This is an important mistake that even experienced technical folks make while switching from the on-premise world to the cloud. We are conditioned to think that the cloud world is all pay-what-you-use, and elastic, but that is not entirely true. There are numerous GCP services where you pay for resources that you allocate (even if you end up not using them). The three most important ones probably are as follows:

• Block storage (persistent disks and local SSDs)
• BigTable
• Cloud Spanner

BigTable and Cloud Spanner are resources where you provision nodes based on your data size, so at least in those cases you are unlikely to wildly overprovision. Block storage devices, the persistent disks, and local SSDs are quite likely to drop below your penny-pinching radar though. These disk abstractions can always be resized, so don't feel compelled to start by provisioning the biggest size you can find. Start small and add capacity as you need to.