For twenty years the answer was VMware and the question was which edition. The licensing change that followed the Broadcom acquisition ended that, not because the product got worse but because the buying model stopped fitting the customers it was sold to.
Perpetual licenses were withdrawn in favor of subscription. The product line collapsed into a small number of bundles. Minimum core counts per processor were introduced, and the free hypervisor was discontinued. A small estate that used to buy a few sockets now buys a bundle with a floor under it.
Who is affectedThe size where it bites
The change is close to neutral for large estates. A company with hundreds of hosts was already on an enterprise agreement, already buying the higher editions, and already subscribing to most of what is now bundled.
It lands hardest on the small end: three hosts, a couple of hundred virtual machines, previously running a modest edition bought outright years ago and renewed for support. That buyer now faces a subscription with a core minimum, for features they did not ask for, at a multiple of what they were paying.
The middle is where the analysis actually has to be done, and it turns on how much of the bundle you would have bought anyway.
The comparisonWhere each one is genuinely better
Stripped of vendor argument, the differences are narrow and specific.
| Criterion | Hyper-V | VMware |
|---|---|---|
| Cost for a small estate | Included with Windows Server | Subscription, with a core floor |
| Third party backup support | Good | Best in class |
| Storage features | Storage Spaces Direct, capable | vSAN, more mature |
| Live migration | Works well | Works well, better tooling |
| Ecosystem and appliances | Fewer vendors ship for it | Almost everything ships for it |
| Skills in the market | Common | Common, and better certified |
| Linux guest support | Fine, occasionally fiddly | Excellent |
| Management at small scale | Simple, if you know Windows | Better, and costs more |
Read that table without the price column and VMware wins on most rows, mostly by small margins. Read it with the price column and the small estate case reverses, because those margins do not justify the difference any more.
ArithmeticThe license that does not change either way
One number belongs on both sides of the comparison and is regularly put on only one.
Windows guests are licensed per host regardless of which hypervisor runs them. Standard edition covers two virtualized instances per licensed host and Datacenter covers as many as fit, and neither figure moves because the hypervisor changed. An estate running mostly Windows is already paying that bill and will keep paying it.
Comparing a hypervisor subscription against zero therefore overstates the saving, sometimes by a lot, because the alternative is not free either: it is the same guest licenses plus the Windows Server licenses that carry the hypervisor role.
On a Windows estate that is close to free at the margin, which is the real argument, and it is a different argument from free.
The comparison that holds up puts the same guest licenses on both sides, then compares what is left: a subscription with a core floor against Windows Server licenses you were largely buying anyway, plus the migration, plus the year after it.
The third optionEverything else on the shortlist
Two alternatives now appear in serious evaluations that would not have three years ago.
Proxmox is open source, built on KVM, and has matured into something that runs production workloads for organizations that are comfortable without a vendor to call. The support subscription is inexpensive. The gap is ecosystem: fewer appliances are certified for it, and some backup vendors treat it as a second class target.
Nutanix and other hyperconverged platforms answer the storage question at the same time, which changes the comparison from hypervisor against hypervisor to a whole architecture decision. They are not cheap, and they compete on operational simplicity rather than price.
Neither is a drop in replacement, and treating either as one is how a migration becomes a year long project.
MethodHow to actually decide
The question is not which hypervisor is better. It is whether the difference between them is worth what it now costs, for this estate.
Price the real renewal, not the list price. Get an actual quote for the actual core count. The core minimums change the arithmetic in ways a spreadsheet built on old assumptions will not show.
List what you use, not what you have. Most small estates use a fraction of the features in the bundle they are now required to buy. That fraction is the honest comparison against an alternative.
Price the migration, including the year after it. Retraining, rewritten runbooks, backup reconfiguration, and the monitoring that has to be rebuilt. A migration that saves money in year one and costs a specialist in year two has not saved anything.
Check what your backup product supports. This kills more migrations than any other single factor, and it is the cheapest thing on this list to check.
ScopeWhat converts, and what is rebuilt from nothing
The estimate everybody gets wrong is the migration, and it is the one part of this decision that can be scoped precisely before committing to anything. Virtual machines are the easy half. The things arranged around them are the half that takes the quarter.
| What you have | On the other side | Effort |
|---|---|---|
| Virtual machine disks | Convert reliably, one tool, one pass | Time, not thought |
| Snapshots and checkpoints | Do not travel | Consolidate before touching anything |
| Guest tools | Removed on one side, installed on the other | Per machine, and order matters |
| Templates | Rebuilt | A day each, and a good time to update them |
| Resource pools and vApps | No equivalent | Rethink, not port |
| Affinity and placement rules | Different mechanism, needs the management layer | Rewrite |
| Distributed switching and overlays | Different product entirely | Design work |
| Backup jobs | Rebuilt against a new target | Plus a test restore of every class of machine |
| Monitoring and alerting | New checks, new thresholds | The part that gets skipped |
Read the right hand column as the project plan. The disks are a weekend. Everything below them is the reason the migration takes a quarter and the reason it is worth pricing honestly before deciding, because a saving that is smaller than the work is not a saving.
PracticeThe traps in the move itself
Five things go wrong often enough to plan for, and none of them is difficult once it is expected.
The network address of every machine changes underneath it. A converted machine gets a new virtual network adapter with a new hardware address.
Every DHCP reservation keyed to the old one stops matching, every license bound to it needs reissuing, and anything with a static entry in a switch or a firewall stops resolving. Collect the list before the move rather than discovering it one application at a time.
Remove the old guest tools before converting, not after. A machine that arrives with the previous platform's drivers still installed can boot to a machine that has no working storage or network driver and no way to install one, which is a recovery job rather than a retry.
Snapshots have to be gone first. A machine with a chain of them converts the base disk and quietly leaves the changes behind, and the failure is silent: the machine boots, and it is the machine from three weeks ago.
Firmware type has to match how the guest boots. A machine that boots UEFI needs the equivalent generation on the other side, and picking the wrong one produces a virtual machine that will not start with no useful message about why.
This is decided at creation and cannot be changed afterwards, so it is worth checking per machine rather than assuming a default.
Both platforms need their capacity at the same time. The storage cannot be shared: one side cannot read the other's file system, so the disks are copied rather than handed over. That means enough room for both copies of everything still to move, for as long as the move takes.
The unpopular answerWhen staying put is correct
A stable estate, a renewal that is expensive rather than impossible, and a team that knows the platform is a case for paying and moving on. Migrations consume attention that could go somewhere it produces something.
The version worth avoiding is the decision made twice: pay this renewal while telling everyone you are migrating, do nothing for a year, and arrive at the next renewal with the same choice and less time. If staying is the answer, say so and stop evaluating.
- Get a real quote for your real core count
The core minimums change the arithmetic. A spreadsheet built on the old socket model will not show what the renewal actually costs.
- List the features you use, not the ones you own
Most small estates use a fraction of the bundle they are now required to buy. That fraction is the honest comparison against an alternative.
- Price the migration and the year after it
Retraining, runbooks, backup reconfiguration and monitoring. Saving money in year one and needing a specialist in year two has saved nothing.
- Confirm your backup product supports the target
This kills more migrations than anything else on the list, and it is the cheapest item to check before any of the rest.
The three numbers that decide it
Everything else in the evaluation is detail hanging off these.
FAQQuestions from the comments
Is Hyper-V free?
The standalone free hypervisor is discontinued, but the role is included with Windows Server, which most organizations already license. For a Windows estate that is close to free at the margin.
Is Proxmox ready for production?
For organizations comfortable without a vendor to call, yes, and many run it. The gap is the ecosystem rather than the hypervisor: fewer certified appliances, and uneven backup vendor support.
How hard is a VMware to Hyper-V migration?
The virtual machines convert reliably. The work is everything around them, and the surprise is usually the backup product, the monitoring, and the appliances that only ship as VMware images.
Does staying on VMware carry risk?
The commercial risk of another change in terms, more than a technical one. The product is not going away and neither is the support for it.
What about running both?
Common, and it works, at the cost of two sets of skills, two backup configurations and two runbooks. It is a transition state rather than a destination.
Do I still pay for Windows Server licenses after moving off VMware?
Yes, and unchanged. Windows guests are licensed per host whichever hypervisor runs them, with Standard covering two virtualized instances per licensed host and Datacenter covering as many as fit. Putting that number on only one side of the comparison is the most common way the saving gets overstated.
What actually breaks when a virtual machine is converted?
The network adapter is new, so its hardware address changes, and everything keyed to the old one stops matching: DHCP reservations, licenses bound to it, static entries in switches and firewalls. Collect that list before the move rather than finding it one application at a time afterwards.
Can I move the storage rather than copying it?
No. Neither platform can read the other file system, so every disk is copied, and both sides need their capacity for as long as the move takes. On a full array that constraint decides the order of the whole project.
Why did a converted machine come back with three weeks of changes missing?
A snapshot chain that was not consolidated first. The conversion took the base disk and left the differencing files behind, and the machine boots perfectly as the machine it was before the snapshot was taken. Consolidate everything before the first conversion.
How long should I budget for a three host estate?
The disks are a weekend and the rest is a quarter. Templates rebuilt, backup jobs recreated with a test restore of every class of machine, monitoring rewritten, and the applications whose owners only appear when something stops. The disks are never what runs over.
Is it worth waiting to see whether the terms change again?
Only with a date attached. The decision that costs the most is the one made twice: pay this renewal while telling everyone a migration is coming, do nothing for a year, and reach the next renewal with the same options and less time. Either commit to the move or commit to staying and stop evaluating.