Cloud egress fees: how much they add, and strategies to reduce them

The short answer

Egress fees are what a cloud provider charges for data leaving its network, mostly responses sent to your users on the internet. On a small web stack it can become a significant and growing line item as traffic grows.

Here's what we found:

• In Cloud World Model's September 2026 pricing snapshot, tripling traffic from 100 to 300 RPS raised the modeled cost of one small web stack by about 48% on AWS, 54% on GCP and 12% on DigitalOcean. All of that increase came from modeled egress.
• On list prices at 600 RPS with 5 KB responses, internet data transfer comes to roughly $0.89/hr on AWS and $1.12/hr on GCP. On DigitalOcean it can be $0.00/hr, assuming the full 8,000 GiB pooled transfer allowance from two Droplets is available, both Droplets run the full month, and that outbound traffic draws on the pool.

You can reduce egress in five main ways:

1. Reduce unnecessary cross-zone and cross-region traffic.
2. Serve repeat content from a CDN or cache.
3. Send fewer, smaller bytes by compressing and trimming responses.
4. Watch NAT gateway processing charges.
5. Use the transfer allowances that come with your plan.

Then model egress before deploying or changing infrastructure, so it shows up in the estimate rather than on the invoice.

This post is about cost only. It makes no latency, performance or capacity claims.

How egress is billed, by provider

These are list prices for data transfer out to the internet, retrieved November 3–4, 2026.

Three billing differences matter more than the headline rates:

• Tiers apply per month. You only reach the lower per-GB tiers at high volume, so most small stacks pay the first-tier rate on nearly every byte.
• DigitalOcean bundles transfer with compute. Each Droplet adds its allowance to a team-wide pool. Two General Purpose 2 vCPU / 8 GiB Droplets contribute 8,000 GiB a month.
• Load balancers can add their own data charges. In our cost-only multicloud comparison, GCP's load balancer added $0.008/GiB of data processed, inbound and outbound, on top of internet egress.

How much egress grows with traffic

From the September 2026 pricing snapshot (modeled)

Cloud Pricing Trends, September 2026 modeled one load balancer, two app servers and a MySQL database at 100 and 300 RPS. Quoted from that post:

As that post puts it: "Tripling traffic from 100 to 300 RPS raises the modeled cost by about 48% on AWS, 54% on GCP and 12% on DigitalOcean. All of that increase is modeled egress." Fixed infrastructure stayed at about $0.45/hr on AWS, $0.39/hr on GCP and $0.27/hr on DigitalOcean.

What these modeled figures include. They aren't a direct calculation of internet fees at 100 and 300 RPS. From the same post:

• The simulator reports one hourly "egress and data transfer" cost. "It combines internet egress and same-region cross-zone transfer."
• "What it doesn't expose: the assumed response size, the gigabytes transferred and the per-GB rate for each provider."
• "On AWS the benchmark calibration models about 88% of offered requests as served, so it's about 88 and 263 RPS."
• No provider's included allowance was deducted, and DigitalOcean's included Droplet bandwidth wasn't credited. As the post says, "Real DigitalOcean egress at these volumes could be lower."

On list prices: 200 vs 600 RPS

The September simulation did not deduct included transfer allowances. The example below applies published allowances, so its DigitalOcean estimate can be $0 even though the simulation reports an egress charge.

Assumptions: steady traffic for all 730 hours of the month, an average response of 5,000 bytes, internet egress to North American users only (no cross-zone or load-balancer data charges). That works out to:

• 200 RPS: about 2,448 GiB a month.
• 600 RPS: about 7,343 GiB a month.

*Assumes the full 8,000 GiB pool is available (not used by other Droplets on the team), both Droplets run the full month, and this outbound traffic draws on the pool. Beyond the pool, Droplet transfer is $0.01/GiB.

These are list-price estimates, not bills. The AWS line is our own arithmetic on AWS list prices. CWM's AWS model gave $0.3117/hr and $0.8724/hr for the same two traffic levels in our multicloud comparison. The model doesn't subtract AWS's 100 GB included transfer, and its byte accounting differs slightly.

Five strategies to reduce egress

1. Reduce unnecessary cross-zone and cross-region traffic

Internet egress isn't the only transfer charge. On AWS, data crossing Availability Zones in the same region costs $0.01/GB in each direction, and traffic between regions costs more: from us-east-2, $0.01/GB to us-east-1 and $0.02/GB to most other regions. Start by finding traffic that crosses zones or regions without needing to, such as chatty service-to-service calls, replication you don't use, or requests routed to a distant zone when a local copy exists.

Placement decisions should account for availability requirements. Putting everything in one zone can lower transfer charges, but it also creates a shared failure point, so we don't recommend it by default. Weigh the transfer saved against the resilience you'd give up.

2. Put a CDN or cache in front of repeat content

Every response served from a cache is one your origin doesn't send, and your origin's egress is the line that grows with traffic. Static assets, images and cacheable API responses are the usual first candidates. CDNs have their own per-GB pricing, though, and we haven't verified those rates for this post. Compare the CDN rate against your origin rate before assuming you'll save money.

3. Compress and shrink responses

Egress scales directly with bytes. Under the same assumptions as the list-price table, halving the average response from 5,000 to 2,500 bytes at 600 RPS cuts monthly transfer from about 7,343 GiB to about 3,671 GiB. On list prices that takes:

• AWS from about $0.89/hr to about $0.44/hr.
• GCP from about $1.12/hr to about $0.57/hr.

Ways to get there:

• Turn on gzip or Brotli compression.
• Paginate large lists.
• Drop unused fields from API payloads.
• Serve modern image formats.

4. Watch NAT gateway charges

On AWS in US East (Ohio), a NAT gateway costs $0.045 per hour, plus $0.045 per GB of data it processes. Standard data transfer charges apply on top. Private instances that pull large updates, images or third-party API responses through a NAT can add a noticeable line. We haven't verified GCP Cloud NAT pricing for this post.

5. Use the transfer allowances you already pay for

AWS includes 100 GB of internet data transfer out each month, shared across services and regions. GCP Premium Tier includes the first 1 GiB to North America. DigitalOcean is the outlier: every Droplet plan contributes an allowance to a team-wide pool, and Droplet overage is $0.01/GiB.

In our 600 RPS example, the 8,000 GiB pool covers about 7,343 GiB of transfer, which is why DigitalOcean's line can stay at $0.00 where AWS and GCP grow with traffic. That holds only if the full pool is available, both Droplets run the whole month, and the traffic draws on the pool. Other Droplets on the same team share it.

FAQ

What are cloud egress fees? Charges for data leaving a provider's network, mainly to the internet, and in some cases between regions or zones. Inbound data transfer is at no charge on AWS, GCP and DigitalOcean.

Why does egress grow faster than the rest of the bill? VMs, databases and load-balancer base fees stay the same when traffic rises. Egress is billed per byte. In the September 2026 snapshot, fixed infrastructure didn't change when traffic tripled, and all of the modeled increase was egress.

Which strategies have the biggest impact? It depends on your architecture. Start by finding unnecessary cross-zone or cross-region traffic, then look at caching and smaller responses, which cut the bytes you're billed for. After that, check NAT traffic and use any transfer allowance your plan includes.

Does DigitalOcean charge for egress? For Droplets, only beyond the pooled allowance that comes with them; Droplet overage is $0.01/GiB. Other DigitalOcean products can have different transfer rates.

Are the September snapshot figures and the list-price figures comparable? Not directly. The snapshot comes from a simulation at 100 and 300 RPS with no allowances deducted, and its egress line combines internet egress and same-region cross-zone transfer. The list-price table covers internet egress only at 200 and 600 RPS, with 5 KB responses and each provider's included transfer.

Methodology and sources

• CWM September 2026 snapshot: Cloud Pricing Trends, September 2026, published October 8, 2026, fetched November 4, 2026. Its figures are quoted as published. In that post's words, the run used "simulator engine 1.2.15, seed 20240601".
• AWS: Amazon EC2 On-Demand pricing, Data Transfer section, and AWS's public price data for US East (Ohio), retrieved November 4, 2026.
  • Internet data transfer out: $0.09/GB first 10 TB, $0.085 next 40 TB, $0.07 next 100 TB, $0.05 above 150 TB.
  • 100 GB a month included.
  • Cross-AZ: $0.01/GB in each direction.
  • Inter-region from us-east-2: $0.01/GB to us-east-1, $0.02/GB to most other regions.
• AWS NAT gateway: Amazon VPC pricing, US East (Ohio), $0.045/hr and $0.045/GB processed, retrieved November 4, 2026.
• GCP: Network pricing, Premium Tier, us-central1 to North America, retrieved November 3, 2026. Cloud Load Balancing pricing, retrieved November 3, 2026.
• DigitalOcean: Droplet pricing and Bandwidth billing, retrieved November 3, 2026.
• List-price estimate assumptions:
  • 5,000 bytes per response, about 2,448 GiB a month at 200 RPS and about 7,343 GiB a month at 600 RPS.
  • North American clients.
  • 730 hours a month.
  • AWS GB rates applied to the same byte counts.
  • Pricing is on-demand with no discounts, and taxes are excluded.

Model your egress

Egress depends on your traffic, your architecture, where your clients are and how big your responses are. Cloud World Model lets you simulate your architecture keyless and see modeled egress next to compute, database and load-balancer cost, then change traffic levels, regions or layout and rerun before you commit to anything. CWM doesn't take response size as an input, so calculate response-size sensitivity separately with your provider's per-GB rates, as in the compression example above.

MCP quickstart →