Why the edge is disconnected
The default assumption in software is that the network is there. Every architecture diagram with an arrow to “cloud” makes it. And for the teams whose systems live in offices and data centers, it mostly holds.
At the edge, it does not. The numbers from 2025 are worth sitting with, because they describe a trend line, not an anomaly.
The year connectivity broke records
Access Now’s KeepItOn coalition documented 313 internet shutdowns across 52 countries in 2025. That is a record, and it is the third straight year the record has been broken. These are not brief blips: they are deliberate, extended disruptions that take entire regions offline, and they landed on top of 180+ documented network disruptions from other causes.
The economic weight is measurable. Blackouts cost the global economy $19.7 billion in 2025, and a Surfshark study put the number of people affected by shutdowns at 4.62 billion. That is not a rounding error in global GDP or a niche humanitarian statistic. It is more than half the planet experiencing, at least once, the thing our architectures assume never happens.
The cloud is not exempt
It is tempting to file all of this under “fragile infrastructure elsewhere.” Then October 2025 happened. A major cloud outage took down a long list of consumer and enterprise services at once, and Forbes reported estimates of roughly $75 million lost per hour for the businesses affected.
The lesson is not that any one provider is careless. It is that centralization concentrates failure. When the hub goes down, every spoke goes down with it, no matter how healthy the devices at the edge are or how close they are to each other. Two machines a meter apart, both fully powered, both working, could not exchange a byte because a data center on another continent was having a bad day.
Where this bites hardest
The teams we talk to do not experience connectivity failure as a headline. They experience it as an operating condition. Warehouses with dead zones between racks. Field operations beyond cell coverage. Fleets of robots that need to coordinate in environments where nobody will ever install an access point. Events where 50,000 phones saturate every tower in range. Public-sector teams whose communication plans have to assume infrastructure is the first thing to fail in a disaster.
For all of them, “add retry logic and wait for the network” is not resilience. It is a queue in front of the same single point of failure.
Designing for the disconnected case
Offline-first is the inversion: treat local, peer-to-peer connectivity as the base layer and the internet as an enhancement, not a prerequisite. Devices connect directly over BLE and WiFi Direct, relay for each other, and treat an internet link, when one exists, as simply the longest link available.
That is the design premise of the Offline Protocol platform: a transport layer that fails over between transports instead of failing, identity that verifies without a certificate authority lookup, and service discovery that works with no DNS. It runs in production today on 350,000+ devices across 80+ countries, a network that grew through the exact conditions in the statistics above.
The edge is not waiting for better connectivity. The record-setting year says it is getting worse. Systems that operate there need a network that survives losing the internet, and that network can be built from the devices already on site.
If some part of your operation matches this description, talk to us.