Bare Metal Cloud, Explained

A cloud made of whole machines

Order it off a list, have it about half an hour later, reinstall it whenever you like, stop at the end of the month. That is the part of a cloud most people actually use, and you get all of it. What you do not get is a hypervisor — and for nearly everyone who ends up on this page, that is the reason they came.

One tenant per machine No hypervisor Unmetered, unshared port

What you keep, and what you give up

A cloud built out of whole servers is better than a virtual machine at some things and worse at others. Which list matters is entirely down to what you are running, so here are both.

You keep

Ordering without speaking to anyone. Pick a line, pay, done.
Handover in about half an hour on anything already racked, and 4 to 24 hours on a machine we build for you.
Wiping it and reinstalling the operating system from the control panel, as often as you like, at no charge.
A console and a power switch that still work when the operating system has stopped answering.
Another machine whenever you need one, and no notice to give when you stop.
A choice of region, on one network.

You give up

Live migration. There is no hypervisor to slide you onto another host, so hardware maintenance gets scheduled with you instead of hidden from you.
Resizing in place. More memory or another disk is a scheduled visit to the machine, not a slider.
Snapshots of a running system. You back it up from inside, the way you would a server, because that is what it is.
Per-second billing. The month is the unit.
An autoscaling group. You add capacity by ordering a machine.

Leaning on any of those? You want a hypervisor — and you can be the one running it. Proxmox and VMware hosts are ordinary lines in the catalogue, with nobody underneath you sharing the hardware.

Why anyone takes that trade

None of these reasons is that it works out cheaper per gigabyte of memory. They are all about what stops being somebody else’s decision.

Nobody else is on it

No steal time, no neighbour saturating the disk queue at the wrong moment, nothing to raise a ticket about. The whole controller and the whole network card are yours, so performance on a Tuesday afternoon looks like performance at four on a Sunday morning.

You know what the processor is

Every line names the actual chip, how many cores it carries and how much memory it will take, so you can size against a part number instead of a vendor’s letter and digit. The catalogue lists all of them.

Per-core licensing stops hurting

Software billed by the core costs less on a few fast cores than on a wide virtual machine sized for the same throughput. Here you choose which processor you are licensing.

The traffic is not metered

Every server ships on an unmetered, unshared port, raisable to 100 Gbps for a fixed monthly amount. No transfer allowance to exceed and no per-gigabyte egress line — usually the thing that decides it for anyone moving video, backups or downloads. Ports and speeds.

The bill does not move

The same figure every month, agreed before you order. Nothing is charged by the request, the gigabyte or the hour, so there is no invoice to reconcile at the end of it and no spend alert to configure at the start.

It is your operating system

Root or Administrator from the first minute — your kernel, your modules, your firewall rules. Nothing is running underneath it that you did not put there.

Frequently asked questions

About the category. Delivery times, locations and support hours live on the pages that own them.

Is bare metal cloud really a cloud?

Half of it is. It is self-serve, it arrives in minutes without a sales call, it is billed by the month with nothing to sign, and you add or drop machines as you go. It is not virtualised, not billed by the second, and not resizable while it runs. The industry settled on “bare metal cloud” for that combination, and the name is fair as long as you know which half is which.

How is this different from an EC2 or Azure virtual machine?

The machine is physical and it is only yours. A hyperscaler instance is a slice of a host shared with strangers, sized in virtual CPUs that map onto threads of somebody else’s processor. Here you get the processor. In practice that means steady performance, no egress bill and a fixed monthly price, against no live migration, no resize slider and no per-second billing.

Can I run virtual machines on it?

Yes, and for a lot of customers that is the point. Install Proxmox, VMware or plain KVM and you are running the hypervisor yourself, with nobody underneath you sharing the hardware. Both are ordinary configurations in the catalogue rather than a special order.

How do I scale without an autoscaling group?

By ordering another machine, which takes about half an hour when the line is in stock. That is slower than an API call and far faster than a procurement cycle. A workload that genuinely bursts minute to minute wants a virtual machine, or a small elastic tier in front of bare metal carrying the steady load.

What does it cost to move data out?

Nothing. The port is unmetered and unshared, so there is no transfer allowance to exceed and no per-gigabyte egress charge — typically the largest single line on a hyperscaler invoice for anyone shifting media or backups. Copying data in costs nothing either, at any volume.