To migrate from VMware to OpenStack, build the OpenStack side first (Nova for compute, Neutron for networking, usually Ceph for storage), then move VMs in waves. Each VM gets its VMDK disks converted to RAW or QCOW2, VirtIO drivers in place of VMware's, and a Neutron network on its old VLAN. Its source copy stays switched off but intact until the new one has proved itself. virt-v2v converts powered-off VMs one at a time; tools built on VMware's Changed Block Tracking copy running VMs and cut the outage to a final sync. Whether you should go at all depends less on OpenStack than on the size of your estate and on who will run the platform afterwards.
Where things stand in October 2026
- VMware is sold only as subscriptions, mostly in the VMware Cloud Foundation (VCF) and vSphere Foundation (VVF) bundles, counted per core with a 16-core minimum per CPU.
- vSphere 7 left general support on 2 October 2025, so many renewals this year are forced upgrades too.
- OpenStack has a counterpart for vCenter, vMotion, DRS, HA, vSAN and NSX; the DRS and HA ones are the least polished.
- Disk conversion is the easy part. Windows drivers, networks and staffing are where projects slip.
- With a few hosts and one admin, Proxmox VE or hosted servers are usually the saner exit.
Why are companies leaving VMware after Broadcom?
Broadcom completed its purchase of VMware in November 2023. The software did not get worse; the terms did, and they kept moving.
| When | What changed | What it means for vSphere users |
|---|---|---|
| 11 December 2023 | Perpetual licence sales end, and so do support renewals for perpetual products. Everything becomes a subscription inside the VCF and VVF bundles, priced per core. | The bill follows your core count, not the features you use. |
| 10 April 2025 | A free ESXi returns with 8.0 Update 3e: non-production only, no support, no vCenter, at most 8 vCPUs per VM. | Fine for a lab, not for production. |
| 15 July 2025 | Cloud providers not invited into the new partner programme get non-renewal notices and the white-label route ends on 31 October, as The Register reported. In January 2024 providers under 3,500 cores had already been cut. | Hosted VMware from a small provider gets hard to find. |
| 2 October 2025 | ESXi, vCenter and vSAN 7.x reach end of general support. | Staying means upgrading under the new terms. |
| 1 November 2025 | VMware at the hyperscalers moves to bring-your-own VCF subscriptions. | A public cloud's VMware service still needs the licence. |
| Early 2026 | The partner programme in Europe shrinks to a few hand-picked providers, last transactions by 31 March. CISPE files an EU competition complaint on 19 March, alleging cost increases above 1,000%. | European providers re-platform, and their customers move with them. |
| February to August 2026 | On 26 February the European Commission sends Broadcom a formal request for information over alleged tying and abusive pricing; on 3 August the General Court refuses Broadcom's request to suspend part of it. | Regulators are looking, but there is no ruling to plan around. |
| September 2026 | Broadcom executives promise a new vSphere Standard for environments of about 128 cores, with no price, metric or date. | Details are due at VMware Explore on Tour in Frankfurt on 13 and 14 October 2026. |
The per-core rule hurts small clusters most. A host with two 8-core CPUs is billed as 32 cores, because each CPU counts as at least 16. In March 2025 a distributor even told partners the minimum order would rise to 72 cores; Broadcom said it had never announced a price change, and later reports say the 16-core floor still applies. That episode explains the mood better than any price list: nobody can predict the terms of the renewal after next.
Should you wait for the new vSphere Standard? If your estate is small and otherwise happy, wait for the details from Frankfurt on 13 and 14 October. If you no longer trust the vendor, a cheaper edition will not fix that.
Can OpenStack replace VMware?
For compute, networking and block storage, yes, and replacing VMware with OpenStack works at scales VMware itself rarely reaches. The OpenInfra Foundation, part of the Linux Foundation since 2025, counted more than 55 million OpenStack cores in production in its 2025 survey, with Walmart, Workday and CERN each above a million. OpenStack is released under the Apache 2.0 licence, so there is no licence fee; you pay for hardware, people and, if you want one, a support contract.
A new release ships every six months. The OpenStack releases page lists 2026.1 Gazpacho (1 April 2026) and 2026.2 Hibiscus (30 September 2026) among the maintained series. Gazpacho added parallel live migrations, which move a VM's memory over several connections at once and which the foundation said VMware leavers had asked for, plus live migration of instances with a virtual TPM, as Windows 11 guests need. For a new build, take Gazpacho if your distribution supports it: it is a SLURP release, so it can upgrade straight to 2027.1 Indri (expected 24 March 2027) and your team upgrades once a year instead of twice. Put the storage on Ceph Tentacle (v20); Ceph's release list gives Squid an estimated end of life of 31 October 2026.
What OpenStack does not replace is the single vendor. With vSphere, one company answers for hypervisor, storage, network virtualisation and management. OpenStack is a set of services on top of KVM, a message queue, a database cluster and usually Ceph, and you own the seams. The licence line becomes a staffing line. Our rule of thumb: if you cannot name two people who will be on call for the control plane, buy a supported distribution (Canonical, Red Hat, Mirantis and VEXXHOST all sell one) or pick something smaller. The upstream installers, Kolla-Ansible and OpenStack-Ansible, are solid but assume someone can debug RabbitMQ at 3 a.m.
VMware vs OpenStack: what replaces vCenter, vMotion, DRS, HA, vSAN and NSX?
The VMware to OpenStack Migration Guide on openstack.org compares the two platforms side by side for compute, networking and storage. Here is how the main VMware features map onto OpenStack, with our view of how close each match is.
| VMware | OpenStack counterpart | How close, in practice |
|---|---|---|
| vCenter Server | Keystone, Nova and Placement, with Horizon or Skyline as dashboard | A different model: API first, dashboards are thin clients. Teams that already script vCenter adapt fastest. |
| vMotion | Nova live migration (shared storage, volume-backed or block), with auto-converge and post-copy | Close. On Ceph, draining a host feels much the same. Post-copy does not work with vGPUs. |
| DRS | Watcher | The weakest match. Run it in manual mode, where it only proposes action plans, and read each plan before letting it act. |
| vSphere HA | Masakari (instance, process and host monitoring) plus Nova evacuate | Works, but needs reliable host fencing. Pull a power cord in testing before you trust it. |
| vSAN | Ceph RBD behind Cinder, Glance and Nova | Mature. Plan for three or more storage nodes and a fast cluster network. |
| NSX | Neutron with OVN, Octavia for load balancing, VPNaaS | Good for routing and segmentation. Firewall rules must be rewritten as security groups. |
| vSwitch and DVS | Open vSwitch or OVN, provider networks | Easy. Keep VLAN IDs and VMs keep their addresses. |
| Templates | Glance images plus flavors | Arguably cleaner: image and size are separate. |
| Aria Operations | Prometheus, Grafana, Ceilometer | You assemble it, and get exactly the dashboards you build. |
For vSphere HA we list Masakari, because instance high availability is its documented job. And many clusters lean on DRS less than their owners think: with sensible flavors and anti-affinity server groups, the Nova scheduler places VMs well from day one.
How to migrate from VMware to OpenStack, step by step
Every migration from VMware to OpenStack runs in the same order: learn what you have, build where it will land, prove the conversion on boring machines, then schedule the hard ones. Plan in quarters, not weekends.
Step 1: inventory every VM
Export the list from vCenter, or let the vmware-migration-kit's discovery phase read it. Record each VM's OS, firmware (BIOS or UEFI, Secure Boot, vTPM), CPU, RAM, disks and their change rate, NICs with VLAN, and an owner who will sign off the result. Flag the awkward ones early: VMs with VMware snapshots (virt-v2v refuses them over SSH until you consolidate), shared disks, passthrough devices, appliances shipped only as VMware OVAs, software licensed to a MAC address, and anything whose vendor supports vSphere only. Then sort into waves: stateless Linux first, ordinary Windows next, databases and clusters last.
Step 2: build the landing zone
- Flavors that mirror your common VM sizes, so nobody argues about CPU and RAM during a cut-over.
- Provider networks on the same VLAN IDs as the old port groups, so VMs keep their addresses. Our overview of SDN options in OpenStack Neutron compares OVS, OVN and the alternatives.
- Ceph pools for Glance, Cinder and Nova, sized for three copies of every block (Ceph's default) plus room to recover a failed node. Our article on Ceph block and object storage explains RBD.
- Keystone projects and quotas per team, the self-service layer vCenter never really offered.
- Monitoring and backups, working before wave one.
Step 3: migrate a VM from VMware ESXi or vCenter with virt-v2v
virt-v2v, from the libguestfs project, is the open-source converter underneath most migration tools. It reads a powered-off guest from vCenter, from ESXi over SSH, through VMware's VDDK library, or from an OVA or VMX file, then installs VirtIO drivers, removes VMware Tools and fixes the bootloader. With -o openstack it writes each disk into a Cinder volume, and the virt-v2v OpenStack output manual says that mode must run on a conversion instance inside the target cloud, usually as root, with the project's openrc credentials loaded. With SSH enabled on the ESXi host, key login through ssh-agent and the VM shut down:
# as root, on the conversion instance inside the target OpenStack project
virt-v2v -i vmx -it ssh "ssh://root@esxi.example.com/vmfs/volumes/datastore1/guest/guest.vmx" -o openstack -oo server-id=v2v-hostUse your own host and paths, and the conversion instance's name or UUID as server-id. virt-v2v prints progress lines that start with the elapsed seconds in square brackets and ends with Finishing off; then boot a new server from the resulting volumes. For a whole vCenter the VMware input manual calls VDDK usually the fastest method, but VDDK is a proprietary Broadcom download.
Step 4: convert VMware disks to OpenStack images with qemu-img
For an exported OVA or a copied VMDK, qemu-img does the format change. Point it at the small descriptor .vmdk, with the large -flat.vmdk beside it. Choose RAW when Ceph stores your images, because Ceph's OpenStack documentation advises against QCOW2 for the VM disks it hosts and RAW lets RBD clone copy-on-write; QCOW2 suits file-based backends. After sourcing your openrc file:
# as root
qemu-img info guest.vmdk
qemu-img convert -p -f vmdk -O raw guest.vmdk guest.raw
openstack image create --disk-format raw --container-format bare --file guest.raw guest-from-vmwareqemu-img info should show file format: vmdk and a virtual size; the RAW file grows to that full size, so check free space. The upload ends with a table showing status as active. qemu-img installs no drivers, which is the next step.
Step 5: VirtIO drivers and the first boot
VirtIO devices play the role of VMware's PVSCSI and VMXNET3. Linux kernels include the drivers, but an older guest's initramfs may not load them; virt-v2v rebuilds it, and on the qemu-img route you regenerate it with the virtio modules before shutting down on VMware. Windows has no VirtIO drivers at all. virt-v2v injects them from /usr/share/virtio-win (the virtio-win package); converting by hand, install virtio-win inside Windows while it still runs on VMware, or the first boot ends at a blue screen reading INACCESSIBLE_BOOT_DEVICE. Then expect small surprises:
- Linux interface names change (
ens192becomes something likeens3orenp1s0), breaking network files written for the old name. - Windows sees the VirtIO adapter as a new NIC; the old static IP stays on the hidden VMware adapter and Windows complains the address is taken.
- VMware Tools can survive on Windows, since the virt-v2v manual admits automatic removal often fails. Uninstall it and add the QEMU guest agent.
- Windows may ask to be activated again after the hardware change.
Step 6: networks, addresses and security groups
Map each port group to a Neutron network. Provider networks sit on your existing VLANs, so a VM keeps its address if you create its port with that fixed IP. Overlay networks change addresses, and encapsulation pushes the MTU below 1500 unless the physical network carries jumbo frames. Neutron also drops traffic from any address not on the port, which quietly breaks keepalived VIPs until you add them as allowed address pairs. Rewrite NSX firewall rules as security groups, and move NSX load balancers to Octavia, covered in our article on OpenStack load balancing as a service.
Step 7: cut over in waves, with a way back
Run a pilot of ten or so dull Linux VMs end to end, including a restore test on the new side, and lower DNS TTLs a week before each wave. Quiet VMs can go cold: shut down, convert, boot. Large or busy ones need a tool built on Changed Block Tracking, which copies the disk while the VM runs and then syncs only changed blocks, so the outage is the last sync plus a reboot. Keep each source VM powered off, not deleted, until its owner signs off. And time the last wave against the renewal date: finish a month late and you pay for another year.
Which VMware to OpenStack migration tools are worth a look?
Most tools wrap virt-v2v, nbdkit (a network block device server that can read VMware disks) and virtio-win, adding discovery, orchestration and warm sync. The openstack.org guide lists nine. We would start with these:
- virt-v2v alone: open source, cold migration, one VM per run. Right for the pilot and for estates of a few dozen VMs.
- os-migrate's vmware-migration-kit: an Apache 2.0 Ansible collection (Ansible 2.16 or newer) maintained by Red Hat's OpenStack migration team. It discovers VMs in vCenter and migrates through nbdkit with CBT for near-zero downtime, through virt-v2v, or by direct conversion.
- VEXXHOST migratekit: an Apache 2.0 command-line tool doing a full copy, a changed-block sync, then cut-over. It needs VMware's VDDK 8.0.2, which you download with a Broadcom developer account, and vCenter rights for CBT and VMware snapshots.
- Platform9 vJailbreak: a web interface over virt-v2v, nbdkit and virtio-win. Often called open source, it is in fact under the Business Source License 1.1, which converts to MPL 2.0 four years after publication.
- Commercial: Coriolis, Mirantis Migration Service, ZConverter, FishOS MoveIt and CONTRABASS Legato, worth paying for when the window is tight.
What goes wrong in a VMware to OpenStack migration?
The conversion rarely fails. These do:
- Nobody owns operations. RabbitMQ, the database cluster and the OVN databases need patching, monitoring and regular upgrades, and a cloud handed over to no one degrades fast.
- A database that lived on an all-flash SAN feels slower on a Ceph pool sized for capacity. Benchmark the real pool with fio before moving it.
- Your VMware backup product works through vSphere's APIs. Confirm it supports OpenStack, or pick its replacement, before wave one.
- Aria Operations goes, Prometheus arrives, and for weeks nobody gets alerts. Our guide to Uptime Kuma, Prometheus and Grafana monitoring from a second location builds a watcher outside both platforms.
- Windows Server Datacenter and several database licences count physical host cores, so check how they treat the new hosts.
- "While we're at it, let's containerise it" doubles the timeline. Lift and shift first; modernise after the renewal date.
Is Proxmox or a hosted VPS the smarter move for a small estate?
Often, yes. OpenStack earns its overhead when many teams need self-service projects, quotas and APIs across dozens of hosts. Three hosts and one admin is a different problem, and there Proxmox VE has been closing the gap with vSphere. Version 8.2 (April 2024) added an ESXi import wizard that can even import live, 9.0 (August 2025) brought HA affinity rules, and 9.2 (21 May 2026) added a dynamic load balancer that moves HA-managed guests by real-time load, Proxmox's answer to DRS. Proxmox Datacenter Manager 1.0 (December 2025) puts several clusters in one view, with live migration between them. Our comparison of Proxmox and OpenStack as virtualization platforms goes deeper.
| Your situation | Best exit | Why |
|---|---|---|
| One to five hosts, one or two admins, ordinary Linux and Windows servers | Proxmox VE | ESXi import, HA, Ceph and load balancing in one web interface; vSphere admins find their way around in a day. |
| Several sites or clusters needing one console | Proxmox VE with Datacenter Manager | One view, live migration between clusters, central SDN. |
| Dozens of hosts, many teams, chargeback per project | OpenStack | Projects, quotas, APIs and multi-tenant networking are what it was built for. |
| A few internet-facing servers on one ESXi box | Hosted VPS or VDS | No hardware, hypervisor or cluster to patch; you look after the guest OS only. |
| A certified appliance, or vendor support that names vSphere | A small VMware footprint for that workload | Move everything around it and license fewer cores next time. |
Where RS Computers fits
RS Computers runs KVM virtual servers, VPS and VDS, in Amsterdam (Netherlands), Prishtina (Kosovo) and Dublin (Ireland). A virtual server will not replace a vSphere cluster, but it helps in three places during a VMware exit.
Tooling: the Ansible control node for the vmware-migration-kit, Prometheus and Grafana watching both platforms from outside, the runbook wiki and a WireGuard jump host all fit on a VPS Micro or VPS Mini. The virt-v2v conversion host is the exception, since it must run inside the target cloud.
Learning: before the hardware arrives, an all-in-one DevStack on a VDS Medium (8 vCPU, 16 GB RAM, 480 GB NVMe) lets the team practise the APIs, Horizon and Neutron on a cloud it may break. Without hardware virtualisation inside the server, set Nova's libvirt virt_type = qemu: slow emulation, fine for learning.
A simpler home: many small vSphere estates are a web server, a mail server, a VPN and an app on one ESXi host. Rebuilding each on a fresh Debian or Windows Server VPS and moving the data is usually cleaner than carrying a converted disk over. Every plan has NVMe storage, its own IPv4 and IPv6 address and free weekly backups; VPS plans get a 1 Gb/s port and VDS plans 10 Gb/s, and Windows Server is available on VPS Mini and every VDS. The vCPUs are shared, so a busy database belongs on a VDS with headroom. The VPS and VDS plans page lists the specifications and which of the three cities can take a new server today.
Unsure which plan suits a workload leaving VMware? Message us on Telegram or email info@rscomputers-ks.com with what it runs, and we will suggest a plan and quote the setup.
Frequently asked questions
Can OpenStack replace VMware?
Yes for compute, networking and storage: Nova live migration covers vMotion, Ceph covers vSAN, Neutron with OVN covers NSX networking, Masakari covers vSphere HA and Watcher covers DRS. It does not replace a single vendor answering for the whole stack, so you need a platform team or a supported OpenStack distribution.
What is the OpenStack equivalent of vMotion?
Nova live migration. It moves a running instance between hosts on shared storage such as Ceph, with volume-backed disks or by block migration, and offers auto-converge and post-copy for VMs whose memory changes quickly. OpenStack 2026.1 Gazpacho added parallel memory-transfer connections for faster large migrations.
How do I convert a VMDK to QCOW2 or RAW for OpenStack?
Run qemu-img convert -f vmdk -O qcow2 guest.vmdk guest.qcow2, or -O raw when Ceph stores your images, then upload it with openstack image create. qemu-img changes only the format, so install VirtIO drivers in the guest first, or use virt-v2v, which handles disk and drivers together.
Do Windows VMs need VirtIO drivers after moving from VMware to OpenStack?
Yes. Windows has no VirtIO drivers built in, and without the storage driver it stops at boot with INACCESSIBLE_BOOT_DEVICE. virt-v2v injects them from the virtio-win package; for a manual conversion, install virtio-win while the VM still runs on VMware.
Can VMware VMs be migrated to OpenStack with near-zero downtime?
Yes, with tools built on VMware's Changed Block Tracking, such as os-migrate's vmware-migration-kit or VEXXHOST migratekit. They copy the disk while the VM runs, sync changed blocks, and stop the VM only for the final sync and first boot. Plain virt-v2v is a cold migration.
Does VMware still sell perpetual licences?
No. Broadcom ended perpetual licence sales, and support renewals for them, on 11 December 2023. VMware is now sold as subscriptions, mostly in the VCF and VVF bundles, counted per core with a minimum of 16 cores per CPU.
Is OpenStack or Proxmox the better VMware replacement?
Proxmox VE for small and mid-sized estates run by a few admins, since it imports VMs straight from ESXi and includes HA, Ceph and a DRS-style load balancer. OpenStack for large, multi-team environments that need self-service projects, quotas and APIs and can staff a platform team.
Before the next renewal lands
Whichever way you lean, do three things this month. Pull the VM inventory with host core counts. Run virt-v2v on two or three dull VMs against a small test cloud, so the team sees drivers, networks and a first boot for itself. And put the renewal date on the project plan. After VMware Explore on Tour in Frankfurt you will know whether a new vSphere Standard changes the arithmetic; by then you should already know whether OpenStack, Proxmox or plain hosted servers fit what you run.