Cloud Economics · 7 min
Where cloud costs actually come from
Most cloud bills are architecture and idle capacity, not the list price of a virtual machine.
Unit price negotiations feel like FinOps. They rarely move the bill. The large lines are usually idle compute, over-specified GPUs, cross-zone data, orphaned disks, and architectures that treat every environment as production-sized.
Cost is a lagging indicator of design. A service that cannot autoscale, a batch job that holds GPUs overnight, a logging pipeline that ships everything at full verbosity — each is an architecture decision with a monthly invoice.
The useful program starts with attribution. Who owns the spend, which product it supports, and which metrics would tell you the spend is doing work? Without that, rightsizing is a one-off cleanup.
Then change the delivery path: environment sizing, lifecycle of ephemeral stacks, and a review when a new GPU SKU or a new multi-region pattern is proposed. Economics belongs in architecture review, not only in finance.
CIFL does not publish invented savings percentages. The work is to make the bill intelligible and the next architecture cheaper to operate than the last — on purpose.