There is finally a number on leaving VMware, and it will be misquoted all autumn. Gartner puts at 55% the share of enterprises that by 2029 will have initiated proofs of concept with alternatives, up from 25% this year. That verb, initiated, is where two years and a full invoice fit. Our experience is that most of those pilots will not help anyone decide anything.
It was published yesterday by The Register, reading Gartner's new Magic Quadrant on distributed hybrid infrastructure. The assumption, in its literal wording, is that "by 2029, 55 percent of enterprises will initiate proofs of concept for alternative distributed hybrid infrastructure products to replace their VMware-based deployments", up from 25 percent in 2026. It is The Register's article — not the report, which is paid for and which we have not read — that takes care to underline that this means investigation and evaluation, not completed migrations.
The interesting number is the other one
The 55% for 2029 is a projection: it may or may not happen, and nobody will go back to check. The 25% for 2026 is the baseline Gartner gives for this year, and on that baseline one in four has already started some kind of proof of concept. That part does not get argued about in a committee, it shows up in the market: vendors are reworking product tiers, advisories exist that did not exist before, and the catalogue of tools that depended on the ecosystem has become a delicate subject.
There is a second signal from the same week that we find more telling than any percentage. On 14 September Gartner published a Magic Quadrant for server virtualization platforms, signed by five analysts. According to The Register, it had not published one for roughly a decade. A research firm publishes a quadrant when there is a purchasing decision to arbitrate. For ten years there was none: server virtualization was not a decision, it was a renewal. The quadrant coming back says more than the 55%.
And since we are citing that article, let us cite all of it, including the part that is awkward for us. In the distributed hybrid infrastructure quadrant, VMware sits as a leader, alongside AWS, Microsoft, Oracle and Nutanix. Proxmox appears as a niche player, and Gartner gives two reasons: lack of support for top-tier enterprise applications, and a small headcount that makes the availability of that support uncertain (The Register adds that Proxmox has since opened a North American office and offers 24×7 support). We migrate people to Proxmox and we operate it daily, so we say this from experience: the second reason is precisely what has to be settled in the support contract before anything gets signed, and the first is what a pilot has to answer, not a brochure.
A virtualization pilot always succeeds
And that is the problem. It is the easiest experiment to get approved in a committee — it does not touch production, it costs no licences, the team builds it out of written-off hardware — and also the easiest to rig without meaning to. The usual script: three old machines, an afternoon of installation, create a VM, it boots, somebody runs a disk benchmark, a pretty number comes out, and the meeting closes on "it works perfectly".
That proves the hypervisor boots virtual machines. Nobody doubted it. A pilot designed to succeed answers a question nobody asked, and the answer gets presented to the committee as if it were a decision. Then the real migration arrives and the three or four things the pilot never touched show up — which are precisely the expensive ones.
Six things the pilot has to try to break
A pilot that helps you decide is not designed to show the platform works: it is designed to find where it stops working, during office hours and with somebody watching. This is the order we set it up in, and the first one surprises almost everybody.
- The restore, before the migration. Your backup software talks to the hypervisor through a specific interface, and that interface changes with the platform. The first thing to test is not moving a machine: it is leaving a workload running for two weeks on the new platform, restoring it from backup and opening it to check there is data inside. That this is not theoretical we saw in August, when the kit almost every tool on the market depended on was pulled from public download: the first thing put at risk was not anybody's migration, it was the backups of people who had not gone anywhere and would one day need to reinstall.
- Booting is not migrating. Move a real Windows machine, one that has been in production for years, not one installed for the occasion. The disk arrives intact and the system will not boot, because the storage controller it expected to find is no longer there. It is fixable, but you want to know before you plan a weekend window.
- What stays behind on the array. If your plan reuses the storage you already have — and in most cases that is the sensible plan — test what travels and what does not. Snapshots do not travel, and some people find that out on the day they need to roll back an hour.
- Kill a node by hand, on a Tuesday morning. Do not shut it down from the interface: pull the power. What you want to measure is whether the rest of the group agrees on what happened, how long it takes to restart what was running there, and what becomes of the machine that was left halfway. A pilot without a deliberate failure is a vendor demo run by your own team.
- Day 2, inside the pilot window. Upgrade the new platform while you are testing it: a minor version, a node reboot, a repository change. What a hypervisor buys you is two years of updates and the procedure for applying them without cutting anything off. If the pilot runs six weeks and you upgraded nothing in those six weeks, you have not tested the part you will do two hundred times.
- The rollback, timed. Write down what you would do if the pilot workload has to be back on the old platform by Monday at eight, then do it and measure how long it takes. It is the half of the project nobody prepares, and the only half that earns you the right to risk the other one: we go through it in the rollback plan.
None of the six needs licences or extra budget: they need someone to decide the pilot is allowed to fail, which is exactly what does not get decided once the project has a name and a slide. Breaking your own system on purpose, during office hours and with people watching, is an old and boring practice used in places where going down costs real money. It works for the same reason a fire drill works: because on the day it happens for real, nobody is reading the procedure for the first time.
The variable that is not in the report: your renewal date
A market percentage does not tell you when to start; your contract does. The Register recalls a detail that explains much of the industry's calendar, and incidentally where the 2029 date comes from: many customers signed new three-year subscriptions just before the acquisition closed, and if those same customers renew for another three years during 2026, 2029 lands by itself. Which is to say the projection does not describe a collective awakening. It describes when contracts expire.
With three months to go you do not have an alternative, you have a negotiating position you cannot defend: if the other side asks "and if you do not?", the honest answer is that you do not know. Our rule, no mystery to it: start the pilot twelve months before the renewal. Twelve months is enough to test, break things, get it wrong, roll back and still decide calmly, including deciding to stay. But having decided it yourself.
When we still recommend staying
We do not sell VMware licences, and we do not sell Proxmox licences either. We are resellers of neither, which lets us say this without it costing us money: there are cases where leaving now is the worst option. We went through them a month ago in when NOT to migrate from VMware to Proxmox — software certified on a single platform, the number that hurts not being the licence line — so here we only add the one missing there, which is the one we run into most.
It is the one-and-a-half-person team. Changing platform is not only about moving machines: it pours away the operational intuition that person has built up over years, the kind that lets them look at an alert at seven in the evening and know whether it can wait until morning or the laptop has to come out. That intuition is on no TCO spreadsheet, it takes a couple of years to grow back, and for those couple of years the company is more fragile even though the licence bill went down. When the person deciding is also the person on call, we have told them to wait and spend that year on something else: documenting the infrastructure and testing the backups. It costs less and you feel it sooner.
What the public figures show, and we went through them two months ago in the exodus that never quite became one, is a phased reduction of dependency: move out what hurts least first, keep the rest, and renegotiate holding that. Gartner's projection, read literally, describes exactly that way of moving.
How we do it
We run Proxmox VE with Ceph storage in production, spread across several data centres, and we have worked with VMware since versions nobody mentions in press releases any more. We have migrated companies from one platform to the other, and we have also told companies to stay when staying made sense. That is why our consulting is vendor-agnostic: what we hand over at the end of a pilot is the report of what broke, how long it took to come back, and what it would have cost had it happened on a real Tuesday. The recommendation falls out of that on its own, and sometimes it contradicts what we wanted.
And when the decision has been made and it is to leave, what we do is migrate in layers, without switching the business off: first what does not bill, then what bills a little, and the weekend window only for what allows nothing else. The boring part is the good part. If your renewal lands in 2027, the time to set up the pilot is this autumn.
Sources (consulted on 23 Sep 2026): the strategic planning assumption of 55% by 2029 — up from 25% in 2026 — and its literal wording ("initiate proofs of concept"), the clarification that it refers to evaluation rather than completed migrations, the mention of three-year subscriptions signed before the acquisition closed, and the note that Gartner had not published a server virtualization quadrant for a decade all come from The Register's article of 22 Sep 2026, which cites the Magic Quadrant for Distributed Hybrid Infrastructure. The existence and date of the Magic Quadrant for Server Virtualization Platforms (14 Sep 2026, by Tony Harvey, Daniel Bowers, Paul Delory, Tony Iams and Owen Marino) are documented in a press release from one of the vendors included, dated 16 Sep 2026. What this post does NOT claim: we have not read the Gartner reports — they are paid for and we are not subscribers — so every figure and quote attributed to Gartner comes from those two secondary sources and not from the originals; we do not know the sample or method behind the 25% and the 55%, and a strategic planning assumption is by definition a forecast, not a measurement. The six checks in the central section are our own criteria, drawn from migrations we have run, not a recommendation from any vendor or analyst.
When does your renewal fall due?
If it falls within the next twelve months, the pilot gets built now. We design one that tries to break, run it with your team, and hand you the results in writing, including when the result is that staying suits you better.
Talk to everyWAN
Microsoft 365
ES 
CAT 
EN