Notesfoundations🌱 seedling

Cloud vs Self-Hosting

Also: Cloud Repatriation · Cloud Exit · Build vs Buy · Managed Services · On-Premises

Running infrastructure yourself trades a lower unit cost and more control for fixed capacity and owning the operations; renting it from a cloud provider trades a higher unit cost and less control for elasticity, speed and managed services. Load shape and operational skill decide which wins.

Why it exists

Every component is a build-or-buy decision. In infrastructure that usually means self-hosted servers and software, or cloud VMs and managed services.

How it works

  • Owned capacity is sized once for the expected peak. Steady load uses it well; a spike above it is unserved.
  • Cloud capacity follows demand (autoscaling, serverless), billed per unit-time at a premium.
  • Managed services remove operational work but reduce visibility: when one misbehaves you can't log into it.
  • Hybrid is common: steady base load owned, bursts and experiments in the cloud.

Trade-offs

Self-hosting Cloud
cheaper per unit at steady scale elastic; no hardware lead time
full control and visibility managed services, less ops work
you own hardware, upgrades, on-call higher unit price, vendor lock-in, provider outages

In practice

  • 37signals (2022–2025): left AWS for owned Dell servers (~$700k) and then moved ~6 PB of S3 files onto 18 PB of owned storage capacity. Cloud spend fell from a $3.2M/yr run rate to $1.3M in 2024 (S3 only) with the same ops team; projected savings over $10M in five years. Their caveats: steady load, existing data-centre space.

Pitfalls

  • Comparing list prices only: staff, hardware refresh, space, power and egress all count.
  • Choosing self-hosting with spiky or unknown demand, or the cloud for large, steady, well-understood load, out of habit.

Recall

These cards are also in your review deck, scheduled with spaced repetition.