AnalysisDevelopersSeptember 15, 2026

Sunk Cost calculates when a local LLM rig pays for itself

Sunk Cost takes a machine, a model, and daily token usage, then estimates how long the hardware takes to pay back versus renting the same model per token. Where local speed isn't measured, it's estimated as memory bandwidth ÷ bytes read per token and labeled as such.

1 source

More stories today

Open the live feed