Writing
Notes on what AI inference actually costs. We maintain a price catalog for every model our customers run, so most of what follows is a by-product of having to get those numbers right.
·8 min read
The storage hiding in your token bill
AI pushed enterprise SSD prices up several-fold, but that is not your bill. The storage on yours is context kept warm and gigabytes held — where a cache read costs a tenth of base input and a cache nobody reads costs more than not caching at all.
Read·6 min read
The rate card you built is already wrong
There were at least 28 changes to text-model API prices between July and September 2026. Three separate ways that breaks an internal cost dashboard, and what to do instead.
Read