The Long Haul: Fiber and Subsea Systems
The map of AI compute is drawn in fiber, and most of the important lines were laid before anyone was thinking about AI. Subsea cables built for streaming and finance now define where models can be trained near their data and served near their users. Faraday's read: the interesting scarcity is not raw fiber, of which there is a great deal, but lit capacity on the right routes, and the permission and physical access to build new landings. Geography, not just silicon, will shape which regions win compute.
Knowledge check
Why can the location of subsea cable landing stations influence where AI compute is economically sited?
Show the answer
Because data must travel between where it is generated or consumed and where it is processed, and that travel runs over long-haul and subsea fiber. Regions with abundant, low-latency cable connectivity can move data to and from clusters cheaply and quickly, while poorly-connected regions face higher latency and cost. Landing-station geography therefore shapes where siting compute makes economic sense.