Gartner Says 55% of VMware Users Will Test an Exit by 2029. The Hardware Bill Is the Real Story.

Everyone ran the same number this week. Gartner's Magic Quadrant for Distributed Hybrid Infrastructure opens with a "Strategic Planning Assumption" that "By 2029, 55 percent of enterprises will initiate proofs of concept for alternative distributed hybrid infrastructure products to replace their VMware-based deployments and embrace hybrid cloud infrastructure delivery, up from 25 percent in 2026."

Read on its own, that sounds like an exodus. Gartner's own document doesn't. The same report files Broadcom's VMware unit in the Leaders quadrant, next to AWS, Nutanix, Microsoft and Oracle, and lists its core virtualization technology, its sovereign cloud ecosystem and its AI-native infrastructure among its strengths. The analyst house predicting that more than half of VMware's customers will start building exit plans still thinks the platform they are leaving is the best one available.

Both things are true at once, and the reason they are true is the actual story. The 55% figure is not a migration forecast. It is a calendar.

Proof of concept is not an exit

Start with the verb Gartner chose. "Initiate proofs of concept." An enterprise running a proof of concept has not moved a workload. It has built a lab, staffed a project and started the assessment work that Paul Delory, a Gartner research vice president, described in Sydney in May as a job that takes a midsized organisation two years and a large enterprise up to four.

That is the gap the headline hides. Gartner's number counts people opening the door. It does not count anyone walking through it.

There is a second tell in the same report. Gartner published a Magic Quadrant for Server Virtualization Platforms this year, and it had not published one for a decade. The category VMware defined for twenty years had become so settled that Gartner stopped ranking it, and now it is contested enough to rank again. Whatever else the 2026 documents say, they mark the end of an uncontested market.

Why 2029 and not sooner

The date is the interesting part, and Simon Sharwood at The Register did the arithmetic first. Broadcom closed its acquisition of VMware on 22 November 2023. That makes 22 November 2026 the third anniversary, which matters less as an anniversary than as an expiry: plenty of VMware customers rushed to sign multi-year subscriptions before the deal completed, and a large block of those three-year terms come up around now.

Renew during 2026 for another three years, and 2029 is when the paperwork returns. Gartner's forecast is not describing when enterprises will want to leave. It is describing when the last of the pre-acquisition contracts stop insulating them from a renegotiation.

Two more dates crowd the same window. On 31 March 2027, VMware's contracts with many members of its Cloud Service Provider programme expire, leaving those partners unable to sell clouds built on VMware Cloud Foundation. Broadcom signalled it in January 2026, when it told its partner community that the Broadcom Advantage VCSP programme was closed and no contracts would be renewed after 26 January. On 11 October 2027, support ends for version 8 of VCF, forcing every remaining customer to decide whether the estate follows the upgrade path to version 9.

Three deadlines, none of them technical milestones. All of them commercial.

What customers are doing while they wait

The survey evidence has been consistent for a year, and it describes neither a stampede nor a market at peace. CloudBolt's January study of 302 IT decision-makers at North American enterprises, a vendor-commissioned piece of research and worth reading with that in mind, found 86% actively reducing their VMware footprint and 4% having completed a full exit. Costs rose by more than 25% for 59% of them, against the 73% who had expected increases of 100% or more two years earlier. Roughly 72% of the workloads that did move went to public cloud infrastructure, not to a rival hypervisor.

Forrester's Naveen Chhabra put a number on how that reconciles with Broadcom's own claims. Broadcom says more than 87% of its 10,000 largest customers have adopted VCF, a figure chief executive Hock Tan has repeated on earnings calls since March. Chhabra's read of the revenue line is that customers are renewing, but on a fraction of their former estate, by his estimate around a quarter of it. Both statements can hold at once. You keep the contract and shrink the surface it covers.

The bill that lands in hardware

Here is where this stops being a software licensing conversation. Delory's advice to VMware customers in May was blunt: few organisations can leave entirely, no rival hypervisor matches VMware's virtual machine density, and moving therefore means buying more hardware to run the same load. He put the typical licensing increase at 300% to 400%, and warned that the alternatives have costs of their own. Gartner's newest quadrant spells some of them out. AWS's on-premises Local Zones carry a 15% to 35% price premium over their parent regions in expensive metros. Nutanix licensing is "complex and less competitive," and its pricing "frequently exceeds expectations." Microsoft's customers "must navigate multiple disjointed management consoles."

So the migration wave is, mechanically, a hardware refresh wave. More nodes per workload, a second platform running alongside during the transition, and procurement that has to be modelled three years out. That demand lands in the middle of a memory and datacenter capacity squeeze already being driven by AI buildouts, and it arrives on calendars that are now public.

A cash cow inside a chip company

The strategic tension sits inside Broadcom's income statement. In the quarter reported on 2 September, the company posted $29.6 billion of revenue, up 86% year over year. Semiconductor solutions brought in $20.8 billion of it, up 127%. Infrastructure software, the segment that is mostly VMware, contributed $8.8 billion, up 29%, and its share of the group fell from 43% to 30% in a single year. Tan told investors the software business is running at 15% annual recurring revenue growth and guided its next quarter slightly down, to $8.7 billion.

Broadcom's growth story is now custom accelerators sold to a handful of hyperscalers. VMware is the high-margin, slow-growing, cash-generating asset sitting beside it. That is not a reason for the product to fail. It is a reason to expect it to be managed for margin rather than for market share, which is roughly what the last three years have looked like.

The counterweight from Broadcom is real and worth stating. VCF 9 is a substantial release. The company's Cloud Provider programme spans some 50 independent local providers, which is the shape of an answer to European sovereignty demands. Gartner lists sovereign AI on VMware infrastructure as a genuine strength.

There is also a European problem that does not appear in any quadrant. CISPE's monitoring body, the European Cloud Competition Observatory, published its fourth report this month and kept Broadcom at Red, its most severe rating. The group, which has a live complaint before the European Commission and is calling for interim measures, reports renewal cost increases of ten times or more, and points to a Compliance Reporting mechanism introduced in VCF 9.0 and later that requires a report every 180 days with management-plane degradation as the documented penalty for missing one. CISPE calls it a kill switch. Its own assessment scores VCF against 52 sovereignty criteria and finds three compliant. Treat all of that as an interested party's characterisation, because that is what it is. Broadcom disputes it and is contesting the process.

What it means if you build the hardware

The 55% number will get repeated all week as evidence of mass defection. The more useful reading is narrower and more concrete. Enterprises are not leaving VMware. They are learning to run less of it, across more platforms, and the contracts that decide when that happens are already dated.

For anyone selling hardware into enterprise infrastructure, that is a demand signal with a schedule attached. Refresh planning, node counts per workload, the storage and networking that arrives with a second hypervisor, and the cost modelling that decides whether a move makes sense at all: those questions sit downstream of a forecast that reads like a software story and behaves like a procurement one. The awkward part is that the node counts, the migration cost curve and the sourcing plan usually sit with three different owners. At DMC, we help hardware companies turn that kind of forecast into a sourcing and production plan they can commit to, including when the transition collides with a component market that is already tight. If you are working through a refresh cycle of your own, let's talk.