A hyperscaler bill looks reasonable on day one. By the time a workload is steady, it is often the largest line in the infrastructure budget — and a surprising share of it is leaving the country entirely. Here is where the money actually goes, and how to check your own numbers.
The costs you do not see at signup
The sticker price is the compute. The real bill is the compute plus several quieter taxes that do not appear on the pricing page:
- Foreign-exchange exposure: billing in dollars means your local-currency cost moves every month with the exchange rate, regardless of your usage.
- Egress fees: you are charged to move your own data out. The more data you have, the more it costs to leave — which is the point.
- Inter-zone and inter-region transfer: traffic between availability zones and regions is often metered too, and adds up in distributed architectures.
- Support latency: a team eight time zones away turns a one-hour problem into a one-day problem. That delay has a cost even if it is not on an invoice.
A worked example
Consider a steady production workload — a database and application tier that runs 24×7. The illustrative shape of the comparison usually looks like this:
| Cost component | Always-on cloud instance | Dedicated bare metal |
|---|---|---|
| Compute | Per-hour, billed in USD | Fixed monthly, quoted locally |
| Data transfer out | Metered per GB | Bundled bandwidth |
| FX risk | Carried by you, every month | Removed if billed in-region |
| Scaling cost | Linear with usage | Add servers as needed, month-to-month |
| Typical result at steady scale | Highest total | Usually lower once egress + FX are included |
Technical detail
Egress is deliberately priced to discourage leaving, so the more you grow the harder it is to move — a form of lock-in that does not show up until you try to migrate. Bundled, flat-rate bandwidth on local infrastructure removes that lever. For steady workloads, a dedicated server at a fixed monthly price is usually cheaper than the equivalent always-on cloud instance once egress and FX are added to the comparison. The crossover point is lower than most teams expect.
How to audit your own bill
Before you change anything, get the real number. Pull your last three months of cloud invoices and split them into three buckets: compute and storage; data transfer (egress, inter-zone, inter-region); and everything else. Then identify which workloads are genuinely variable and which run flat all month. The flat, transfer-heavy workloads are where local infrastructure almost always wins; the variable ones may be worth keeping in the cloud.
What local infrastructure changes
Predictable monthly pricing, bandwidth bundled rather than metered, no penalty for moving your own data, and a team you can reach in your own working hours. The bill stops being a moving target, and more of it stays in the region.
Key takeaway
For early, bursty workloads, the cloud premium can be worth it. For steady ones, do the full sum — compute plus egress plus FX — and the local number often wins by a margin that surprises people. The honest way to settle it is to price your actual workload both ways.
Ready to talk specifics?
Get a Quote