TL;DR: Cross-region traffic over the public internet bills ~$0.09/GB; over Transit Gateway peering it rides AWS's private backbone at $0.02/GB — 77% less, plus flat fees of ~$36/month per VPC attachment and per peering link. The second win is topological: 5 VPCs across 2 regions need 10 VPC-peering connections for full mesh, or 2 TGWs and 1 peering link. But TGW is a scale tool — at 1–2 VPCs per region, plain VPC peering (same $0.02/GB, no attachment fees) is cheaper.
The numbers
- 50 TB/month US↔EU: internet $4,500 vs TGW
$1,292 (data + 4 attachments) → **$3,200/month saved** - Field examples: a 3-region SaaS moving 80 TB/month went $7,200 → ~$2,140 ($60k/year); a 4-region log-aggregation pipeline at 120 TB/month went $10,800 → ~$3,084 ($92k/year)
- The honest counter-example: a 2-VPC DR pair at 30 TB/month already on VPC peering ($600) would pay $852 on TGW — attachment fees made it worse; they correctly kept VPC peering
Do this
-
Find inter-region flows on the public path: Cost Explorer inter-region data-transfer usage types per region pair. Anything meaningful still crossing the internet is the immediate 77% fix.
-
Build hubs where scale justifies them: TGW per region → attach VPCs → create the peering attachment → update TGW route tables. Traffic never touches the internet.
-
Peer every pair that must talk. TGW peering is non-transitive: US-East↔EU-West plus EU-West↔AP-Southeast does not connect US-East to AP-Southeast. Book direct flights, not connections.
-
Re-check the math against plain VPC peering for small topologies — same per-GB rate, zero attachment fees. TGW earns its fees through mesh elimination, not the data rate.
-
Bonus once the hub exists: centralized inspection (one firewalled VPC on the TGW instead of appliances in every VPC), and the NAT consolidation pattern rides the same hub.
Gotchas
- Attachment fees are always-on — $36/month per VPC and per peering link, whether you move 100 GB or 100 TB. This is what kills low-volume setups.
- No transitive routing, no exceptions — plan a direct peering per communicating region pair.
- Same partition only: no peering into GovCloud or AWS China TGWs.
- 50 Gbps default ceiling per peering attachment (raisable); single-digit-ms added latency vs direct peering — irrelevant except for ultra-low-latency niches.
Skip this if
- You're single-region — nothing to peer (bookmark for when DR or data residency forces the issue; retrofitting routing is the painful path).
- 1–2 VPCs per region with modest traffic — VPC peering is cheaper and simpler.
- The flows are AWS↔on-prem — that's Direct Connect; many teams run both (DX for hybrid, TGW peering for cross-region).