Week 2: The Last Customer Costs More Than All the Others Combined

Borrowing the term from telecoms and rail, we look at why running a line 99% of the way is cheap and the final stretch to every actual door is where the budget dies

1 March 2027

“Last mile” is a telecoms and rail term before it was ever a metaphor, and the literal version is worth taking seriously, because it’s a cleaner demonstration of the thesis than software gives you. A software team’s last mile is expensive because of decisions they made. A fiber operator’s last mile is expensive because of geometry nobody chose: density falls as you move outward, and the network has to reach every point in the coverage area, including the ones with almost nobody on them.

Here’s the number that makes this concrete rather than vibes-based. Aerial fiber construction runs on the order of $40,000–$60,000 per mile before you’ve connected a single household to it. Spread that over a dense city block and it’s nothing per customer. Spread it over the last few rural properties on a rural loop and the same mile of cable might serve one household. California’s public utilities commission has literally paid $54,100 to connect a single unserved location under its broadband programs — compare that to a subsidized satellite alternative costing roughly $1,086 per location, a fifty-times difference, for exactly the same outcome: one more address that can get online. The 2021 federal BEAD broadband program goes as far as defining an “Extremely High Cost Per Location Threshold” in its rules — an explicit, written-down admission that past some point per-customer cost stops being a curve you can budget for and becomes a number you have to decide whether to accept.

This is where the domain complicates the thesis rather than just restating it. In software, the last mile is expensive because of debt — shortcuts taken earlier that compound. Nobody took a shortcut here. Nobody cut a corner to make rural fiber expensive; the expense was baked in by the map before anyone laid a single conduit. That’s a different species of last-mile cost: not the bill for corners cut, but the bill for a resource — density — that was never available to begin with. A team can avoid the software version by not taking the shortcuts. No fiber operator can avoid this version by planning better; the only tools available are subsidy, a cheaper technology for the hardest addresses, or accepting that some addresses stay unconnected. The lesson isn’t “finish better.” It’s that some last miles are structural, not behavioural — and a course that only ever blamed procrastination or corner-cutting would be wrong about this one.