Every team facing a Broadcom renewal ends up comparing Proxmox vs OpenStack vs VMware, and most of the answers they find were written in 2021.
The received wisdom goes: VMware is the serious one, Proxmox is the cheap one for small shops,
OpenStack is the complicated one for people building clouds. That was a fair summary once. In 2026 it’s
misleading, because all three moved — and because the field is now considerably wider than three.
Racktrail runs OpenStack, Proxmox and Ceph in production, and we move estates between them. That’s the only reason we can compare them without it being a sales position — we’re paid either way. Here’s
what’s actually true now.
What changed in 2026?
Proxmox picked up the enterprise features people said it lacked.
Proxmox VE 9.0 shipped in August 2025 on Debian 13. 9.2 followed on 21 May 2026 with the feature that closes the most-cited gap: a dynamic load balancer. The cluster now automatically migrates HA-managed guests to reduce imbalance across nodes while respecting the HA rules you’ve set. If someone tells you Proxmox has no DRS equivalent, that objection expired in May.
The same release added WireGuard as a native SDN fabric protocol and native BGP support — the
difference between a cluster that sits awkwardly beside your network and one that integrates with it.
Proxmox SDN now does roughly what vSphere distributed switches do: one control plane, consistent
config across hosts.
Less glamorous but arguably more important: the surrounding ecosystem matured. Proxmox Backup
Server reached 4.x, and Veeam Backup & Replication 13.1 added Proxmox VE support including replication. Enterprise backup vendors don’t invest in platforms they think are hobby projects. That signal matters more than any feature.
OpenStack got easier to live with.
The historical complaint was never capability. It was operational burden — too many components,
upgrades every six months, a full-time job to stay current.
SLURP, the Skip Level Upgrade Release Process, addressed that directly. 2026.1 “Gazpacho”, released 1 April 2026, is a SLURP release: operators on 2025.1 Epoxy upgrade straight to it and skip everything between. You upgrade once a year rather than twice.
Gazpacho also shipped parallel live migrations in Nova — multiple memory transfer connections, materially faster workload movement — and live migration with vTPM, which OpenInfra describes as enabling secure movement of sensitive workloads. In practice that matters for any guest depending on a virtual TPM, Windows 11 among them. The parallel migration work was explicitly flagged as in high demand from operators seeking a VMware alternative, which tells you where the project’s attention went.
Worth noting for anyone thinking about sovereignty: 40% of contributions to Gazpacho came from European contributors, which the OpenInfra Foundation links directly to digital sovereignty initiatives driving adoption in the region.
And VMware got more expensive and less flexible.
Perpetual licences ended in January 2024. Subscription bundles replaced à-la-carte licensing, per-core pricing raised costs for many deployments, and customers report being pushed toward VCF editions containing far more than they need. The technical product didn’t get worse. The commercial relationship did.
Proxmox vs OpenStack vs VMware, side by side
| VMware vSphere | Proxmox VE | OpenStack | |
|---|---|---|---|
| Best at | Feature depth, ecosystem, “it just works” | VM consolidation, straightforward virtualisation | Self-service multitenancy, API-driven infrastructure |
| Operational weight | Moderate, welldocumented | Low — fewest moving parts | Highest, but SLURP cut the upgrade burden |
| Multi-tenancy | Good | Single-tenant or lightly segmented | Full — projects, quotas, RBAC |
| API maturity | Mature, proprietary | Functional REST, lighter tooling | Comprehensive, Terraform-native |
| Bare metal | Separate products | Not its job | Ironic, a genuine differentiator |
| Kubernetes underneath | Via Tanzu | Workable, less integrated | Strong — Cluster API, native |
| Cost model | Subscription bundles, percore | Free; optional support subscription | Free; cost is people or a partner |
| Lock-in risk | High and now demonstrated | Low — open, standard, portable | Low — open, standard, portable |
| Time to production | Most probably already there | Weeks for a VMwareshaped team | Months, more if self-taught |
When staying on VMware is the right answer
Worth saying first, because almost nobody selling an alternative will say it. VMware remains the most technically complete platform on this list. Stay if:
- Your renewal increase is tolerable and your estate is small enough that migration cost exceeds the saving.
- You depend on features with no clean equivalent — some NSX and vSAN designs, specific DR tooling, or third-party appliances shipped VMware-only.
- Your team has no capacity for a migration this year, and forcing one creates more risk than the invoice does.
- You can negotiate. Multi-year commitments move the number, and a costed migration plan is the leverage that makes that conversation work — whether or not you use it.
Hypervisors are becoming a commodity, and Broadcom’s pricing is the main reason people leave. If the
price is survivable for you, “stay and negotiate” is a legitimate outcome.
When Proxmox is right
- You’re consolidating VMs, not building a platform. The mapping from vSphere is close enough that a VMware-shaped team is productive in weeks.
- Your team is small. Fewer moving parts, fewer things to understand at 3am. Underrated.
- You want the shortest credible migration. For a straightforward estate, it’s the fastest route off VMware.
- Tenancy is simple. One organisation, a few segmented environments — not hundreds of isolated tenants with independent quotas.
- You want Ceph without a separate project. Integrated directly; VE 9.2 offers Ceph Tentacle 20.2 as a stable option alongside Ceph Squid 19.2.
When OpenStack is right
- You need genuine multi-tenancy. Projects, quotas, role-based access, isolation that holds when the tenants are paying customers.
- The API is the product. If your workflow is infrastructure-as-code, this fits far better.
- Kubernetes at scale sits underneath. Cluster API integration and a networking model built for it.
- Bare metal matters. Ironic is a real differentiator, and Gazpacho added autodetect deploy interfaces and trait-based port scheduling.
- Accelerators are on the roadmap. Cyborg now ships a driver configuration guide covering FPGA, GPU, NIC, QAT, SSD and PCI passthrough.
Frequently asked questions
Yes, for the workloads it suits. Proxmox VE 9.2 added dynamic load balancing and native BGP, and the
backup ecosystem matured — Veeam Backup & Replication 13.1 supports Proxmox VE including
replication. The remaining limitation isn’t reliability, it’s multi-tenancy: Proxmox is excellent virtualisation
but it isn’t a self-service cloud.
It used to be a stronger objection. SLURP releases cut upgrades to once a year rather than twice, which
removes most of the ongoing burden. It’s still the heaviest option here, and if nobody outside your
infrastructure team needs to provision their own resources, you’ll be paying that weight for capability
you never use.
Load-balanced and horizontally scaled tiers move with no user-visible interruption. Stateful singleinstance VMs need a brief stop — minutes, scheduled — to guarantee a consistent final sync. Anyone promising literally zero downtime for a stateful VM is either describing an HA pair you already have, or hasn’t done it.
Both are free to license. The real cost is people. Proxmox is cheaper to operate because there’s less of it. OpenStack costs more in expertise, whether you hire it or buy it from a partner. Compare total cost
against your VMware renewal over the same period, not against zero.
It depends on estate size, how much is undocumented, and how fast your change process allows
cutovers. Discovery usually takes longer than people expect and delivers more value than the cutover
itself. Get a plan costed before you commit to a date — it’s also the only real leverage you have in a
renewal negotiation.