How to read a GPU-hour line on a Malaysian cloud invoice
The GPU-hour row is rarely one number. Here is how we unpack it when a Kuala Lumpur team brings last month’s bill to Jalan Thambi Dollah.
Notes
Pages we wish teams had before the first large bill
These notes come out of sittings: how to split a GPU-hour line, why token counters lie, what a campaign leaves running, and the four questions finance will ask after go-live.
Token counts and the bill are not the same story
A product team can watch tokens rise slowly while the invoice jumps. The gap is usually context length, retries, or a second model that never appeared on the dashboard the team actually looks at.
Idle inference capacity after a campaign week
Hari Raya and year-end campaigns leave replicas running. The invoice in the following month is often a campaign story, not a new product story.
Splitting a shared model endpoint across three product teams
Shared endpoints are common. Shared bills without a method are how three teams end up arguing about a number none of them fully caused.
What finance in Kuala Lumpur asks after an AI feature ships
The questions are predictable. Preparing the answers before the first full invoice arrives is cheaper than reconstructing a launch month from memory.