What egress is and why it dominates a cloud bill
Egress is data leaving a provider's network for the internet. Inbound transfer is normally free, movement inside a single availability zone is usually free or nearly so, and everything that goes out to a user, a partner or another cloud is metered per gigabyte. The rate looks small — single-digit cents — and that is exactly why it escapes planning until the invoice arrives.
The arithmetic is simple and the sums are not. A service delivering 50 TB a month at list rates costs roughly $4,300 a month, or over $51,000 a year, for bytes that cost nothing to produce. Nothing else on a typical bill scales so directly with success: double the users and compute may not double, storage certainly will not, but egress does.
Two structural features shape the bill. Pricing is tiered and marginal — the first block of gigabytes is charged at the highest rate, and later blocks at lower ones, with each rate applying only to the volume inside its own band. And the discount is shallow: a step from nine cents to five cents across a two-hundred-fold increase in volume means large customers pay a similar order of magnitude per gigabyte as small ones. That is the opposite of how compute and storage behave, and it is why egress becomes a larger share of the bill as an application grows.
Prices change and vary by region and provider, so every rate on this page is an input rather than a constant. Fill the bands from your own price list, and the calculator does the rest.
How a tiered bill is assembled
Subtract the free allowance first. Providers commonly grant a modest monthly amount — the default here is 100 GB — and it comes off the top before any band is charged. On a small service it can mean the bill is zero; at any real volume it is a rounding error.
Fill the bands in order. The billable volume is poured into the first band until it is full, then the second, then the third, with the remainder at the final rate. Each band's cost is its gigabytes multiplied by its own rate, and the total is the sum. This is marginal pricing: crossing into a cheaper band does not retroactively reduce what you already paid for earlier gigabytes. Anyone who has read an income tax table will recognise the structure exactly.
Add request charges. Object stores bill per operation as well as per gigabyte, quoted per 1,000 or per 10,000 requests. For large objects this is negligible; for a service serving millions of small files it is not, and it is the charge most often left out of an estimate entirely.
Add CDN delivery separately. A CDN does not remove egress — it changes who charges for it. Traffic served from cache leaves the CDN's network at the CDN's rate, and only cache misses reach your origin. So the total becomes origin egress at your tiered rates plus CDN egress at the CDN's rate, and the comparison against all-origin delivery is a genuine comparison rather than an automatic saving.
Divide at the end for the blended rate. Total cost divided by total gigabytes delivered is the single number to track over time. It is the figure to quote in a comparison, to use when estimating the cost of a new feature, and to compare against a committed-use discount or a private interconnect. Watching the blended rate rather than the invoice separates a price change from a volume change.
Worked example: 50 TB a month, with and without a CDN
A media service delivers 50 TB a month. The account has a 100 GB free allowance and a four-band price table at $0.09, $0.085, $0.07 and $0.05 per GB, with bands of 10,000 GB, 40,000 GB and 100,000 GB.
- Find the billable volume. 50 TB = 50,000 GB, less the 100 GB allowance = 49,900 GB.
- Fill the first band. 10,000 GB × $0.09 = $900.00. 39,900 GB remain.
- Fill the second band. Its capacity is 40,000 GB and only 39,900 GB are left, so all of it lands here: 39,900 × $0.085 = $3,391.50. Nothing reaches the third or fourth band.
- Total. $900.00 + $3,391.50 = $4,291.50 a month, which is $51,498 a year.
- Blended rate. $4,291.50 ÷ 50,000 GB = $0.08583 per GB — barely below the first-band rate, because almost all the volume sits in the top two bands.
- Now offload half to a CDN at $0.06/GB. Origin drops to 25,000 GB, less the allowance = 24,900 GB billable: 10,000 × $0.09 = $900.00, then 14,900 × $0.085 = $1,266.50, giving $2,166.50 of origin charges. CDN delivery is 25,000 × $0.06 = $1,500.00.
- Compare. $2,166.50 + $1,500.00 = $3,666.50 against $4,291.50 all-origin, a saving of $625.00 a month.
- Check why the saving is modest. The 25,000 GB moved to the CDN were being billed at $0.085, and are now billed at $0.06, so the gain is 25,000 × ($0.085 − $0.06) = $625. The saving is exactly the rate difference multiplied by the volume moved — which is why a CDN priced above your marginal origin rate makes the bill larger, not smaller.
Step 8 is the whole decision in one line. Compare the CDN's rate against the marginal band your traffic currently sits in, not against the first-band rate, because the volume the CDN removes comes off the top.
Reading the result and finding the savings
The blended rate tells you where you sit on the curve. A blended rate close to the first-band price means most of your volume is in the expensive bands and a modest reduction moves real money. A blended rate near the bottom rate means you are already deep in the discounted bands, and further volume is comparatively cheap — at that point architecture, not tuning, is where the savings are.
Compression and caching act on volume, and act first. Serving compressed responses, right-sizing images and setting sensible cache headers all reduce gigabytes before any rate applies, and they reduce the most expensive gigabytes because the reduction comes off the top band. Halving delivered bytes saves more than half the marginal cost when the traffic straddles bands.
The largest egress bills are often architectural. Cross-region replication, chatty inter-zone traffic, and analytics that pull raw data out for processing elsewhere all generate egress that no cache can absorb. Moving the computation to the data rather than the data to the computation removes the charge entirely, and it is usually the only change that alters the number by an order of magnitude.
Model the exit before you need it. The cost of a full data extract at your current rate is a number worth knowing, because it is the practical measure of how expensive leaving a provider would be. Regulatory pressure has pushed providers to waive egress charges for customers actually terminating service, but that concession applies to leaving — not to running a workload that reads from one cloud and writes to another every day, which is billed normally and can quietly dominate a multi-cloud design.
How the blended rate moves with volume on a four-band table
| Monthly egress | Billable GB | Monthly cost | Blended $/GB | Annual cost |
|---|---|---|---|---|
| 1 TB | 900 | $81.00 | $0.08100 | $972 |
| 10 TB | 9,900 | $891.00 | $0.08910 | $10,692 |
| 50 TB | 49,900 | $4,291.50 | $0.08583 | $51,498 |
| 100 TB | 99,900 | $7,793.00 | $0.07793 | $93,516 |
| 200 TB | 199,900 | $13,795.00 | $0.06898 | $165,540 |
| 500 TB | 499,900 | $28,795.00 | $0.05759 | $345,540 |
| 1 PB | 999,900 | $53,795.00 | $0.05380 | $645,540 |
Volume rises a thousand-fold between the first and last rows while the blended rate falls by only about a third. That shallow curve is the defining feature of egress pricing.
What this model leaves out
- Inter-region and inter-zone transfer. Traffic between regions, and often between availability zones, is charged separately at its own rates and does not touch the internet egress bands at all.
- Per-destination pricing. Several providers price egress differently by destination continent, so a single blended rate hides real variation if your audience is spread widely.
- Committed-use and negotiated rates. Above a few hundred terabytes a month, list pricing is a starting point for a conversation rather than what you will actually pay.
- Private interconnect. A direct connection to the provider is billed on port hours plus a lower per-GB rate, which changes the shape of the calculation entirely once volume justifies it.
- Load balancer and NAT gateway data processing. These are charged per GB processed in addition to egress, and on some architectures they are the larger line.
- CDN request and feature charges. CDNs bill per request and for features such as TLS, functions at the edge and log delivery, none of which appear in a per-GB comparison.
- Cache-miss traffic. The CDN share modelled here is what reaches users; misses still pull from origin and are billed as origin egress, so a low hit ratio erodes the saving.
Where egress sits in a cloud cost model
Cloud spend divides into three shapes. Compute is elastic and can be tuned continuously with rightsizing and scheduling — the server vCPU sizing calculator and the cloud VM monthly cost calculator deal with that side. Storage accumulates and is managed with lifecycle policies and reduction ratios. Egress is different from both: it is driven by user behaviour rather than by anything you provision, so the levers are architectural rather than operational.
That is why the three effective responses are all structural. Put a cache in front of the origin so bytes are served once and delivered many times. Reduce the bytes themselves, through compression, adaptive bitrates and smaller images, which shrinks the most expensive band first. And keep data and computation in the same place, so that analytical workloads never pull raw data across a billing boundary.
Volume estimates for any of this start from content size and audience. For media, the video bitrate and file size calculator turns a bitrate and a duration into gigabytes per stream, which multiplied by concurrent viewers is your egress. For bulk transfers, the data transfer time calculator answers the companion question of how long a given volume takes to move, and the data compression ratio calculator converts a reduction ratio into the gigabytes you no longer pay for.
A final discipline: recalculate the blended rate every month rather than reading the invoice total. A bill that rises because volume grew is a success; a bill that rises because the blended rate moved is a pricing or routing change, and the two demand completely different responses.