Three changes did the damage and they compound. Perpetual licences stopped being sold, the product line collapsed into bundles, and minimum core counts were attached to orders. Any one of them is an adjustment to absorb. Together they mean a renewal quote can arrive for a multiple of last year's cost without a single virtual machine having been added. This is what changed, how the arithmetic works, and what the alternatives actually cost.
None of it was concealed. Broadcom said publicly that it would concentrate on the largest VMware customers and simplify the portfolio, and both things happened much as described. What caught most organisations out was the size of the effect on a mid-sized estate, which is precisely where the core minimums bite hardest and the bundles carry the most that goes unused.
MassiveGRID runs Proxmox with Ceph in production, across a partner footprint of 85+ metros in 30+ countries, on HA clusters with automatic failover.
Proxmox support plans — from $99/node/month
HA Private Cloud · Colocation for hardware you already own
A note on figures. Licensing terms have changed repeatedly since the acquisition closed and vary by region, contract and negotiation. The mechanisms below are stable and worth understanding; the numbers in your quote are the only ones that describe your situation. Treat this as a guide to what to ask, not a price list.
What Actually Changed
Broadcom completed its acquisition of VMware in November 2023 and moved quickly.
Subscription only. Perpetual licences were withdrawn from sale in December 2023. Existing perpetual licences remain valid, but support for them ends, and a perpetual licence without security patches has a shelf life rather than a lifetime. This is the change that removes the option of simply not renewing.
Bundling. Dozens of individually purchasable products were consolidated into a small number of bundles, principally VMware Cloud Foundation and VMware vSphere Foundation, with narrower vSphere Standard and Essentials Plus editions reintroduced later for smaller customers. If you previously bought vSphere and nothing else, the replacement may include storage, automation and management components you neither asked for nor wanted, priced accordingly.
Core minimums. Licensing moved to per-core subscription with a floor. A 16-core-per-CPU minimum arrived first, so a socket with 8 physical cores was billed as 16. In April 2025 a 72-core minimum per order followed, which sets a floor on small purchases regardless of the hardware behind them.
Free ESXi withdrawn, then partly restored. The free hypervisor was discontinued in February 2024, which stranded a large population of lab, edge and small-business installations. A free ESXi 8 offering returned in 2025, but the interruption taught many of those users to look elsewhere, and most did not come back.
Channel consolidation. The partner programme was overhauled and many resellers lost authorisation, so the person who had handled an account for a decade often could no longer quote it.
The Arithmetic That Catches People
Core minimums are the mechanism worth modelling, because their effect is largest on exactly the hardware small and mid-sized estates run.
Consider three hosts, two sockets each, and different core counts per socket:
| Cores per socket | Physical cores, 3 hosts | Billed cores at a 16-core floor | Paying for |
|---|---|---|---|
| 8 | 48 | 96 | Twice the cores you own |
| 12 | 72 | 96 | A third more than you own |
| 16 | 96 | 96 | Exactly what you own |
| 32 | 192 | 192 | Exactly what you own |
The pattern is that the floor penalises modest hardware and is invisible on large hardware. An estate of older two-socket eight-core servers pays double. This is also why "consolidate onto fewer, larger hosts" became common advice, and why that advice trades a licensing problem for a bigger failure domain.
Layer the order minimum on top and a genuinely small deployment, a two-host cluster for a branch office, can find itself buying licence capacity for hardware several times its size.
What to Ask Before You Renew
Get these in writing, because each one has caught organisations that assumed continuity:
What is the billed core count, and how is it derived? Physical cores, sockets, minimums applied per CPU and per order.
Which bundle, and what does it include that we do not use? Components you cannot decline still appear in the price.
What is the term, and what happens at the end of it? A favourable first term followed by an uncapped renewal is a common shape.
Is there a cap on the increase at renewal? If not, this year's quote sets no precedent for next year's.
What happens to our perpetual licences? Specifically, until when do they receive security patches.
What are the audit and true-up terms? Growing past the licensed core count mid-term needs a known process rather than a surprise.
The Realistic Alternatives
There are four, and only two are usually serious for an on-premises estate.
| Option | Licensing model | Honest assessment |
|---|---|---|
| Stay on VMware | Per-core subscription with minimums | Least disruption, least leverage. Defensible if the tooling is deeply embedded |
| Proxmox VE | AGPLv3. Optional support subscription per CPU socket per year | Closest functional match for KVM virtualisation with clustering and HA |
| Hyper-V | Windows Server per core, 16-core minimum per server | Sensible where Windows Server and Datacenter licences already exist |
| Public cloud | Per-instance, plus egress and storage | A different cost structure, not a cheaper one. Rarely wins on lift-and-shift |
Two points on that table. Public cloud is frequently proposed at this moment and frequently abandoned once egress and always-on instance costs are modelled, because a stable predictable workload is the case on-premises hardware handles most economically. And Proxmox has no licence cost at all: the software is fully functional from the no-subscription repository, and a support subscription buys the enterprise repository and vendor support rather than the right to run it.
What a Proxmox Move Actually Costs
The software being free does not make the migration free, and pretending otherwise leads to a stalled project. The real costs are:
Learning. Corosync quorum, Ceph placement and watchdog fencing are unfamiliar, and mistakes in the first two can cost data. Budget genuine engineering time.
Migration work. Per-VM preparation, mostly VirtIO drivers on Windows guests and initramfs on Linux, plus a maintenance window each. Our migration runbook covers the failure modes.
Feature gaps. There is no equivalent of Fault Tolerance, and nothing that replaces DRS directly. Anything depending on those needs a design decision rather than a translation.
Hardware. Ceph wants enterprise SSDs with power-loss protection and a 25 GbE network. If the current estate is a RAID controller and 1 GbE, that is a real purchase.
Support. Either an internal on-call capability or a vendor. This is the line item most often left at zero, and it is the one that matters at 03:00.
The honest summary: Proxmox eliminates a recurring licence cost and replaces part of it with operational responsibility. Whether that trade is favourable depends on the size of the quote and whether you have the people.
A Decision Shortcut
If the renewal increase is modest and VMware tooling is embedded across backup, monitoring and automation, staying is defensible. If the increase is a multiple, the estate is dozens of VMs rather than thousands, and the workloads are ordinary Linux and Windows servers, migration pays back quickly.
The case that needs care is the middle: a large estate with a painful quote and deep VMware integration. There, migrate in phases, starting with development and internal tooling, and keep VMware for the workloads that genuinely depend on features Proxmox lacks. A hybrid for a year is a legitimate outcome, not a failure to commit.
Leaving Without Building a Cluster First
The hardest version of this project is migrating off VMware and standing up a first production Proxmox cluster simultaneously, under a licensing deadline. That is when corosync, Ceph and fencing get learned in a hurry, and those are the areas where errors destroy data rather than waste time.
MassiveGRID has operated Proxmox with Ceph in production for years. Workloads can land on clusters that already have HA, automatic failover and three-way replicated NVMe storage, with no hardware to buy and no cluster to commission. Because CPU, RAM and storage scale independently, there is also no incentive to consolidate onto oversized hosts purely to satisfy a licence.
To keep your own hardware and hand over operations, Proxmox support plans run from $99 per node per month for patching and monitoring, through $249 with HA and Ceph management and a four-hour critical response SLA, to $449 with a one-hour SLA and a named engineer. The Enterprise tier includes migration work. The tradeoffs are compared in managed Proxmox versus self-managed, and the hypervisors themselves in Proxmox versus ESXi versus Hyper-V.