Every server Unmetered traffic, DDoS protection and IPMI included. No setup fees, no contract. Unmetered traffic on every server See pricing

Private cloud

Build your private cloud on single-tenant hardware

Run your own virtualization or Kubernetes platform on hardware nobody else touches. Proxmox VE installs in one click, and free private VLANs connect your nodes in each location.

Top pick for private cloud

Pro 64

A balanced cluster node: 64 cores, 768 GB of RAM and four NVMe Gen5 drives for local storage or Ceph OSDs, with 2 × 25 Gbps ports. Ready in minutes at $1,349/mo.

Processor
AMD EPYC 9555P64 cores · 128 threads
Memory
768 GBDDR5-6000 ECC
Storage
4 × 3.84 TBNVMe Gen5
Network
2 × 25 GbpsUnmetered traffic

$1,349/mo

Deploy

Ready in minutes · 6 locations

Benefits

Dense hosts, up to 192 cores

The Pro 64, Max 96 and Max 192 give you 64 to 192 Zen 5 cores and 768 GB to 1.5 TB of DDR5 per node for high VM density.

Proxmox VE in one click

Deploy Proxmox VE straight from the OS library, or boot any hypervisor or Kubernetes distribution from a custom ISO over the KVM console.

Free private VLANs

Cluster, storage and VM traffic between your servers in the same location runs on private VLANs, and internal traffic is not billed.

Single-tenant by design

There is no shared hypervisor under yours: your VMs run on hardware dedicated to you, which keeps isolation and capacity planning simple.

Recommended servers

Three configurations that fit.

Prices per month in USD, excluding taxes, with unmetered traffic and DDoS protection included.

Pro

Pro 64

A balanced cluster node: 64 cores, 768 GB of RAM and four NVMe Gen5 drives for local storage or Ceph OSDs, with 2 × 25 Gbps ports. Ready in minutes at $1,349/mo.

  • AMD EPYC 9555P64 cores · 128 threads
  • 768 GBDDR5-6000 ECC
  • 4 × 3.84 TBNVMe Gen5
  • 2 × 25 Gbps portUnmetered traffic

Ready in minutes · 6 locations

$1,349/mo

Max

Max 96

96 cores and 1,152 GB of RAM on a single socket, so there is no cross-socket NUMA to plan around. Four 7.68 TB NVMe Gen5 drives give 30.72 TB raw per node.

  • AMD EPYC 9655P96 cores · 192 threads
  • 1,152 GBDDR5-6000 ECC
  • 4 × 7.68 TBNVMe Gen5
  • 2 × 25 Gbps portUnmetered traffic

Same-day delivery · 4 locations

$1,899/mo

MaxFlagship

Max 192

Two EPYC 9655 CPUs with 192 cores, 1.5 TB of RAM and 2 × 100 Gbps ports: the most consolidation per node, for large VM fleets or Kubernetes clusters built from a few very large workers.

  • 2 × AMD EPYC 9655192 cores · 384 threads
  • 1.5 TBDDR5-6000 ECC
  • 4 × 7.68 TBNVMe Gen5
  • 2 × 100 Gbps portUnmetered traffic

Same-day delivery · 3 locations

$2,999/mo

Architecture and sizing

How to run it well.

01

Designing a Proxmox VE cluster

Proxmox VE is in the one-click OS library, so a cluster can be running the same day. Plan for three nodes as a minimum: Corosync needs a majority to keep quorum, and Ceph needs three hosts to keep three replicas apart. With two nodes, add an external quorum device (QDevice) on a small third machine.

  • Networks: keep Corosync, Ceph and VM traffic on separate VLANs or at least separate subnets of the private network, and expose only management and public services on the public interface.
  • Storage: with four NVMe Gen5 drives per node, Ceph with 3× replication gives about a third of raw capacity. As an estimate, three Max 96 nodes have 92 TB raw and roughly 30 TB usable, before keeping 20 to 30% free for recovery.
  • Without Ceph: ZFS on each node with Proxmox storage replication works well for two or three nodes, if you can accept asynchronous copies.

Moving off VMware? Proxmox VE includes an import wizard for ESXi virtual machines. See our guide to Proxmox on a dedicated server.

02

Kubernetes on bare metal

Running Kubernetes directly on bare metal removes a virtualization layer and gives pods the full CPU, memory and NVMe bandwidth of the node. You also take over what a cloud provider would otherwise handle:

  • Load balancing: without a cloud load balancer, use MetalLB or kube-vip to announce service IPs. Additional IPv4 addresses are available on request.
  • Storage: local NVMe volumes for databases that replicate themselves, or Rook-Ceph or Longhorn for replicated block storage.
  • Networking: Cilium or Calico as the CNI, with pod-to-pod traffic on the private VLAN.
  • Control plane: three control-plane nodes for etcd quorum; on large hosts, run them as small VMs under Proxmox VE instead of dedicating three servers.

kubeadm, k3s and RKE2 install on Ubuntu or Debian. For Talos Linux or another immutable OS, boot its ISO through the KVM console. On dual-socket workers like the Max 192, enable the kubelet's Topology Manager so latency-sensitive pods stay on one NUMA node.

03

Networking and IP planning

Every server includes one IPv4 address and a /64 IPv6 block, and additional IPv4 addresses are available on request. Plan addressing before you build:

  • Public: give each node its own IPv4 address for management, and use the /64 for VMs and services where your users have IPv6.
  • Private: pick RFC 1918 ranges for the private VLAN, one subnet per role (cluster, storage, tenants), and document them early.
  • Tenant isolation: Proxmox VE SDN can build VXLAN or EVPN zones on top of the private network, giving each team or customer its own segment.
  • Edge: route VM traffic through a firewall VM such as OPNsense, or nftables on the host, and keep the Proxmox web UI and SSH off the public internet.

The private network covers servers in one location. To link clusters across locations, run WireGuard or IPsec over the public interfaces: traffic is unmetered, so replication between sites adds no transfer fees. Every IP stays behind always-on DDoS protection.

Your stack

Runs what you already use.

Full root access and one-click images for Ubuntu, Debian, AlmaLinux, Rocky Linux, Windows Server and Proxmox VE, or your own ISO through the KVM console.

  • Proxmox VE
  • Ceph
  • ZFS
  • Kubernetes
  • Talos Linux
  • OpenStack
  • MetalLB
  • Cilium
  • Longhorn
  • OPNsense

FAQ

Private cloud, answered.

How do I get Proxmox VE on a server?

Choose Proxmox VE from the one-click OS library when you order or reinstall, and the server is delivered with it installed. You can also boot the official ISO, or any other hypervisor installer, through the HTML5 KVM console. Either way you get full root access.

How many nodes do I need for a cluster?

Three is the practical minimum: Corosync needs a majority for quorum, and Ceph needs three hosts to keep three replicas apart. Two nodes can work with an external quorum device, but without shared storage you rely on asynchronous replication for failover.

Can servers in different locations share a private network?

No. Private VLANs connect your servers within the same location. Between locations, use an encrypted tunnel such as WireGuard over the public network; traffic is unmetered, so there is no transfer cost.

Can I get more IPv4 addresses for my VMs?

Yes. Each server includes one IPv4 address and a /64 IPv6 block, and additional IPv4 addresses are available on request. Open a support ticket with your requirements and we send you a quote.

Is Proxmox VE a realistic alternative to VMware?

For many workloads, yes. Proxmox VE offers clustering, live migration, high availability, Ceph integration and backups with Proxmox Backup Server, and it includes an import wizard for ESXi virtual machines. Test the features you depend on, such as backup integrations and guest drivers, before migrating production.

Still have a question?Tickets for customers, written quotes for projects.How to reach us

Start with Pro 64.

Our top pick for private cloud costs $1,349/mo with unmetered traffic and DDoS protection included. Or let an engineer size the whole setup with you.

  • No setup fees
  • Unmetered traffic
  • Cancel anytime