Most networks have a middle: a tower, an exchange, a cable landing where everything converges. A mesh has none. Every node is also a relay, and a message picks its way across the network one hop at a time, taking whatever path happens to be alive. This piece comes in two halves — first how that works today, then what we'd salvage if the centralised stack ever fell away.
A mesh is a network with no centre. Every node relays for its neighbours, so a message hops from node to node across whatever links are up — and reroutes the moment one goes down.
It’s deciding which hop. Three stances divide the field:
Before any clever routing, the radio sets the budget. Every mesh link picks two of three things — range, bandwidth, power — and pays for it in the third. There is no radio that gives you all three.
Buys range, pays in bandwidth.
Buys battery life, pays in range.
Buys throughput, pays in reach & power.
On top of the radio sits the routing logic — the part that decides where each hop goes. The three stances are genuinely different philosophies, not just tunings.
AODVkeeps no route it isn’t using. When it needs to reach someone new, it floods a request outward and waits for a reply. Cheap when idle; a beat of latency the first time you call.
OLSR and BATMAN keep the map warm, gossiping constantly so a route is always ready. BATMAN goes further: no single node holds the whole map — each only learns the best direction.
When there is no path at all, DTN gives up on routing in the moment. It stores the message, carries it, and forwards it when a link finally appears — store, carry, forward.
| Trait | Centralised | Star | Mesh |
|---|---|---|---|
| Single point of failure | The centre | Each hub | None inherent |
| Path diversity | One | Few | Many |
| Adding a node | Provision at centre | Attach to a hub | Just join |
| Partition tolerance | Poor | Limited | Degrades gracefully |
Here is the part prose can’t do. Below is a small mesh: a client at the lower left, a gateway at the upper right, and the shortest live path between them drawn in gold. Take relays offline and watch the route bend around the damage — until it can’t.
Tap a relay to take it offline. The gold line is the shortest surviving path from client to gateway. Cut the right relays and the path reroutes; cut enough and it fails outright.
Moving packets turns out to be the easy half. The web you’re reading this on trusts identity through certificate authorities — your browser ships with over a hundred root certificates and believes whatever they sign. It’s a clean model with a central dependency baked in. A mesh can’t assume those authorities exist.
The alternatives are older and stranger. A web of trust— Phil Zimmermann’s idea from 1992 — lets people sign each other’s keys until trust emerges from the graph. Stacks like Reticulum drop names entirely: a destination simply is a 128-bit hash of a public key, and everything is encrypted by default. No authority, no permission, no centre.
None of this is hypothetical. Mesh networks already carry real traffic at wildly different scales — a continental commons, a borough-sized ISP, a mesh you can fit in a backpack.
Guifi.net is the giant. Born in rural Catalonia in 2004 when villages got tired of waiting for an ISP, it now runs roughly 37,600 working nodes over some 73,000 kilometres of links — telecom infrastructure held as a commons.
NYC Mesh looks more like an ISP than a cloud. Rooftop antennas feed about seventy hubs, which feed a handful of data-centre supernodes that speak BGP to the wider internet. All volunteers; 450-odd installs in 2024 alone.
Meshtastic is the lower bound: cheap LoRa radios paired to a phone, flooding messages a few hops with no internet, no infrastructure, and no permission. Its cousin, Freifunk in Germany, is where BATMAN was born.
Now the second half — what a mesh could be. Everything from here on is a thought experiment, not a report. The trick is that none of the building blocks are invented: each one already runs today. The only speculation is in composing them into a replacement for a stack that vanished.
So: suppose the cell networks, the trans-Atlantic fibre, and the consumer ISPs all went dark tomorrow. A plausible rebuild wouldn’t start from a blank page — it would reach for the parts already lying around. You’d probably run LoRa for the long-haul backbone: kilobits per second, but kilometres per hop. Inside a neighbourhood, density would likely call for something BATMAN-shaped — self-configuring, no node needing the whole map. Across regions with no live path, traffic would fall back to delay-tolerant sneakernet: a drive carries the bundles the radio can’t. And identity would have to be cryptographic rather than certified — a web of trust, or Reticulum-style hashes of public keys, standing in for the certificate authorities that are no longer there.
That’s the honest part. What you’d rebuild isn’t a copy of the internet we have — it’s a slower, choppier, message-shaped one. Different, not restored. But, on the evidence of § II, entirely buildable.
The most serious recent screen treatment of post-cloud infrastructure. Its second season opens with a homemade local network stitched together after the global internet goes dark — very nearly this article’s thought experiment. Credited as inspiration only: no dialogue, frames, or character designs appear here.
Cory Doctorow’s Walkaway (2017) supplies the political imagination of bottom-up infrastructure; the cypherpunk lineage — May, Hughes, Gilmore — supplies the older conviction that trust can be built from cryptography rather than granted by an authority.
The current cyberpunk revival — the salvaged-hardware, post-everything mood drifting across TikTok edits and Are.na channels — is the aesthetic weather this piece was written in. It is described here in our own words. Nothing was scraped or embedded.