What the platform costs

you pay the full hourly rate for capacity you can give back at any second, which is the most expensive way to run anything steady.

On-demand pricing

you promise a year of spend in advance and pay perhaps forty percent less, and you owe it whether or not you use it.

Reserved capacity

the machine costs a fraction of the normal price and can be taken away from you with two minutes' notice.

Spot instance

a two-minute warning arrives and everything on that node has to drain, checkpoint and be gone before it is killed.

Spot interruption

every service asked for four cores because that is what the template said, and the fleet averages eight percent CPU.

Overprovisioning

you look at a fortnight of real usage and cut the requests in half, and nothing at all happens.

Rightsizing

the unattached disks, the reserved addresses and the load balancer for a service deleted last year are all still billing.

Zombie resource

objects older than ninety days move to cheap cold storage automatically, and reading one back now costs money and minutes.

Storage lifecycle policy

the compute was predictable and the surprise on the invoice was gigabytes leaving the provider, at a rate nobody had modelled.

Egress charge

spreading replicas across zones for resilience quietly added a per-gigabyte charge to every internal call between them.

Cross-zone traffic cost

you tried to answer which team spends what and forty percent of the bill was in a bucket labelled untagged.

Resource tagging

each team is told what they spent last month, and nothing moves between budgets as a result.

Showback

the spend actually lands on your team's budget, and now somebody senior cares about the idle staging cluster.

Chargeback

a job that was accidentally left running was flagged on day two rather than found in the invoice five weeks later.

Cost anomaly detection

the bill went up thirty percent and the cost per order fell, which is the only version of that sentence worth reporting.

Unit economics

engineering, finance and product look at the same spend numbers weekly, and the engineer who can change it is in the room.

FinOps