Why it matters
The ISP as a single point of failure
What depending on one access provider means for resilience.
Updated 29 Oct 2026
The dependency
A typical home has one connection to one provider. Behind that connection are shared pieces that most customers never see: the provider's routers, its routing configuration, its connections to other networks, and its management systems. A fault in any shared piece affects every customer behind it, however good their own equipment is.
This is a structural point rather than a statistic. We did not find a sourced figure for how concentrated last-mile provision is, so none is given here.
Outages that show the pattern
Rogers, Canada: 8 July 2022
The outage began at about 4:30 a.m. EDT and lasted from about 15 hours to several days for some customers. More than 12 million users were affected. Rogers told the regulator, the CRTC, that "the deletion of a routing filter on its distribution routers caused all possible routes to the internet to pass through the routers, exceeding the capacity" of core equipment, during phase 6 of a 7-phase update. Interac debit payments and some 911 calls on Rogers phones were disrupted. The CRTC published an executive summary in July 2024 and set mutual-assistance and emergency-roaming requirements.
Optus, Australia: 8 November 2023
The outage began at 04:05 AEDT and lasted about 12 to 13 hours, with 88% of service restored by 13:00. More than 10 million people and 400,000 businesses were affected. A BGP routing problem followed a planned software upgrade at a Singtel exchange in North America, and about 90 edge routers disconnected as a protective measure. Landline emergency 000 calls failed.
Facebook (Meta): 4 October 2021
This was not an access provider failure, but it shows the same class of fault. A maintenance command on Meta's backbone "unintentionally took down all the connections in our backbone network", and a faulty audit tool failed to stop it. Meta's DNS servers then withdrew their BGP advertisements when they saw unhealthy connections, which made them unreachable. Engineers had to go on site to restart systems. We have not given a duration because the source text we read does not state one.
What the three have in common
Each began with a routine change or maintenance action rather than an attack or a storm. Each took down a large group of users at once. In the Rogers and Optus cases, emergency calling was affected. The details differ, but in every case the users could do nothing locally to restore service.
Deliberate shutdowns
Governments also cut connectivity on purpose. Access Now's #KeepItOn campaign recorded 313 shutdowns in 52 countries in 2025, which it describes as "the highest number since 2016". The report, "Rising repression meets global resistance", was launched on 31 March 2026, and about half the shutdowns were in Asia Pacific. Shutdowns exploit the same centralisation: a handful of licensed operators can be instructed to switch off an entire region.
What a mesh changes, and what it does not
| Situation | Conventional ISP customer | Member of a community mesh |
|---|---|---|
| Provider's core network fails | Offline | Usually offline from the internet, if the mesh uses a single upstream |
| Local street or building link fails | Offline | Traffic may re-route over other nodes |
| Neighbours want to talk to each other during an outage | Needs the provider's network | Can work if local nodes and power are intact |
| Government orders shutdown | Offline | Local traffic may continue; internet access normally does not |
The benefit is mainly local connectivity: nodes that can still reach each other when the uplink is gone. NYC Mesh, for example, describes itself as a neutral network that does not "monitor, collect, store or block any user data or content", with volunteer-run uplinks. A mesh with several independent uplinks can lose one and keep the rest, but that depends on how its operators have built it.
LoRa meshes are different again. They never offered general internet access, so an ISP failure does not take anything from them; equally, they cannot replace the lost service. See LoRa vs Wi-Fi mesh and what you can actually do.
Practical conclusions
- Treat a mesh as a complement to an ISP, not a replacement, unless you live in or near a network with its own independent uplinks.
- For emergency communication, assume that local messaging is the realistic goal: resilience in practice covers what has and has not worked.
- If you want a first step that needs no hardware, see zero-hardware start or find a local mesh.
- Mesh networks also depend on power and on volunteers; neither is guaranteed. A mesh is a different set of dependencies, not an absence of them.