Where your cloud bill actually goes: a field guide to the seven usual suspects

In over a hundred cost audits, we have never once found the waste where the client expected it. The big line item everyone argues about is usually fine. The money leaks out in places nobody owns.

Below are the seven patterns we find most often, roughly in order of how much they cost. Percentages are typical ranges from our audits — your numbers will differ, but probably not by much.

1. Idle capacity nobody dares to touch

The largest single source of waste — often 20–40% of compute spend. Instances sized for a launch two years ago, "temporary" clusters that outlived three product pivots, and machines kept running because nobody is certain what they do. The fix is unglamorous: an inventory, an owner for every resource, and a default-off policy for anything unowned after 30 days.

2. Instances sized by anxiety

Teams provision for the worst day they can imagine, then never look again. In practice most services sit under 20% utilization. Rightsizing is a spreadsheet exercise, not an engineering project — the hard part is giving one person the authority to actually do it.

3. Storage that only ever grows

Snapshots without expiry, logs kept at the most expensive tier forever, orphaned volumes from deleted instances. Storage rarely dominates a bill, but it compounds: it is the only cost category that grows even when your business doesn't. Lifecycle policies fix 90% of it in an afternoon.

4. Data crossing boundaries it doesn't need to

Egress and cross-zone traffic are priced like fine print and treated the same way. A chatty service placed in the wrong zone can quietly add five figures a year. If your traffic costs exceed 10% of the bill, the architecture is telling you something.

5. Non-production running production hours

Staging and development environments that run nights, weekends, and holidays — serving no one. Scheduling them off outside working hours typically saves 60–70% of their cost, and modern IaC makes the schedule a ten-line change.

6. Paying list price out of caution

Commitment discounts (reserved instances, savings plans) routinely cut 30–50% off steady workloads — but teams avoid them for fear of lock-in. The fear is usually miscalibrated: your baseline load is far more predictable than you think. Commit to the floor, stay flexible on the peaks.

7. No one owns the bill

Every pattern above survives for the same reason: cost is everyone's problem and no one's job. The single highest-leverage change is assigning ownership — one person who reviews the bill monthly, with the authority to ask "what is this?" and turn things off when the answer is silence.