and its own resolvers for your ISP's, which is exactly the swap those failures call for. That is also why a VPN does nothing for the total failures: there, no alternative path exists on your side of the problem, or no server exists on the other side of it.
What should you actually do in each case?
A short, honest decision path for the moment things stop loading. If nothing works on any device and your router's lights are wrong, it is your line or the power: switch to mobile data or a second carrier, and no software will help. If some sites work and others do not, try another DNS resolver or connect your VPN; if the dead site comes back through the tunnel, you have found a resolver or routing failure, the one class where a VPN's separate network path genuinely fixes the outage. If one big service is down everywhere, check an outage tracker and wait; nobody's tunnel revives a crashed cloud region, as the September 2026 AI outages demonstrated to millions of users at once. And during a deliberate shutdown, understand what regime you are under: a full shutdown cuts everyone, while filtering leaves the line up, and the difference decides whether anything, including a VPN, can matter. There is a quieter, year-round reason to keep the tunnel on anyway: on a working line, it carries your traffic encrypted past whatever the local network is doing, outage or not.
Why does every outage feel like "the internet is down"?
Because concentration has made unrelated failures look identical from a couch. Cloudflare's Radar team counted 174 major internet outages in 2025, and the biggest items on that year's list were not cables or ISPs but three cloud incidents in five weeks: AWS in October, Azure in October, Cloudflare in November. When a provider that fronts millions of sites fails, the symptom at home, everything broken at once, is indistinguishable from your ISP dying, even though the fixes are opposites. That is the practical reason to learn the classes in the second table: the first diagnostic question is never "is the internet down", it is "whose part of it".
What changes next year
Three files to watch. The FCC's rules tightening outage reporting after the AT&T failure continue to produce better public postmortems, which is where this ledger's future rows come from. Subsea-cable protection moved up NATO's and the EU's agendas after the Baltic cuts, with repair-ship capacity the acknowledged bottleneck for the next West Africa-scale event. And the cloud concentration question got its first real test series in late 2025; whether AWS, Azure and Cloudflare's post-incident changes hold will show in whether 2026's outage count beats 2025's 174. This page keeps its ledger format and will take new rows as the postmortems land.
About the author
Le VPN Blog Editor
Alan Summers has been writing and editing for the Le VPN blog for years, covering online privacy, cyber security, and the best ways to get the most out of a VPN. He keeps a close eye on the news that affects internet freedom around the world and turns it into practical advice for Le VPN readers.
Articles by Alan Summers →
