Cloud Repatriation: The Hidden Costs of Public Cloud Architecture
Table of Contents
When 37signals reported in late 2024 that they had saved over $2 million by leaving AWS, a wave of CTOs ran the numbers and quietly came to the same conclusion: for predictable, steady-state workloads, the public cloud has gotten expensive in ways that are hard to defend on a finance call.
Two years later, repatriation is not a fringe move. It is a default consideration on any infrastructure review at a company past Series B with stable demand.
The Three Cost Drivers Behind Repatriation
The math has not really changed. The visibility of the math has.
For people who want to think better, not scroll more
Most people consume content. A few use it to gain clarity.
Get a curated set of ideas, insights, and breakdowns — that actually help you understand what’s going on.
No noise. No spam. Just signal.
One issue every Tuesday. No spam. Unsubscribe in one click.
Egress Fees
AWS charges roughly $0.09 per GB out to the internet beyond the first 100 GB. For a company moving 200 TB/month, that is over $18,000 monthly — for the privilege of letting your data leave. Hetzner charges zero. Cloudflare R2 charges zero. The egress tax is now the single most cited reason for moving.
GPU Markups
H100 instances on AWS list at roughly 3-4x the bare-metal cost amortized over depreciation. For AI workloads, this delta is often the entire margin of the business.
Predictability Premium
Cloud is priced for elasticity. If your workload is not elastic, you are paying for an option you never exercise.
When Repatriation Does Not Pay
Repatriation is a trap for:
- Pre-Series-A startups (your time is more valuable than the savings)
- Genuinely bursty workloads (the cloud was actually designed for these)
- Compliance-heavy industries without dedicated platform staff
- Teams without at least one engineer who has run a fleet before
The Modern Hybrid Stack
The teams getting this right do not pick a side. They run a hybrid:
| Workload | Where it runs | |----------|---------------| | Steady-state web tier | Hetzner / OVH / colo | | GPU training | Lambda Labs, Crusoe, dedicated | | Bursty traffic | AWS, Cloudflare Workers | | Object storage | Cloudflare R2, Backblaze B2 | | Edge | Cloudflare, Vercel |
Tooling like Pulumi, Crossplane, and the maturing Talos Linux distribution have made multi-environment Kubernetes nearly turnkey for teams of any size.
The Real Takeaway
Cloud repatriation is not about hating AWS. It is about matching workload shape to pricing shape. The cloud is exquisite for spiky, unpredictable demand and miserable for steady-state at scale. Most companies have both — the question is whether your bill reflects that.
Related Reading
- migrating from DigitalOcean to Hetzner — A real migration story with concrete cost numbers.
- why hyperscalers are outspending megaprojects — Capex flows that explain why cloud margins are squeezed.
- DevOps and SRE for small teams in 2026 — The tooling stack that makes hybrid infra manageable.
💡 Key Takeaways
- When 37signals reported in late 2024 that they had saved over $2 million by leaving AWS, a wave of CTOs ran the numbers and quietly came to the same conclusion: for predictable, steady-state workloads, the public cloud has gotten expensive in ways that are hard to defend on a finance call.
- Two years later, repatriation is not a fringe move.
- The math has not really changed.
Ask AI About This Topic
Get instant answers trained on this exact article.
Frequently Asked Questions
Nilesh Kasar
Community MemberAn active community contributor shaping discussions on Startups.
You Might Also Like
Enjoying this story?
Get more in your inbox
Join 12,000+ readers who get the best stories delivered daily.
Subscribe to The Stack Stories →Nilesh Kasar
Community MemberAn active community contributor shaping discussions on Startups.
The Stack Stories
One thoughtful read, every Tuesday.
Responses
Join the conversation
You need to log in to read or write responses.
No responses yet. Be the first to share your thoughts!