Stop Doing Cloud

Cloud vs. Self‑Hosting: Cost vs. Time

• Many argue cloud is more expensive per CPU/storage than a VPS or bare metal, but worth it because it removes worries about hardware failures, disks filling up, and low‑level ops.
• Others say cloud is “overpriced most of the time” and that dedicated hardware gives more predictable, better performance for much less money.
• Several solo founders explicitly accept higher monthly cloud bills to free time for product work rather than server setup, patching, and backups.

Startups, PMF, and Scale

• One camp: early‑stage startups should focus on product–market fit and iterate fast; managed services and serverless are ideal at this phase.
• Opposing camp: most startups never reach massive scale; a single server or a few VPSs are enough and over‑architecting for hyperscale is wasted effort.
• Some note that in leaner funding environments, infra cost can matter enough that cloud at list prices can make a business non‑viable.

Reliability, HA, and Downtime

• Critics of “one dedicated box” highlight availability: upgrades, reboots, failed deploys, or broken hosts can create unacceptable downtime for businesses where the tech is the product.
• Others respond that 2–3 “nines” from a simple LAMP/monolith setup is often sufficient, especially early on, and that many services do not truly need multi‑AZ, 5‑nines architectures.
• Suggested mitigations: two similar servers behind a load balancer or simple A/B failover; however, some say good, automated failover is non‑trivial.

Complexity: Kubernetes, Microservices, and Over‑Engineering

• Several stories describe startups that prematurely adopted Kubernetes and microservices and later regretted the complexity; a simple monolith on a single VM or managed container service would have been faster.
• Others counter that once you manage fleets of VMs, container orchestration and standardized tooling (K8s, Prometheus, etc.) becomes less painful than bespoke setups.
• Some emphasize that cloud does not require K8s; a monolith on Fargate/Cloud Run with RDS/Cloud SQL can stay simple.

Performance and Hardware Choices

• Some argue a modern single machine is extremely powerful and can handle vast traffic unless workloads involve heavy analytics, stream/batch processing, or GPUs.
• Counterexamples are given of systems processing tens of TBs of data per day with ephemeral clusters of thousands of machines—clearly beyond “one box”.
• Debate over vCPUs vs. dedicated cores: some blame shared vCPUs and network storage for poor real‑world performance despite “OK” specs on paper.

Security, Liability, and Compliance

• Pro‑cloud voices note that managed databases, private subnets, and automated backups give better default security postures and ease audits (SOC2/ISO).
• Others point out that providers usually limit liability to service cost; if risk matters, you need insurance regardless of hosting model.

Middle‑Ground and Pragmatic Strategies

• Several recommend hybrids: cheap VPS or bare metal for apps, managed or specialized services (hosted DB, email, auth) where they add clear value.
• Cloud credits are mentioned as a strong incentive to start in the cloud while avoiding deep lock‑in to expensive proprietary services.
• Some see the main value of cloud in unified APIs and automation tooling (Terraform, CDK) that simplify lifecycle management.

Critiques of the Article’s Approach

• Supporters find it “refreshing” and aligned with “don’t overcomplicate early.”
• Skeptics say it glosses over OS installation, firewalls, backups, monitoring, disaster recovery, and consistent, tested procedures—turning hardware failures into potentially week‑long outages.
• There’s concern that the article underestimates operational discipline needed to run self‑hosted infra “correctly,” especially for anything beyond hobby scale.