Why Companies Are Ditching the Cloud: The Rise of Cloud Repatriation

Original Article ↗ Hacker News Discussion ↗

Claims that “major companies are ditching the cloud” are met with skepticism, as public cloud providers continue to post strong growth while high‑profile repatriations like 37signals and GEICO look more like special cases than a broad shift. Commenters argue that runaway egress and storage costs, poorly managed “lift‑and‑shift” migrations, and a lack of cloud‑native practices are driving some workloads back on‑prem, especially bandwidth‑heavy or stable, predictable systems. Many see a nuanced future where organizations mix hyperscale cloud, colocation, and bare‑metal or Kubernetes-based platforms, choosing based on workload economics, control, and available expertise rather than ideology.

Reality of “cloud repatriation” as a trend

  • Many commenters see the article’s “companies are ditching the cloud” framing as overblown or clickbait.
  • Cloud provider earnings (AWS, Azure, GCP) are cited as evidence that overall cloud usage is still growing strongly.
  • Others argue multiple trends can coexist: late adopters still moving in, while some veteran users move specific workloads out.
  • Examples like mid-sized SaaS firms or large insurers are seen as anecdotes, not proof of a broad exodus.

Main cost drivers: when cloud gets expensive

  • Egress bandwidth and storage are repeatedly described as the biggest unexpected cost, especially for video/streaming, image hosting, and data-heavy services.
  • “Lift-and-shift” migrations (just moving VMs) without re-architecting often produce 2–10x higher costs vs on‑prem, with no reliability gain.
  • Poor cloud asset management (abandoned dev environments, idle servers, no cost visibility for engineers) inflates bills.
  • Some insist that with negotiated discounts, reserved instances, autoscaling/serverless, and right-sizing, cloud can be very cost‑competitive; others counter that even optimized setups can be far pricier for steady, bandwidth-heavy workloads.

Lift‑and‑shift vs cloud‑native

  • Consensus that simply replicating legacy environments in the cloud is a bad idea except as an emergency move.
  • “Cloud‑native” is described as adopting managed services, autoscaling, serverless, and re-architected apps to exploit scale-to-zero and consumption pricing.
  • This is seen as both the path to cloud cost efficiency and a major source of vendor lock‑in.

Alternatives: bare metal, colo, hybrid, and “open clouds”

  • Many argue there is a large middle ground between hyperscalers and DIY datacenters: bare‑metal hosting, colocation, Hetzner/OVH‑style providers, or OpenStack/Kubernetes‑based private clouds.
  • Hybrid models are popular: baseline or bandwidth-intensive workloads on dedicated/colo; bursty or experimental workloads in public cloud.
  • Some new projects aim to offer open-source cloud stacks on cheaper hardware providers.

Operational and organizational factors

  • Several note that big organizations carried old processes into the cloud (committees, click‑ops, no automation), negating many benefits.
  • Cloud is praised for rapid experimentation and low initial headcount; on‑prem is favored for long-lived, predictable workloads once expertise is in place.
  • Frustration with cloud vendor support, opaque pricing, and lock‑in drives some desire for more control, even when not strictly cheaper.