VMware licensing in 2026: what has actually changed and the decision 2027 puts in front of you

Comparativa Proxmox vs VMware

VMware licensing under Broadcom has been generating headlines for almost three years, and by now it is hard to tell what is actually in force from what was announced, withdrawn, and still circulates as if it were standing. Making that distinction is worth doing before any infrastructure decision, because the two things that really put pressure on a budget are not the ones that made the most noise.

The short version: what changed structurally is the purchasing model, with mandatory subscription, per-core licensing, and a catalog reduced to two bundles. What remains in force as a licensing floor is 16 cores per physical socket, not the 72 per order that get quoted so often and that were withdrawn days after being announced. And the date that really matters is October 11, 2027, when vSphere 8 reaches end of life with no supported path for the Standard and Enterprise Plus editions, with the added twist that the three-year contracts signed in 2024 and 2025 expire together just before it.

What Gartner says, and what it does not

The figure that circulated most this week comes from a Gartner report covered by The Register on September 22, 2026: 55 % of enterprise VMware users will have started proof-of-concept work with alternatives before 2029, up from the 25 % doing so in 2026. It is a forecast rather than a market-share measurement, and that distinction matters. It does not say 55 % will migrate, it says more than half will evaluate.

The gap between evaluating and migrating is exactly where the decision sits. A proof of concept costs weeks and commits nothing; a migration commits the platform for years. That the number of companies willing to do the first should double in three years is significant for what it reveals about the climate, not because it announces an exodus.

Licensing timeline, with dates

The facts, in order. The reversals are included, because listing them is what makes it possible to separate a condition that applies from an announcement that never took effect.

DateWhat happened
November 2023Broadcom closes the VMware acquisition.
Late 2023Perpetual licenses stop being sold. Those already purchased keep working; no new ones are issued.
Late 2023The catalog goes from more than 160 products to two bundles: vSphere Foundation and VMware Cloud Foundation. The Essentials kits are retired.
Late 2023Licensing moves from socket to core, with a minimum of 16 cores per physical socket.
Early 2024Free ESXi is withdrawn.
2024Channel restructuring: most customers move to buying through an authorized reseller.
March 28, 2025Partners are told of a minimum of 72 cores per order line, effective April 10.
~April 10, 2025That 72-core minimum is withdrawn after the channel reaction. The 16-core per socket floor applies again.
April 2025Free ESXi returns with vSphere 8.0 Update 3e, limited to non-production use: no vCenter, a maximum of 8 vCPU per virtual machine, and no support.
~May 2025Notices begin reaching holders of perpetual licenses with expired support, asking them to remove patches applied after expiry.
July 31, 2025vSphere Standard reaches end of sale.
December 2025Broadcom withdraws vSphere Foundation from several EMEA countries, leaving only Cloud Foundation available.
vSphere 9Ships exclusively inside the subscription bundles. There is no standalone edition.
September 2026Broadcom is reported to be preparing a return of vSphere Standard, subscription-based and licensed per core with the 16-core per processor floor. Price, availability and commercial terms have not been published.
October 11, 2027vSphere 8 end of life. Standard and Enterprise Plus have no supported continuation.

What applies today and what never took effect

Of everything above, this is what shapes a budget today:

  • Mandatory subscription: there is no route back to a perpetual license. Stop paying and the hosts stop being managed.
  • 16 cores per physical socket as the licensing floor. This is the one that applies.
  • Two bundles, and in several EMEA countries only one, Cloud Foundation, since December 2025.
  • The 72-core per order minimum is not in force. It was announced and withdrawn in April 2025, and it was a per order line floor rather than a per socket one. It still gets quoted as though it were a live condition.
  • Free ESXi exists but is useless in production: 8 vCPU per machine, no vCenter, no support. It is a lab, not a platform.
  • The return of vSphere Standard is an announcement, not an offer. As of September 2026 there is no published price, availability or commercial terms, so it cannot be put into a budget yet.

It is worth checking the actual quote against what has been reported. Several of these conditions have been applied differently depending on channel, region, and how the offer was structured, so the only reliable version is the one written into the specific quote.

The two facts that bite in Europe

Public debate has focused on pricing and on the reversals, but for a mid-sized European company what changes the planning is something else.

vSphere Foundation withdrawn in EMEA countries

Since December 2025, in several EMEA countries the catalog narrows to VMware Cloud Foundation. For an environment that fit the middle bundle, that is not a price increase, it is a jump to a different product category, with everything that implies in cost and in features that will go unused. It is the point that hits hardest for anyone running between two and ten hosts who does not need a full cloud platform.

The 2026 and 2027 renewal wave

The three-year contracts signed in 2024 and 2025, once the door to perpetual licenses closed, expire together across 2026 and 2027. They coincide with the end of life of vSphere 8 on October 11, 2027, which leaves Standard and Enterprise Plus without a supported continuation. Renewing without having evaluated alternatives commits another three years; evaluating them requires starting with room to spare, because a serious proof of concept does not get built in a fortnight.

At Stackscale we see this in practice: migration conversations that arrive with time work out well, and the ones that arrive with the renewal already due turn into a price negotiation with no real alternative on the table.

What migrating actually involves

This is where the debate usually stays general, and it is the part that weighs most on the decision. Migrating from VMware to Proxmox VE is not a tooling problem, it is a planning and prior testing problem. With those two things done, the cutover is measured in minutes.

The cutover is short, and the order matters

The usual procedure is to power off at the source and power on at the destination, with the disk format conversion from VMware to Proxmox completed afterward, while the machine is already running. That is why the maintenance window stays short: nothing waits for the conversion in order to boot. Depending on the environment and on the extra work each machine needs, the cutover can run to a few minutes or somewhat longer, but the impact ranges from almost none to very low.

With high availability, RTO and RPO at zero

When the source environment already runs high availability, the migration can happen live. One leg of the cluster is migrated and mixed high availability between VMware and Proxmox is maintained through the transition, which puts the cutover at RTO = 0 and data loss at RPO = 0.

Whether that is possible depends less on the hypervisor than on what sits underneath. At Stackscale it rests on real networking with no dependency on the virtualization layer, on network storage, and on synchronous replication between nodes, which is what allows both platforms to stay up at once without losing state.

Where the time goes, and what happens when something breaks

The time is not in the cutover, it comes before: in the inventory, in the preparation, and in testing each type of machine. Those prior tests are precisely the preparation for incidents, which is why migration day holds few surprises. When one does come up, rolling back is as fast as moving forward, because the source environment stays in place until validation at the destination is closed out.

Our customers have already migrated thousands of virtual machines, and in every case the support went wherever it was needed, from cluster design through to final validation.

What to do depending on where you stand

Current contract with a distant renewal

This is the comfortable position, and the one worth using. A real inventory of machines, consumption, and dependencies, plus an unhurried proof of concept, settles the decision before the renewal arrives. The advantage of evaluating early is that the negotiation happens from a different position, whatever the outcome turns out to be.

Renewal within the next twelve months

Here the order reverses: first find out what the alternative costs and how long it takes, then sit down to renew. Without that figure the negotiation has no counterweight. Twelve months is enough for a full proof of concept and a phased migration, but not if the thinking starts in the final quarter.

Perpetual licenses with expired support

This is the most exposed position, and not because of cost but because of patching: applying security updates after support has expired is what prompted the 2025 notices. A production environment without security patches is a different and more urgent problem than a licensing one, and it deserves to be treated as such.

Frequently asked questions

Is the 72-core per order minimum still mandatory?

No. It was communicated to the channel on March 28, 2025, effective April 10, and withdrawn around that same date after the channel reaction. The floor that applies is 16 cores per physical socket, which is earlier and different: one was per order line, the other is per processor.

Can I keep using my perpetual VMware licenses?

Perpetual licenses already purchased keep working. What is no longer possible is buying new ones or, with support expired, applying patches released after that expiry. Since May 2025 notices to that effect have gone out to holders in that situation.

Is free ESXi usable for a small production environment?

No. The version reinstated in April 2025 with vSphere 8.0 Update 3e is limited to non-production use: no vCenter, a maximum of 8 vCPU per virtual machine, and no support. It is fine for a lab and for testing.

What happens on October 11, 2027?

vSphere 8 reaches end of life. The Standard and Enterprise Plus editions have no supported continuation, because vSphere 9 ships exclusively inside the subscription bundles. Anyone on those editions has to decide before that date.

How long is the cutover in a migration to Proxmox VE?

With planning and prior testing, the cutover is measured in minutes: power off at the source, power on at the destination, and the disk conversion completes afterward while the machine runs. If the source environment already runs high availability, it can be done with no cutover at all, keeping mixed high availability across both platforms through the transition, with RTO and RPO at zero.

Can we roll back if the migration does not go as expected?

Yes, and in the same timeframe. The source environment stays available until validation at the destination is closed out, so rolling back is as fast as moving forward.

Is migrating to Proxmox VE only a way to save on licensing?

Savings are why many companies open the conversation. That reading falls short: what gets decided in a migration is which platform the business will operate on for the next few years, who maintains it, and which vendor dependencies come with it.

Where Stackscale fits

VMware to Proxmox VE migrations that go well all look alike: an honest inventory, testing by machine type, and a short maintenance window because the heavy work was done beforehand. What the infrastructure contributes is the ability to keep both platforms running at once, and that depends on networking, storage, and replication rather than on the hypervisor.

At Stackscale we deploy and operate Proxmox VE in production on dedicated infrastructure in our data centers in Madrid and Amsterdam, with our own network, network storage, and synchronous replication. If you are working out what to do before your next renewal, talk to us about your private cloud project and we will size it with your data.

Further reading

Sources

Share it on Social Media!

Cookies customization
Stackscale, Grupo Aire logo

By allowing cookies, you voluntarily agree to the processing of your data. This also includes, for a limited period of time, your consent in accordance with the Article 49 (1) (a) GDPR in regard to the processing of data outside the EEA, for instead, in the USA. In these countries, despite the careful selection and obligation of service providers, the European high level of data protection cannot be guaranteed.

In case of the data being transferred to the USA, there is, for instance, the risk of USA authorities processing that data for control and supervision purposes without having effective legal resources available or without being able to enforce all the rights of the interested party. You can revoke your consent at any moment.

Necessary Cookies

Necessary cookies help make a web page usable by activating basic functions such as the page navigation and the access to secure areas in the web page. The web page will not be able to work properly without these cookies. We inform you about the possibility to set up your browser in order to block or alert about these cookies, however, it is possible that certain areas of the web page do not work. These cookies do not store any personal data.

- moove_gdpr_popup

 

Analytical cookies

Analytical cookies allow its Editor to track and analyze the websites’ users behavior. The information collected through this type of cookie is used for measuring the activity on websites, applications or platforms, as well as for building user navigation profiles for said websites, application or platform, in order to implement improvements based on the analysis of data on the usage of the service by users.

Google Analytics: It registers a single identification used to generate statistical data about how the visitor uses the website. The data generated by the cookie about the usage of this website is generally transferred to a Google server in the USA and stored there by Google LLC, 1600 Amphitheatre Parkway Mountain View, CA 94043, USA.

- _dc_gtm_UA-XXXXXXXX-X

- _gat_gtag_UA_XXXXXXXX_X

- _ga

- _gcl_au

- _gid