Estimating egress costs is the single most critical piece of financial due diligence you can do before picking a provider. Get it wrong, and you'll get a bill that makes your eyes water. The sticker price for compute or storage is meaningless if you're moving data out to the internet or to other regions.
Here’s my blunt, practical process.
**First, map your data flows.** You can't estimate what you haven't defined. For a typical web application, trace:
* User downloads (static assets, reports, uploaded files)
* API responses to clients
* Data shipped to another cloud or your own colo for backup/analytics
* Traffic between regions *if* the provider charges for cross-region transfer (some do, some don't)
* CDN egress (often a separate, lower rate)
**Second, get your numbers.** You need volume per month, ideally in GB. Use your current logs. If this is a new service, model it.
* `(Average response size in KB * requests per day * 30) / 1024 = Monthly GB estimate`
* Don't forget cron jobs, data warehouse extracts, etc.
**Third, run the calculations.** Don't trust the pricing page calculator alone. Build your own spreadsheet because tiered pricing is a pain. Here's a simplified example for a hypothetical 100TB/month egress.
```python
# AWS Example (us-east-1, to internet)
first_10tb = 10240 * 0.09 # $0.09/GB for first 10TB
next_40tb = 40960 * 0.085 # $0.085/GB for next 40TB
next_100tb = 51200 * 0.07 # $0.07/GB for next 100TB (we only have 50TB left)
total_aws = first_10tb + next_40tb + next_100tb
# GCP Example (us-east1, to internet)
# GCP has a flat rate after 10TB in some regions, but check carefully.
first_1tb = 1024 * 0.12 # $0.12/GB
next_9tb = 9216 * 0.08 # $0.08/GB for 1-10TB
remaining_90tb = 92160 * 0.06 # $0.06/GB for >10TB
total_gcp = first_1tb + next_9tb + remaining_90tb
print(f"AWS Estimate: ${total_aws:.2f}")
print(f"GCP Estimate: ${total_gcp:.2f}")
```
**Key gotchas to check for every provider:**
* **Intra-region vs. Inter-region vs. Internet egress:** Rates differ wildly.
* **Free tiers:** Often the first GB/month is free. Negligible for serious loads.
* **Commitment discounts:** If you sign a 1-3 year term, egress is often discounted 10-30%.
* **Direct Connect/AWS Private Link/Cloud Interconnect:** You pay for the dedicated port, but egress over it is usually *much* cheaper than internet egress. Calculate the breakeven point.
Finally, pressure-test your model. What if traffic doubles? What if you add a new data export feature? The margin for error is small when you're dealing with $0.05-$0.20 per GB at scale.
Build once, deploy everywhere