Offshore private cloudYour own region, down to the hypervisor.
Three to six whole machines reserved to one tenant, on a private L2 fabric with Ceph underneath and an API on top. We run KVM and pay no licence to anyone — which is why the same cluster costs a fraction of a VMware one, and why nobody can reprice it out from under you.
- PC 3 · XSNodes3 nodesRyzen 9 · 4.4 GHzCores36c / 72tZen 4, dedicatedMemory288 GBDDR4 ECCUsable storage2 TBCeph · 6 TB raw399$/mo
$3 214elsewhere - PC 3 · SNodes3 nodesEPYC 9124 · 3.7 GHzCores48c / 96tZen 4, dedicatedMemory576 GBDDR5 ECCUsable storage4 TBCeph · 12 TB raw749$/mo
$3 694elsewhere - PC 3 · MMost deployedNodes3 nodesEPYC 9224 · 3.7 GHzCores72c / 144tZen 4, dedicatedMemory1.15 TBDDR5 ECCUsable storage8 TBCeph · 24 TB raw1 449$/mo
$5 054elsewhere - PC 4 · LNodes4 nodesEPYC 9354 · 3.8 GHzCores128c / 256tZen 4, dedicatedMemory2 TBDDR5 ECCUsable storage21 TBCeph · 64 TB raw2 699$/mo
$8 014elsewhere - PC 5 · XLNodes5 nodesEPYC 9454 · 3.8 GHzCores240c / 480tZen 4, dedicatedMemory3.1 TBDDR5 ECCUsable storage53 TBCeph · 160 TB raw4 799$/mo
$12 554elsewhere - PC 6 · XXLNodes6 nodesEPYC 9554 · 3.75 GHzCores384c / 768tZen 4, dedicatedMemory4.5 TBDDR5 ECCUsable storage128 TBCeph · 384 TB raw8 499$/mo
$26 533elsewhere
Usable storage is what you can actually write: Ceph keeps every block on three separate chassis, so the raw figure beside it is three times larger. Nodes are whole machines from the bare-metal fleet — the same hardware, the same regions, reserved to you and wired together. Elsewhere is the cheapest published OVHcloud Hosted Private Cloud pack that the cluster on that line matches or exceeds on cores, memory and raw storage; it is the only operator that still prints per-month figures for a single-tenant cluster, and it is a VMware stack.
Included on every cluster
- Whole machines, reserved to one tenant
- Ceph storage, replicated three ways
- Live migration and automatic HA failover
- Private L2 VLANs and a routed VPC
- REST API, Terraform and OpenTofu
- Root on the hypervisors themselves
- No VMware, no Broadcom, no licence
- 10 Gbit cluster fabric, unmetered
- Any of the eight regions
- Monthly billing, no setup fee, cancel anytime
Where the difference actually comes from
It is not our hardware being cheaper. It is that a third of a conventional private cloud bill is software you never see.
The line item that is missing
Nothing on this cluster is licensed by anyone.
Proxmox VE, KVM, Ceph, Linux — open source top to bottom, with no per-core subscription, no entitlement file and no licence server to call home to. Broadcom withdrew public VMware pricing and moved the product to per-core terms; smaller tenants have reported increases of 350 to 450 per cent on renewal. A cluster that is not licensed cannot be repriced by a vendor you have never met.
Configure a cluster- No hypervisor licence Not bundled, not resold, not passed through at cost. There is no licence.
- No per-core subscription Adding cores adds hardware, and nothing else. The price list is the price.
- No third party in the contract One supplier, one invoice, settled on-chain. Nobody upstream to change terms.
Which cluster fits?
PC 3 · XS – S
A control plane, a staging estate, or thirty to sixty small guests that must not all die together.
PC 3 · M
A production platform: databases, application tiers and a Kubernetes control plane on one fabric.
PC 4 · L
Tenants of your own — a hosting business, a SaaS, a customer per VLAN, billed by you.
PC 5 – 6 · XL
A private region. Hundreds of guests, storage measured in hundreds of terabytes, no public cloud anywhere in the path.
Want a node that is not on this list — a GPU in the cluster, a JBOD shelf, a fourth region to replicate into? Specify the node in the builder
The platform
A cluster you could rebuild yourself.
Every component is one you can download, read and run without us. That is deliberate: an estate you could take somewhere else is an estate nobody — including us — can hold hostage.
- Nodes
- AMD EPYC 9004 · DDR5 ECC Whole machines from the bare-metal fleet, reserved to you. Three is the minimum: two nodes cannot hold a quorum, and a cluster that cannot vote cannot fail over.
- Hypervisor
- Proxmox VE 8 on KVM Open source, no per-core subscription, no licence server to phone home to. Take root on the hosts if you want it — most tenants never do.
- Storage
- Ceph, replicated three ways Every block written to three chassis. A drive, a node or a whole shelf can fail without the cluster noticing. Usable capacity is one third of raw, by design.
- Fabric
- Private L2 · 10 Gbit Your own VLANs and VXLAN overlays between the nodes, with a routed VPC above them. Nothing of yours shares a broadcast domain with anyone else.
- Control
- REST API · Terraform Provision, snapshot and destroy programmatically. An OpenTofu provider, an Ansible collection and cloud-init images ship with the cluster.
- Failover
- Automatic, under a minute A guest whose node dies restarts on a surviving node inside a minute. Live migration moves running guests between nodes with no interruption at all.
Installed on the cluster, or waiting in the image library
- Proxmox VE 8
- KVM / QEMU
- Ceph
- Linux
- Terraform
- OpenTofu
- Ansible
- cloud-init
- Kubernetes
- Docker
- Cilium
- WireGuard
- Prometheus
- Grafana
- OpenStack
Proxmox VE 8
KVM / QEMU
Ceph
Linux
Terraform
OpenTofu
Ansible
cloud-init
Kubernetes
Docker
Cilium
WireGuard
Prometheus
Grafana
OpenStackProxmox VE is what we install by default because it is the shortest route to a working cluster. Ask for plain libvirt, for OpenStack, or for bare Linux with Ceph and nothing above it, and that is what you get — with root on the hypervisors either way.
Racked in any of the eight
The price never changes with the region. Split a cluster across two, or replicate into a third.
- ReykjavíkIceland
- BucharestRomania
- SofiaBulgaria
- ChișinăuMoldova
- ZurichSwitzerland
- AmsterdamNetherlands
- Panama CityPanama
- SingaporeSingapore
Included at every size
Four things most of the trade charges for, and one of them twice.
-
Mitigation
1.2 Tbit/s DDoS protection
Always on at every edge, on the cheapest plan as much as the largest. No per-attack fee, no metered scrubbing and no protected-IP upsell.
-
Availability
99.99 % uptime SLA
Credits are worked out from the public incident record and applied to your balance automatically. There is no claim form, because there is nothing to claim.
-
Settlement
Crypto payment, no identity
Eight assets including Monero. No card, no bank detail, no document — not at signup, not at renewal, and not at any amount.
-
Moving in
Free migration
Whole servers, control panels, databases and mail, moved by an engineer at no charge. You test the copy before anything is cut over, and you keep the old host until you are happy.
Before you commit a cluster.
Why three nodes at the minimum?
Because two cannot vote. High availability depends on a surviving majority agreeing which node is dead; with two nodes a network fault leaves each convinced the other has failed, and both try to run the same guest. Three is the smallest cluster that can lose a node and still know it. If you genuinely want one machine, buy a dedicated server instead — it is cheaper and we will say so.
Is there a VMware licence anywhere in this?
No, and that is most of the reason for the price. We run KVM under Proxmox VE with Ceph underneath, all of it open source and none of it subscribed. Broadcom withdrew public VMware pricing and moved to per-core subscriptions; smaller tenants have reported increases of 350 to 450 per cent. Nothing here can be repriced by a vendor you have never met.
Do I have to use Proxmox?
No. Proxmox is what we install by default because it is the fastest way to a working cluster. Ask instead for plain libvirt, for OpenStack, or for bare Linux with Ceph and nothing on top, and that is what you get. Either way you hold root on the hypervisors — we are not a layer between you and your own hardware.
Why is usable storage a third of the raw figure?
Ceph writes every block to three separate chassis, so 24 TB of drives carries 8 TB of data. The listed usable figure is what you can actually put on the cluster; the raw figure is there so you can check our arithmetic. Erasure coding instead of replication is available where the workload suits it and lifts usable capacity to roughly two thirds.
Can I add nodes later?
One at a time, without downtime, billed pro rata from the day it joins. Ceph rebalances onto the new node in the background and live migration spreads the guests. Removing a node works the same way in reverse, provided the remaining ones can hold the data.
Who can see the virtual machines?
Nobody outside your cluster. The disks are Ceph volumes on hardware scheduled to you alone, on VLANs that terminate inside your own fabric. At-rest encryption with a key we do not hold is available on request, which reduces what could be surrendered to what we could not read.