Skip to content

Blog

Egress fees: the line item your comparison is missing

What egress fees are, why vCPU grids miss them, and a Hikube invoice from the catalogue, line by line.

CPU/RAM grids look much the same from one provider to the next. The gap shows up on a line no grid carries: what it costs when data leaves the provider’s network. This article explains that line item and shows its effect on an invoice. The three-year comparison is elsewhere: /blog/tco-cloud-souverain-suisse-vs-aws-azure-le-calcul-complet-sur-36-mois. The range for your own estate is on /tools/tco-calculator.

What egress fees are

An egress fee bills each gigabyte that leaves the provider’s network: to the internet, to your offices, to another cloud, sometimes from one of the provider’s regions to another. Inbound traffic is nearly always free; it is the outbound direction that is metered. Two things carry similar names and are unrelated. Inter-zone transfer bills your data moving internally, including between two datacenters of the same provider. Return of data at contract end is the recovery of your estate when you leave: a contractual obligation, not a monthly line. Conflating the three gives a TCO that is wrong in both directions.

Why comparisons miss them

Because a comparison happens before production, and the outbound volume is only known afterwards. A grid lines up vCPUs, RAM and gigabytes of disk: three quantities you decide. Egress is not something you decide, it is something you incur. And it grows along paths nobody put in the table.

  • Backups written outside the provider, every night
  • Datasets pulled to a GPU machine, then pulled again on the next run
  • The APIs your partners call: their inbound traffic is your outbound traffic
  • Exports to an analytics tool hosted at another provider
  • A CDN cache that misses and re-fetches the object from origin
  • End-of-contract return, which is not an egress fee but is paid once if your contract bills it

A Hikube invoice, line by line

The configurator’s SME scenario: one s1.small instance, 100 GB of replicated volume and one public IPv4. The amounts below are not transcribed into this article, they are read from the published catalogue: they follow the prices shown on the pricing page on every deployment, and the date at the top of the page says when this text was reviewed.

  • Instance s1.small (1 vCPU, 2 GB RAM): CHF 24 per month
  • Replicated block volume, 100 GB at CHF 0.18 per GB: CHF 18 per month, replication across the three datacenters included
  • One public IPv4: CHF 4 per month
  • Inbound traffic: free
  • Outbound traffic: free. This is the line this article is about, and it is zero whatever the volume
  • Total: CHF 46 per month

What this invoice does not include

An honest total means saying what it leaves out. None of these lines is priced here: reading them in the catalogue beats guessing them in an article.

  • Windows Server, which is not enabled by default and licenses per vCPU at the published rate
  • The Kubernetes control plane, billed separately from the workers
  • GPUs, whose production prices are in the catalogue
  • Backup and the secrets vault, which are separate services: managed backup and the secrets vault
  • Any project work, contracted under a SOW and outside PAYG
  • And one line that does not exist: there is no high-availability surcharge. Replication across three datacenters sits in the volume price, so do not add it as a multi-AZ SKU

Which content for which question

Four pages, four questions. This one explains the line item and shows its effect on an invoice. The thirty-six-month comparison between a Swiss sovereign cloud and the hyperscalers is in the 36-month TCO calculation. The range for your own estate, with your quantities, is on the TCO calculator, which does not replace a quote. And the catalogue itself, SKU by SKU, is on the pricing page. For GPUs and Windows, an engineer prices it: a Hidora engineer.

Ready to run on 100% Swiss infrastructure?

14-day trial, no credit card. GPUs included.