Cloud Costs: Compute, Storage & Refreshes

Everything in this course is rented, and rented things send a bill. The good news is that an Azure invoice, however large, comes down to a handful of drivers — and once you can name them, the eye-watering numbers start to make sense. This final module isn’t about budgeting spreadsheets; it’s about understanding why the cloud costs what it does, so you can ask useful questions and avoid the few mistakes that quietly burn money.

The two big buckets: compute and storage

Almost every Azure charge falls into one of two buckets. Storage is what you pay to keep data sitting there — your blobs, your data lake, your warehouse tables. It’s usually the cheaper bucket: holding files is inexpensive, and the cost grows slowly as you accumulate more. Compute is what you pay to do work — running a pipeline, firing up a Databricks cluster, answering a query, refreshing a report. Compute is usually the bigger and more volatile bucket, because it’s billed by how much processing power you use and for how long.

The single most useful idea here: storage is cheap, compute is expensive. Keeping a mountain of data costs surprisingly little; processing it costs real money. That’s why a forgotten file rarely hurts the budget, but a cluster left running all weekend absolutely does.

Pay-as-you-go: the meter is always running

The default way Azure charges is pay-as-you-go — you pay only for what you actually use, billed by the hour or even the second, like a taxi meter rather than a flat monthly fee. This is wonderful when you’re careful: switch things off and the meter stops, so you never pay for idle capacity. It’s brutal when you’re not, because the meter keeps ticking whether or not anyone’s benefiting. The classic horror story is a Databricks cluster (Module 4) left switched on over a holiday weekend — three days of taxi fare with nobody in the cab. Pay-as-you-go rewards attention and punishes neglect.

Storage is cheap; compute is expensive. The fastest way to waste money is leaving compute switched on while it’s doing nothing.

Pausing compute: the off switch that saves the most

Because compute is billed while it runs, the most powerful saving lever is simply pausing or stopping it when it’s idle. Many Azure services have a clear off switch: clusters can auto-shut-down after a few quiet minutes, and some warehouse engines can be paused so they cost almost nothing while no one’s querying — storage charges stay, but the expensive compute charge stops. Refreshes (Module 6) count as compute too, so refreshing a heavy report every fifteen minutes when once a day would do is a quiet money leak. None of this requires technical skill; it’s the cloud equivalent of switching off the lights when you leave the room, and it’s where most real savings come from.

Capacity: pre-paying for a fixed amount of power

The other model you’ll meet is capacity. Instead of paying per use, you reserve a fixed block of computing power for a flat, predictable price — you’ll hear it with Microsoft Fabric and Power BI especially. Think of it as the difference between hailing taxis all day (pay-as-you-go) and leasing a company car (capacity): the lease is a steady, known cost, and it’s cheaper if you’d otherwise be taking a lot of taxis. The catches are real, though. Capacity is a fixed size, so a busy day can hit the ceiling and slow everyone down, while a quiet week means you’re paying for power you didn’t use. Capacity buys predictability; pay-as-you-go buys flexibility, and which is cheaper depends entirely on how steadily you use it.

Spot it: cloud cost drivers

Read each situation and decide for yourself, then tap a card to flip it and check your answer.

Sort the cost concepts

Drag each item into the bucket it belongs to — or tap an item, then tap a bucket. Hit Check placement when you’re done.

How to use it

You won’t approve the invoice, but you can stop it surprising anyone. When a bill jumps, the first question is now obvious: “Is some compute left running?” — a cluster, a warehouse, an over-eager refresh. When a new report is proposed, ask “how often does this really need to refresh?” — every refresh is compute. If the team debates capacity versus pay-as-you-go, you can frame it as steady, predictable usage (lean to capacity) versus spiky, occasional usage (lean to pay-as-you-go). And you can carry the one rule that explains most of the bill: keeping data is cheap, working the data is not. That instinct — switch off what’s idle, refresh only as often as you need — is the most valuable habit in the entire course.