OpenStack
Order the servers, choose Ubuntu or Debian, and they are online with root about half an hour later. What happens next is your call. We do not install OpenStack and we do not operate it, so this page sets out exactly what your cloud gets from the hardware, and the one thing it will not get, while that still changes what you order.
Thirty-three images are on the order form, among them Ubuntu through 26.04, Debian through 13, AlmaLinux 8 to 10, Windows Server through 2025, VMware vSphere Hypervisor 7 and 8, and Proxmox VE 9. OpenStack is not there. Nobody sells it as a checkbox for a good reason: your release, your services, your networking driver and your storage backend are decisions about your cloud, and picking them at build time would only mean picking them wrong.
The deployment tool most people reach for is Kolla-Ansible. Its support matrix lists three host systems: Debian 13, Rocky Linux 10 and Ubuntu 24.04. Two of them are on our order form. Rocky Linux 10 is not, so plan on Debian or Ubuntu and nothing about the build will surprise you.
One boundary, said once and clearly. Support here is 24 hours a day for the machine, the disks and the link. It does not extend to your cloud. If an agent stops or an upgrade breaks a service, that is your side of the line, and we would rather you knew that today than at two in the morning in six months.
OpenStack's own guide gives each node at least two network interfaces. Our servers have one. There is no private VLAN, no extra port and no cross-connect on the price list, at any price, so the second interface has to be created on the host and every kind of traffic shares the same wire. Kolla-Ansible asks for two interfaces, 8 GB of memory and 40 GB of disk per host; only the interfaces are a real question here, and the answer is that you make the second one yourself.
If your design depends on a storage network the instances cannot touch, this is the wrong hardware, and knowing that before you pay is the entire point of putting it in the second paragraph.
On the ProLiant blade plans a disk controller sits in front of the drives and presents one mirrored volume. Perfect for a controller node, perfect for Cinder on LVM, and wrong for Ceph, which wants whole raw devices and says so in its own documentation.
That is what the NVMe cartridge plans are for. The m510 and the m710x have no controller in the way, so each disk arrives raw and an OSD gets what it expects. They are small and inexpensive, which happens to be the right shape for a first storage cluster: three of them cost less than one blade.
No bundles and no packages. Each plan is configured on the order form and priced there, and the configurator is the only number that counts.
| The job | Plans | Why that one |
|---|---|---|
| Runs the control plane | D104, D105 | Memory from 32 GB to a terabyte, mirrored SSDs, and a console you can open from the panel. An everything-on-one-box setup wants this too. |
| Runs the instances | D105, D157, D158 | You run out of memory long before you run out of cores. Two sockets reach a terabyte, four sockets reach two, and no part of that costs a licence fee. |
| Holds the storage | m510, m710x | Raw NVMe with nothing in front of it, which is what Ceph is asking for. Buy three to begin. |
| Somewhere to learn | m710x | A single cartridge with 32 or 64 GB and a reinstall button. Break it as often as you like; it costs about what a big virtual server does. |
Not every plan is in every city. D104, D105, D157, m510 and m710x are in all five, meaning New York, Miami, San Francisco, Amsterdam and Bucharest. D158 is in New York, San Francisco and Bucharest.
Setting up OpenStack rearranges the host's networking, and sooner or later a step does it wrong and the server stops answering. The blade plans and the m510 and m710x have a console you open from the control panel, so that moment is an inconvenience. The small cartridges, m300, m700, m710 and m710p, can only be powered off and on, so there it means reinstalling. Reinstalls are free, unlimited and take roughly as long as the first build. Order a plan with a console if the machine is going to be experimented on, which it is.
The rest is the same deal as any server here. Broken hardware is replaced within four hours, fifteen minutes of network downtime in a row earns a credit, you are billed monthly with no contract and no notice period, and paying for three, six or twelve months at once takes 5%, 10% or 15% off. None of the plans above has a setup fee.
The current release is 2026.1 Gazpacho, out on 1 April 2026. It is a SLURP release, which means you can jump straight to it from 2025.1 Epoxy instead of stopping at 2025.2 Flamingo on the way. Both distributions we install are new enough for it, and the containerised deployments barely care what the host is running.
Do not learn from the installation guide alone. It says outright that it is a teaching document and not meant for production. Kolla-Ansible and OpenStack-Ansible are what people actually run, and both want the same three things from us: a clean system, root, and addresses that behave.
No. We deliver the server with the system you picked, hand you root and the console, and stop there. Plenty of customers pair us with an OpenStack consultancy and that works nicely: they build the cloud, we keep the hardware under it alive.
There is not, and there is no way to buy one. Every server has one public port carrying everything, so lock the API and database ports down to your own addresses in the host firewall and encrypt whatever crosses between nodes. If your architecture assumes a separate back-end network, it needs different hardware than ours.
From the same /24 as the server, delivered on the same port, so an instance can hold a public address itself or sit behind the host. Two extra are $6 a month, four $18, six $30 and eight $42 on the larger plans; the entry blade goes to four and the NVMe cartridges to two. Past that, NAT behind the server, which is how most clouds this size run anyway.
On the cartridge plans, yes, because their NVMe disks are handed over raw. On the blades the controller presents one volume instead of individual drives, and Ceph tells you not to build on that, so use blades for compute and cartridges for storage. Three cartridges make a sensible first cluster.
One is enough to learn on, and an all-in-one install is a genuine OpenStack that fails in all the ways a real one does. Three is the point where other people can rely on it: control plane by itself, two compute nodes next to it. Order one, prove the design, add the rest whenever. There is no minimum.
On a blade or an m510 or m710x you open the console from the panel and put it right yourself. On the m300, m700, m710 or m710p there is only power control, so the fix is a reinstall. Reinstalls cost nothing and are unlimited. This is the reason to pick a plan with a console when the machine is going to be tinkered with.
Whichever you install. 2026.1 Gazpacho is current, dated 1 April 2026, and being a SLURP release you can move to it from 2025.1 Epoxy in a single step. We pin you to nothing, and rebuilding a host from scratch is one click in the panel.
Of course. The hardware is rented to you, the cloud on top is yours, and what you do with it is your business. Windows guests still need your own licences, the same as anywhere else.
A server already in the rack goes live roughly 30 minutes after approval; one built to your specification takes 4 to 24 hours. Cancel in the control panel and it ends with the billing month. No contract, no notice, no early exit fee.
