OpenStack bare metal

OpenStack dedicated servers.Your cloud, on our metal.

Yes, you can rent an OpenStack dedicated server here and run a private cloud on hardware nobody shares with you. We install Ubuntu 24.04 or Debian 13, you get root about 30 minutes after a racked machine is approved, and you deploy OpenStack yourself with Kolla-Ansible, OpenStack-Ansible or your own playbooks.

  • Single tenant
  • Unmetered port
  • Raw NVMe for Ceph
  • No contract
  • Five cities

This page lists what the hardware and the network will and will not do under your control plane: one public port per machine, no private VLAN between them, floating IPs from the same /24, and a console that still works when your bridge config does not. Read it before you order.

Who installs what

OpenStack on bare metal: we build the host, you build the cloud

OpenStack is not one of the images on our order form. It is a deployment, not an operating system: the release, the services, the network driver and the storage backend are decisions about your cloud, and we would get them wrong if we made them for you at install time.

The sequence is short. Pick a distribution your deployment tool supports, take root over SSH, then run Kolla-Ansible or OpenStack-Ansible against the machine. Kolla-Ansible's current support matrix names Debian 13, Rocky Linux 10 and Ubuntu 24.04. We install the first and the third. Rocky Linux 10 is not an image here, and Ubuntu 26.04 is on the order form but not yet on Kolla's list, so a Kolla deployment on our hardware is Debian 13 or Ubuntu 24.04.

Support covers the machine, the disks and the network. Our AI on chat answers in seconds and can check your account and your server, and a person takes over whenever you ask. Nobody here operates your OpenStack: a broken Neutron agent or a failed upgrade is yours to fix.

  1. YouInstances and tenantsProjects, flavors, images, quotas
  2. YouOpenStack servicesKeystone, Nova, Neutron, Glance, Cinder, Ceph
  3. YouDeployment toolKolla-Ansible, OpenStack-Ansible, your playbooks
  4. UsLinux and root accessUbuntu 24.04 or Debian 13, reinstall from the panel
  5. UsHardware, port, consoleDisks, power, network, failed parts replaced

Network

One port per machine, and what that decides

The OpenStack installation guide's example architecture gives the controller and every compute node at least two network interfaces, and Kolla-Ansible's quick start asks for two interfaces, 8 GB of memory and 40 GB of disk per host. Memory and disk are easy here. The interface is the constraint: every machine we sell has one public port and no private VLAN to your other machines. A separate port or VLAN comes only with a cross-connect you bring in from another network, and it does not link your nodes to each other.

So the second interface is made on the host in software, and management, API and tunnel traffic share the wire with instance traffic. If your design needs an isolated storage network, or a management plane instance traffic cannot reach, this is the wrong hardware for it, and it is better to learn that here than from a ticket after you order.

The port

A 1 Gbps port, unmetered and not shared, is included on every line in the role table below. Nothing counts traffic in either direction, so replication, image pulls and backups do not show up on the bill. 10 Gbps is $529 a month and 20 Gbps $1,629 on the lines that carry them.

Floating IPs

One IPv4 comes with the machine. Extra addresses come from the same /24 and arrive on the same port: two for $6 a month, four for $18, six for $30, eight for $42 on the Scalable and four-socket lines. The D104 stops at 4 and the NVMe cartridges at 2. No MAC registration to file.

IPv6

Assigned as individual addresses, not as a routed /64. There is no prefix to delegate to tenant networks, so instances take addresses from your overlay and reach IPv6 through the host. Plan the tenant side with that in mind.

Between nodes

Machines in one city sit on the same network, a few milliseconds apart, which is what a control plane needs. The Geneve or VXLAN tunnels cross the public port, so the host firewall is part of the design, not an afterthought.

Disks and Ceph

Where the disks go, and which machines suit Ceph

Ceph wants whole devices. A RAID controller that shows the operating system one logical volume is the setup every Ceph and ZFS guide tells you to avoid, so the controller you pick in the configurator matters more than the drive size.

The m510 and m710x cartridges carry no RAID card. Their NVMe drives arrive raw, so each OSD gets the device it expects. They are also small, which suits a first Ceph cluster: three cartridges cost less than one large blade, and you learn more from three failure domains than from one.

On the ProLiant blades the 6, 12 and 24 disk controllers are hardware RAID. That suits a controller node and Cinder on LVM. The configurator also offers a JBOD option on the D104 and D105 with up to 2 drives, and on the D158 with up to 4, so a blade can carry a small OSD next to its compute role. For a storage tier built for Ceph, the cartridges are still the simpler answer.

Ceph OSD

m510 / m710x

NVMe, no RAID card, out-of-band console. Start with three.

Controller, Cinder on LVM

RAID 1 pair

Any blade line, mirrored SSDs behind the hardware controller.

Private cloud hardware

Machines for each OpenStack role

OpenStack's own hardware requirements are modest. The installation guide's proof-of-concept figures are one processor, 4 GB of memory and 5 GB of disk for a controller, and one processor, 2 GB and 10 GB for a compute node; Kolla-Ansible's quick start, quoted above, asks for 8 GB and 40 GB per host. Those figures get a cloud running. What you size for is the instances above it, and each role spends its memory differently.

Controller

Memory goes to the API workers, the message queue and the database, so it scales with cores rather than with instances: Red Hat's platform guide offers 1.5 GB per core as a starting figure and 20 GB as a floor. Run three controllers if the APIs have to survive losing one.

Compute

Add up the memory of the flavors you intend to run, then leave a few gigabytes for the host: cores overcommit, memory does not. KVM needs the Intel VT or AMD-V extensions, which every line here has. Disk holds the images Nova caches and the ephemeral roots.

Ceph OSD

Budget about 4 GB of memory per OSD, BlueStore's default target. Ceph's sizing guidance puts total host RAM above twice the sum of those targets, to leave room for recovery. Three hosts is the floor for three copies in three failure domains.

All-in-one

One machine runs the lot, which is how most people start. The quick start figures apply, plus whatever the instances want. It also asks for two interfaces; here the second one is made in software, as the network section explains.

Nothing here is a bundle. Each line is built to the specification you choose and priced by the configurator, which is the only place a total is final. The links open that line's configurator.

RoleLinesWhat decides it
Controller, or a single all-in-one D104, D105 Memory from 32 GB to 1 TB on the D104 and 16 GB to 1 TB on the D105, a mirrored SSD pair and an out-of-band console. An all-in-one needs the memory more than the cores.
Compute D105, D157, D158 Instances run out of memory long before cores. The four-socket D157 takes 128 GB to 1 TB and the D158 128 GB to 2 TB. KVM has no per-socket licence, so four sockets cost nothing extra to virtualise.
Ceph OSD m510, m710x NVMe with no RAID controller in front of it. Memory 32 GB to 128 GB on the m510, 32 GB to 64 GB on the m710x. Both carry a console. Start with three.
A lab you can break m710x One cartridge, NVMe, a console and a reinstall button in the panel. DevStack on a machine nobody else uses, for about the price of a large virtual server.

Cities differ by line. The D104, D105, D157, m510 and m710x are in all five: New York, Miami, San Francisco, Amsterdam and Bucharest. The D158 is in New York, Bucharest and San Francisco. Amsterdam and Bucharest bill in euro: the same figures, with a euro sign. Keep the nodes of one cloud in one city; the tunnels are not meant to cross an ocean.

Recovery

Getting back in after a bad bridge

A deployment rewires the host's networking, and the step that gets it wrong takes the machine off the network with it. The blades and the m510 and m710x cartridges carry an out-of-band console you open from the control panel, so a broken bridge is a console session, not a support ticket. The small cartridges (m300, m700, m710, m710p) have power control only, so the same mistake there means a reinstall. Reinstalls are self-service, as many as you need, and take about as long as the first build.

Everything else works as on any dedicated machine here. Failed hardware is replaced within 24 hours on the included support tier and within 4 hours on Bronze, under the service level agreement (PDF), and network unavailability past 15 consecutive minutes earns a credit. Billing is monthly with no contract; paying 3, 6 or 12 months ahead takes 5%, 10% or 15% off.

A ready server ordered as it stands has no setup fee. Any other configuration is built for you and pays $29 once, shown on the order page before you pay.

Releases

Which OpenStack release to deploy

2026.2 Hibiscus is the newest release, published on 30 September 2026. It is not a SLURP release. 2026.1 Gazpacho, from 1 April 2026, is, so a cloud that upgrades once a year can go from 2025.1 Epoxy straight to Gazpacho and skip Flamingo. Nothing in our build pins you to a release.

The project's own installation guide says it is a learning document, not a production reference, and that is fair. What people run in production here is Kolla-Ansible or OpenStack-Ansible, and both want the same three things from us: a clean distribution, root, and addresses that behave.

Answers

OpenStack dedicated server questions

What is an OpenStack dedicated server?

A physical server rented to one customer that runs the OpenStack services itself: Nova, Neutron, Keystone and the rest are installed on the host, with no hypervisor of ours underneath. Here you get the machine with Ubuntu 24.04 or Debian 13 and root, and you deploy OpenStack on it with Kolla-Ansible or OpenStack-Ansible.

Do you install OpenStack for me, or is this self-managed OpenStack hosting?

It is self-managed. We build the machine, install the distribution you chose and hand over root and the console; everything above that is yours. If you would rather buy the deployment than build it, hire an OpenStack integrator and rent the hardware here. That combination works, and we are glad to be the hardware half of it.

Can I get a private network between my nodes?

No. Each node has one public port and no private network to the others, so its traffic for management, tunnels and storage crosses that port. A separate port or VLAN is given only for a cross-connect you bring in from another network, and it does not join your nodes to each other. Encrypt what matters, use the host firewall to keep the API and database endpoints closed to everything but your own addresses, and treat a design that assumes an isolated back plane as a design for other hardware.

How do floating IPs work, and how many can I have?

Extra IPv4 addresses come from the same /24 as the machine's primary address and arrive on the same port, so Neutron can give one to an instance directly or route a tenant behind the host. Two cost $6 a month, four $18, six $30 and eight $42 on the Scalable and four-socket lines; the D104 goes to 4 and the NVMe cartridges to 2. Beyond that, instances sit behind NAT on the host.

Can I run Ceph?

Yes. The m510 and m710x have no RAID controller, so each OSD gets the raw NVMe device Ceph expects; three of them make the smallest cluster worth running. The blade lines' 6, 12 and 24 disk controllers are hardware RAID and suit the controller and Cinder on LVM, though the D104, D105 and D158 also offer a small JBOD option.

What are the hardware requirements for an OpenStack node?

The installation guide's proof-of-concept minimums are one processor, 4 GB of memory and 5 GB of disk on a controller, and one processor, 2 GB and 10 GB on a compute node; Kolla-Ansible's quick start asks for 8 GB of memory and 40 GB of disk per host. Those are floors for a lab, so size for the workload instead: a controller by cores, about 1.5 GB of memory each by Red Hat's guidance; a compute node by the memory its flavors add up to, since cores overcommit and memory does not; a Ceph host by roughly 4 GB per OSD, BlueStore's default target. The memory each line takes is in the role table above.

How many machines do I need to start?

One, if you are learning: an all-in-one on a single machine is a real OpenStack and fails in all the instructive ways. Three is where it becomes something others can depend on, with the control plane on its own machine and two compute nodes. There is no minimum order and no bundle, so start with one and add the next when the first proves the design.

What happens if I lock myself out?

On the blades and on the m510 and m710x you open the out-of-band console from the control panel and fix it, as you would standing at the rack. On the m300, m700, m710 and m710p there is power control and no console, so recovery is a reinstall, which is self-service and takes about as long as the original build. Pick a line with a console for a machine you will experiment on.

Which OpenStack version should I use, and can I upgrade later?

Whichever you deploy. 2026.2 Hibiscus, released on 30 September 2026, is the newest; 2026.1 Gazpacho is the latest SLURP release, so a yearly upgrade can go from 2025.1 Epoxy straight to Gazpacho. Nothing in our build pins you to a version, and reinstalling the host is a button in the panel.

Can I sell instances from my OpenStack cloud?

Yes. It is your hardware to run, your hypervisor and your cloud. Windows guests need your own licences, as they would anywhere. If the question is what it costs against renting from a hyperscaler, the comparison is on our infrastructure as a service page.

How fast is a server delivered, and how do I cancel?

A racked machine is live about 30 minutes after the order is approved, and a build to order takes 4 to 24 hours. Billing is monthly with no contract and no minimum term. Cancel from the control panel at least five days before the billing month ends, and that month is the last one you pay for.

Start with one machine and prove the design

Order a controller or an m710x lab node, deploy, and add compute and Ceph nodes when the first one works. Billing is monthly with no contract, so a design that does not work out ends with the billing month.

Configure a D105