Worked scenarios
Named fact patterns posted line by line under both frameworks. Every figure is an illustrative example, not client data, shown to make the shape of the postings concrete.
The scenarios
- Prepaid asset
A $2M/yr reserved GPU commitment, posted line by line
The buyer's exact case: a $2M per year reserved GPU commitment over three years. At signing it is a prepaid asset, not a $6M expense. As capacity is drawn down ...
- Prepaid asset
A prepaid inference credit block, set up and unwound
A prepaid block of inference credits is set up as a prepaid asset at purchase and released to the P&L as tokens are consumed. When the inference serves a paid p...
- Split
Building an internal copilot: the capitalizable split
An internal copilot build splits across capitalize and expense. Preliminary evaluation is expensed, development-window engineering and compute are capitalized, ...
- Cost of revenue
Production inference routed to cost of revenue
Production inference feeding a paid product is cost of revenue, not operating expense. Routing it correctly sizes AI gross margin and unit economics. Figures be...