TL;DR: Direct Connect is a private physical line into AWS: Data Transfer Out drops from ~$0.09/GB to $0.02–0.03/GB, plus you get predictable latency and a private path. The trade is a fixed port fee ($216/month at 1 Gbps, ~$1,620 at 10 Gbps) plus a telecom circuit — so it's transformative at 20+ TB/month of sustained hybrid traffic and a money-loser for light egress. The classic post-install failure: routing still sends traffic over the internet while the paid line idles.
The numbers
- 100 TB/month: internet $9,000 vs DX
$2,216 (1 Gbps port) → **$6,800/month saved** - Media-rendering example from the source workflow: 150 TB/month, $13,500 → $4,620 (10 Gbps) — $106k/year
- Counter-example: 5 TB/month = $450 internet vs ~$616 with port + circuit — DX loses; revisit at 20–30 TB/month
- Port fees are hourly-billed (no long-term lock-in); the telecom circuit has its own notice terms
Do this
-
Get the real egress number: Cost Explorer → group by usage type →
DataTransfer-Out-Bytes, trailing 3 months. Sustained multi-thousand-dollar egress to fixed destinations is the qualifying profile. -
Model break-even honestly: (GB × $0.09) vs (GB × $0.02 + port + circuit). Clearing payback within 3–6 months is the usual green light. Get the telecom cross-connect quote before deciding — distance to a DX location drives it.
-
Start smaller with Hosted Connections (50 Mbps–10 Gbps through APN partners) if a full dedicated port feels like overcommitment.
-
After cutover, verify the line is actually used: CloudWatch
ConnectionBitsPerSecondon your VIFs. Near-zero while internet egress persists = route tables/DNS still prefer the public path, and you're paying for both. -
Plan redundancy: a second DX connection for mission-critical paths, or at minimum a site-to-site VPN failover — a single line is a single point of failure.
Gotchas
- DX is a path, not a policy. BGP and route tables must prefer it; the most common outcome of a rushed install is an idle paid line (the quiz-question failure in the full workflow).
- Setup is real networking work: colo cross-connect, Virtual Interfaces, BGP peering. Most teams use a network integrator; once up, it's largely hands-off.
- Location friction is the deal-breaker — far from a DX facility, circuit extension costs can eat the math.
- Direct Connect Gateway shares one physical connection across many accounts/VPCs — don't buy a port per account.
- DX solves AWS ↔ your sites; region-to-region traffic inside AWS wants Transit Gateway Peering instead.
Skip this if
- Egress is modest (< ~10–20 TB/month) — stay on internet rates and revisit as volume grows.
- Your egress is scattered across global end-users — that's CloudFront's job, not a dedicated line's.
- The traffic is AWS-internal — VPC Endpoints and TGW peering address that layer.