Plans & sizing

_{Updated October 27, 2026} A plan is a fixed bundle of CPU, RAM, and base disk, billed at a single rate. Outbound bandwidth is unlimited on every plan. Picking the right plan is the main sizing decision when creating a VM. The full price list lives on Pricing; this page is the how-to-pick guide.

​

The four tiers

In simple terms:

• Standard — the platform may share the underlying physical core with other VMs. Performance is great most of the time but can have brief jitter when neighbors get busy. Best price/CPU. Standard is burstable: all your vCPUs are available whenever you need them, on top of a sustained level you can hold indefinitely. See CPU and bursting.
• Premium — half a physical thread is reserved for every vCPU, with burst to the full thread. Performance is predictable and steady. Higher price/CPU but no noisy-neighbor dips.
• Cost-Optimized: the disk is local NVMe on the one node the server runs on, which makes it fast and cheap. A repair of that node takes the server offline until the node is back, so it has no uptime commitment. CPU is shared and burstable, like Standard. Linux only, in US East.

If you can’t tell which fits, start with Standard. Move workloads to Premium when you can point to a specific symptom (latency p99 spikes, database query timeouts under load, real-time-stream stutter) that guaranteed CPU would solve. Pick Cost-Optimized when price matters more than uptime and your important data also lives somewhere else.

​

What is different on Cost-Optimized

• Linux only. Every Linux template runs on it; Windows does not.
• Resize stays in the tier. A Cost-Optimized server resizes to another Cost-Optimized plan. To move to another tier, create a new server.
• Backups cover the OS disk and are stored off the node, so they survive a node failure. Snapshots stay on the node with the disk.
• Volumes stay on the node. Extra storage added at create, or a volume created to attach to the server, attaches only to Cost-Optimized servers.
• No custom images. Save as image is not available, because the image would stay on one node.
• No uptime commitment. See the SLA.

​

Recommended starting plans by workload

Each row below names a starting plan — the place to begin, then resize up if you outgrow it. All prices are monthly subscription rates; see Pricing for yearly / 24-month locks.

​

Web and application servers

​

Databases

Databases are CPU-and-disk-sensitive — almost always pick Premium for consistent I/O. For databases above ~80 GB, attach a Volume for the data directory so disk size and VM size scale independently.

​

Background workers, batch, ML

​

CI/CD runners

​

Bastion / jump host

​

Windows workloads

Windows Server runs on its own Windows plan family (not on Standard or Premium): Windows plans are placed on licensed hosts, and the Windows Server Standard licence is a separate $15/mo line on every Windows VM. Prices below are the plan plus the licence. Windows has a higher RAM floor than Linux, so start at 4 GiB for anything beyond a single admin session.

​

Game servers

Latency-sensitive — Premium is the right pick for any production game server.

​

How to size in practice

Three rules that beat any sizing chart:

1. Start one size below what you think you need. Resize up is fast (one reboot, prorated billing). Right-size at month-end with real metrics.
2. Watch RAM, not CPU. Most “I need a bigger VM” moments are RAM-bound — the OS starts swapping and everything slows down. CPU saturation is rarer than memory pressure for typical web/app workloads.
3. Move data off the VM. If the disk is filling up, attach a Volume; if backups are the issue, Object Storage is cheaper per GB than VM disk. Don’t size up the VM to gain disk.

You can change plans on a running VM — see Resize. Disk can grow but cannot shrink, so be conservative there.

​

Plan vs storage vs network — what’s billed where

​

Related