Public cloud is the right choice for workloads that scale up and down by the hour. For steady workloads that run all day, every day, the same capacity on dedicated servers usually costs a fraction of the price. This guide covers where the difference comes from, what you take on in exchange, and how to move without downtime.
Where the savings come from
- Compute. A cloud vCPU is usually one hardware thread, half a physical core. A dedicated server's cores are all yours, with no CPU credit system or noisy neighbours.
- Storage. Cloud block storage is billed per gigabyte per month, often with separate charges for provisioned IOPS. A dedicated server's disks are included in the monthly price.
- Egress. Data leaving a cloud provider's network is billed per gigabyte. Dedicated servers have a fixed port, often unmetered, so a busy month does not change the bill.
What you take on
Managed cloud services do real work. On dedicated servers you run that work yourself, or with tools:
- Operating system patching and monitoring.
- Backups, and testing that they restore.
- Database replication and failover, if you used a managed database.
- Load balancing, if you used a managed load balancer.
Most teams handle this with familiar open-source tools: Proxmox or Kubernetes for orchestration, PostgreSQL streaming replication, HAProxy, and Prometheus with Grafana for monitoring.
A migration plan that avoids downtime
- Inventory. List each service, its real CPU, memory and storage use at peak, and its monthly cloud cost including storage and egress.
- Size. Pick servers that cover peak use with headroom. Memory is usually the constraint; cores are cheap.
- Build in parallel. Order the servers, set them up with the same configuration management you use today, and connect them to the cloud environment over a VPN.
- Replicate data. Set up database replication from cloud to dedicated, and sync file storage.
- Shift traffic. Lower DNS TTLs a day ahead, then move traffic service by service. Keep the cloud side running until the new side has handled a full business cycle.
- Decommission. Turn off the cloud resources, and delete snapshots and storage volumes so they stop billing.
Keep what still fits the cloud
A hybrid setup is common: steady workloads on dedicated servers, bursty or experimental ones in the cloud. The aim is to pay cloud prices only where you get cloud value.