Cloud egress: Cross-AZ vs Cross-Region Data Transfer

Path matrix: cross-AZ vs cross-region data-transfer meters (educational)
Meter	Cross-AZ	Cross-region	Watch
Typical trigger	LB→instance, DB replica, chatty microservices	DR replication, multi-region reads	Architecture diagrams
Relative $/GB (educational)	Lower than inter-region (verify)	Higher tiers common	Provider + product
Same-AZ baseline	Often $0 between resources	N/A	Pin locality
User-facing?	Usually internal	Sometimes user geo routing	Don't blend with Internet egress
Control lever	Fewer cross-AZ hops; caching	Async replicate; regionalize	Product constraints

Path matrix: cross-AZ vs cross-region data-transfer meters (educational)

Meter

Cross-AZ

Cross-region

Watch

Typical trigger

LB→instance, DB replica, chatty microservices

DR replication, multi-region reads

Architecture diagrams

Relative $/GB (educational)

Lower than inter-region (verify)

Higher tiers common

Provider + product

Same-AZ baseline

Often $0 between resources

N/A

Pin locality

User-facing?

Usually internal

Sometimes user geo routing

Don't blend with Internet egress

Control lever

Fewer cross-AZ hops; caching

Async replicate; regionalize

Product constraints

Provider source footnotes

AWS EC2 on-demand data transfer — as-of 2026-09-09 Unknown until verified

GCP network pricing — as-of 2026-09-09 Unknown until verified

Azure bandwidth pricing — as-of 2026-09-09 Unknown until verified

Dated starting points only — Unknown until verified means no invented $/GB. PipeToll does not scrape private consoles or guarantee figures.

Path-class vocabulary: path template, Unknown until verified. Treat cross-AZ and cross-region as internal path templates distinct from Internet egress. PipeToll’s FinOps-lite habit is to name the path before comparing any $/GB cell.

Path templates to model

Load balancer → instances across AZs — request/response bytes may bill as cross-AZ depending on product.
Primary DB → standby in another AZ — replication chatter is often invisible in “user egress” dashboards.
Mesh sidecars east-west — chatty RPCs multiply GB even when payloads look small.
Async cross-region replica — DR/backup path; schedule and compression matter more than peak QPS.
Same-AZ pinned pairs — baseline that is often $0 between resources; use as control.

Common traps

Blending cross-AZ with Internet egress in one spreadsheet line hides the HA tax. “We only pay for downloads” ignores replica streams. Multi-cloud comparisons that quote only Internet $/GB miss inter-zone SKUs. Collapsing to one AZ is not a guaranteed savings path — it is an availability tradeoff. Mark unverified rates Unknown.

FinOps checklist

Inventory top cross-AZ talkers (LB, DB, mesh) by bytes — not only by request count.
Separate Internet, cross-AZ, and cross-region lines in the cost model.
Run sensitivity with calculator Mode A (?mode=gb&path=cross-az) using verified regional rates only.
For HA changes, pair cost deltas with explicit RPO/RTO acceptance — no savings guarantees.
Link findings to glossary: Cross-AZ and methodology.

FinOps-lite Unknown until verified

Worked example (educational): 8 TB cross-AZ chatter / month

A multi-AZ service ships ~8,000 GB of replica and mesh traffic inside a region. Using an example cross-AZ $/GB placeholder (treat as Unknown until verified), estimate = GB × rate for that path alone — do not blend with Internet egress. No savings guarantee if you collapse AZs; availability constraints dominate.

FAQ on this path

Is cross-AZ the same as Internet egress?

No. Cross-AZ is usually internal regional transfer between zones. Internet egress leaves the provider toward public clients. Keep meters separate in FinOps reviews.

Why can HA cost more than user downloads?

Chatty replicas, synchronous commits, and east-west microservice calls can move more GB than client responses. Measure both.

Does cross-region always cost more per GB?

Often inter-region tiers are higher, but verify per provider and product. Replication schedules and compression change totals.