Cloud cost is a design problem, not a billing one
When the cloud bill climbs, most teams reach for discounts and dashboards. The durable savings are almost always in the architecture.
When a cloud bill starts climbing, the first response is usually financial: buy reserved capacity, chase committed-use discounts, set up a cost dashboard, hold a meeting about it. All of that helps at the margin. None of it addresses why the bill is the size it is.
Cloud cost is mostly a consequence of design decisions made months or years earlier. You can negotiate the price of the wrong architecture down by twenty per cent. You can’t negotiate your way out of running three times more infrastructure than the workload needs.
Where the money actually goes
In the systems we’re asked to look at, runaway cost tends to come from a short list of causes:
- Idle capacity that’s always on. Environments provisioned for peak, running at peak prices around the clock, serving a load that peaks for two hours a day.
- Data moving when it doesn’t need to. Egress and cross-zone traffic are quiet, recurring taxes that rarely show up in a design review.
- The wrong tool doing the job. A managed service billed per request handling a workload that a small, steady worker would serve for a fraction of the cost — or the reverse.
- No back-pressure. Systems that retry aggressively and scale out under stress can turn a small incident into an expensive one in minutes.
Notice that none of these are billing problems. They’re architecture problems that happen to arrive as a bill.
Optimise the design, then the price
The order matters. Committing to three years of reserved capacity for infrastructure you shouldn’t be running just locks in the mistake. We work the other way around:
- Measure what the workload actually needs — not what was provisioned, and not what someone guessed at the start.
- Reshape the architecture so capacity follows demand, data stays put, and each job runs on the cheapest thing that can do it well.
- Then buy the discounts — on the smaller, right-sized footprint that’s left, where a commitment is safe to make.
Done in that order, the discounts apply to a bill that’s already been cut, and the savings compound instead of papering over the cause.
The number that matters
The goal isn’t the lowest possible bill — a system starved of resources fails in its own expensive ways. The goal is a bill that moves with the business: when usage doubles, cost rises sensibly; when it’s quiet, you’re not paying for peak. That relationship is designed in. It can’t be bought after the fact.