Network transfer costs explained (egress, cross-region, cross-AZ)

Transfer costs are often the hardest part of a cloud bill to predict because they scale with usage and architecture. Use this guide as a framework to separate the transfer types and avoid double counting.

1) Split transfer into clear buckets

• Internet egress: data leaving your provider to end users.
• CDN bandwidth: delivery from CDN edge to users.
• Origin egress (cache fill): origin-to-CDN bytes.
• Cross-AZ: east-west traffic across zones.
• Cross-region: replication, multi-region reads, DR.

2) Estimate GB/month (fast methods)

• From metrics: bytes sent / transfer GB in billing exports is the most accurate starting point.
• From throughput: Mbps to GB/month using Units Converter.
• From requests: requests/day x avg response size to GB/month using API response transfer.

3) Price each bucket separately

• Internet egress: egress cost
• CDN bandwidth: CDN bandwidth cost
• Cross-region: cross-region transfer

Where to get the data (fastest sources)

• Billing exports: best for validating real GB and directionality (what is actually billed).
• CDN analytics: edge bandwidth, request volume, cache hit rate (needed to split origin vs edge).
• Load balancer metrics: data processed and request volume (good for separating heavy endpoints).
• Flow logs / VPC metrics: inter-zone and inter-region traffic (often invisible in app metrics).

4) Validate (avoid double counting)

• Separate CDN bandwidth from origin egress. Guide: origin egress vs CDN bandwidth.
• Separate cross-AZ from internet egress (they are different boundaries).
• Check incident windows: retries/timeouts can multiply transfer.

Worked estimate template (copy/paste)

• Internet egress GB/month = requests/month x avg response size (split heavy endpoints)
• CDN bandwidth GB/month = edge-to-user delivery (from CDN analytics)
• Origin egress GB/month = cache fill traffic (depends on cache hit rate)
• Cross-AZ GB/month = top east-west flows (LB, service mesh, database calls)
• Cross-region GB/month = replication + DR + multi-region reads

Common pitfalls

• Blending all transfer into one average number (you lose the optimization lever).
• Double-counting CDN bandwidth and origin egress (cache fill is the origin driver).
• Ignoring cross-AZ traffic in multi-zone services (steady background cost in microservices).
• Ignoring incident retries/timeouts (same payload transferred multiple times).

Optimization levers (match the lever to the bucket)

• Internet egress: compress payloads, reduce response size, put heavy endpoints behind CDN or signed downloads.
• Origin egress: improve cache hit rate, avoid cache-busting deploy patterns, separate long-tail endpoints.
• Cross-AZ: co-locate chatty services, reduce retries, batch calls, and avoid accidental cross-zone database access.
• Cross-region: reduce cross-region reads, use regional affinity, and validate replication topology.

Related tools

Sources

• AWS pricing (region/transfer pricing varies; validate per provider)
• GCP VPC network pricing

Related guides

Related calculators

FAQ