The biggest practical barrier to leaving the cloud is not hardware procurement or engineering effort — it is Data Gravity: the phenomenon where large datasets become increasingly expensive and difficult to move, effectively anchoring your architecture to the provider that hosts them.
Hotel California Pricing
Cloud providers have deliberately asymmetric pricing for data movement:
- Ingress (uploading data into the cloud): Free. Always. Getting your data in is frictionless by design.
- Egress (downloading data out of the cloud): Expensive. AWS charges $0.05–$0.09 per GB depending on region and volume tier.
This is not accidental. It is a structural moat. The more data you accumulate, the more expensive your exit becomes.
The Exit Tax: A Concrete Example
Suppose you have 10 Petabytes of data in Amazon S3 — a realistic figure for a media company or data platform after several years of operation.
10,000,000 GB × $0.08/GB = $800,000
That is $800,000 in network transfer fees alone, before you have bought a single server or written a single line of migration code. This is the "exit tax" that cloud providers implicitly charge.
Strategies to Reduce Egress Costs
1. AWS Direct Connect / Azure ExpressRoute (Negotiated Rates)
A dedicated private network line between your colo facility and the cloud provider's network. Egress over Direct Connect is significantly cheaper than public internet egress, and for large migrations, the cost difference can be substantial. You negotiate the rate as part of your enterprise agreement.
2. Physical Data Transfer (Snowball / Data Box)
AWS Snowball and Azure Data Box are physical storage appliances — essentially ruggedized hard drive arrays — that AWS ships to you. You load your data onto them locally (no egress fees), then ship them back to AWS (or in this case, to your colo). AWS then loads the data onto your destination storage.
For very large datasets, this is dramatically cheaper than network transfer:
- Network transfer of 1 PB at $0.08/GB ≈ $80,000
- Snowball for 1 PB ≈ $2,500–$5,000 in device fees
3. Incremental Migration
Rather than a "big bang" migration, move data incrementally over time using reserved Direct Connect capacity, spreading the egress cost across months or quarters and staying within negotiated rate tiers.
Interviewers frequently ask: "How would you migrate 5 petabytes of data out of S3 cost-effectively?" Candidates who only mention network transfer fail. The strong answer immediately raises physical transfer devices (Snowball/Data Box), explains the per-GB economics, and then discusses the operational trade-off: physical transfer is cheaper but slower (days to weeks for device shipping), while Direct Connect is faster but requires upfront infrastructure investment. Framing it as a speed-vs-cost trade-off demonstrates senior-level thinking.