How the World Is Wired: building Topologies
Why I built an interactive atlas of global infrastructure across 20 lenses — and what mapping submarine cables, chip fabs, and power grids taught me about how fragile (and concentrated) the modern world really is.
“In the cloud” is a marketing euphemism for “in a datacenter at the other end of a cable under the sea.” I kept running into that gap — between how borderless the internet feels and how concentrated it actually is — and I wanted to make it visible. So I built Topologies: an interactive atlas you can look at the world through, one infrastructure lens at a time.
There are 20 lenses in all — submarine cables, internet exchanges, BGP routing, cloud regions, DNS roots, certificate authorities, time infrastructure, datacenters, supercomputers, chip production and its bottlenecks, energy, rare metals, critical minerals, shipping routes, maritime chokepoints, airports, navigation, satellites, and the political map of nation-states. Switch between them and the same planet keeps revealing new small-N systems: 13 logical DNS roots, a handful of fabs that can make leading-edge chips, a few dozen cable landing hubs that nearly all Europe–Asia traffic funnels through.
The hardest part wasn't the map. It was being honest.
Anyone can scatter dots on a world map. The temptation is to make every number look authoritative. I didn't want that. So every node and every figure carries a confidence level — verified (from public records or operator data), estimated (derived or interpolated), or mocked (a structural placeholder that is explicitly not a real statistic). Where a real value is unknown, the app shows a dash, not an invented number.
The data comes from open sources — TeleGeography for cables, PeeringDB for exchanges and facilities, OurAirports, the Global Energy Monitor, USGS and the IEA for minerals, TOP500 for supercomputers, CelesTrak for satellites, root-servers.org, and more — each under its own licence, all credited on an in-app Licenses & Disclaimer page.
That honesty discipline paid off in unexpected ways. While auditing the data I found that LUMI, Europe's flagship supercomputer, was pinned in Oulu — about 90 km from where it actually sits, in Kajaani — and that it (and eleven other famous machines) had been imported twice, once curated and once from TOP500, because the importer never deduplicated. Refreshing the airports lens from OurAirports surfaced three airports still on the map years after they closed: Berlin-Tegel, Tripoli International, and the old Siem Reap. None of that is visible until you go looking — which is exactly the point of a tool like this.
Dependencies are the interesting part
A datacenter on its own is a dot. A datacenter that you can click to follow to the submarine cable that feeds it, the internet exchange it peers at, and the power plant it draws from — that's a story. Topologies models those cross-lens links explicitly, so you can trace a chain from one domain into another, or run a “what if this node fails?” simulation and watch the cascade ripple outward.
Routing is first-class too: pick two nodes and the app traces the shortest path with Dijkstra, plus a couple of alternatives via Yen's k-shortest-paths, and you can re-optimise by latency, bandwidth, cost, or raw hop count.
A cartographic instrument, not a dashboard
The visual language went through a full redesign. The first version was a generic cyan-on-indigo dashboard; it competed with the data instead of framing it. The current design treats the app as an instrument: quiet cool-graphite chrome, a single survey-brass accent that carries every interaction signal (active lens, primary action, selection, the traced path), and monospace for every number — coordinates, capacities, latencies — like an instrument readout. The map gets a ticked neatline frame and a faint graticule, so it reads like a chart. Dark by default, with a daylight theme.
How it's built
It's a static single-page app — React 19, TypeScript, Vite, Bun — with two renderers sharing one data model: MapLibre GL for the geographic view and d3-force for the force-directed topology view. Everything is static JSON loaded on demand, so switching lenses is instant. It ships to Cloudflare Pages, rebuilding and deploying on every push through a GitHub Actions pipeline (production comes from main; every branch gets its own preview URL).
Topologies is free to use in teaching, talks, and research. Pick a lens, click around, trace a path, and follow a dependency or two.