I took all my projects off the cloud, saving thousands of dollars

Original Article ↗ Hacker News Discussion ↗

Cost and pricing comparisons

  • Many commenters agree AWS is often far more expensive than VPS/dedicated options (Hetzner, OVH, Linode/DO) once you need moderate CPU/RAM/disk, especially for RDS and large block storage.
  • Extremely tiny or highly intermittent workloads can be very cheap on cloud (free/near‑free tiers, Cloud Run, Lambda, tiny S3/ECR usage).
  • For large, cold storage (PB scale), some find Glacier‑class services cost‑competitive vs building massive storage systems; at TB scale, local NAS or rented servers win easily.
  • Several people highlight that “2×” cloud premium is common and acceptable; others report 5–10× or more for comparable capacity.

When cloud is a good fit

  • Widely cited use cases: rapid MVPs with startup credits, bursty/seasonal load, large multi-region services, LLM or GPU-heavy work, and regulated industries needing ready-made certifications and SLAs.
  • Cloud helps bypass slow internal procurement and CapEx constraints; OpEx and “self‑service servers” were a huge part of its original appeal.
  • Managed services (databases, Redis, CI/CD, backups, global distribution) can be cheaper than hiring/retaining infra specialists, especially for fast‑moving startups.

Arguments for self‑hosting / bare metal

  • Many report running sizeable SaaS, forums, or side projects on 1–3 dedicated servers or home hardware with Cloudflare/tunnels, at a fraction of cloud cost and with acceptable uptime.
  • For the majority of businesses that don’t need “five nines,” simple setups (one DB, one app server, maybe a hot spare) are seen as sufficient and much cheaper.
  • Some frame self‑hosting as ideological: resisting “enshittification” and corporate control, promoting independence and decentralization.

Operational complexity and risk

  • Pro‑cloud voices emphasize the hidden labor of self‑hosting: backups, restores, security patching, intrusion detection, audits, off‑site redundancy, and hardware failures.
  • Others counter that much of this work also exists on cloud VMs, and that modern tooling (Docker, Ansible, k3s, etc.) plus AI assistance lowers the barrier.
  • A recurring worry with cloud is surprise bills and account lockouts; with self‑hosting, the main “catastrophe mode” is getting hugged to death during traffic spikes.

Lock‑in, architecture, and semantics

  • Several note that many “leaving the cloud” stories were barely using cloud‑specific services (mostly EC2/RDS/Redis), so migration was straightforward.
  • There’s disagreement over what “the cloud” even means: some treat any remote VPS/dedi as cloud; others reserve the term for hyperscalers and their proprietary services.
  • Hybrid and multi‑provider strategies are popular in the thread: keep compute or data where it’s cheap, use cloud only where its unique features matter.

Tone and meta‑discussion

  • Multiple commenters find the article ranty, straw‑manny, and needlessly antagonistic toward “cloud people,” even if they broadly agree AWS is often a bad deal for small projects.
  • Others see it as a useful counterweight to the default “everything must be on AWS” mindset, but wish for more rigorous TCO comparisons and fewer culture‑war vibes.