In the Net ⛓ Episode 04
You bought VMware once, outright, the way you buy a tool. You paid for a perpetual licence, you owned it, and it kept running whether or not you ever spoke to the vendor again. In November 2023 a 69 billion US dollar (around 64 billion euro) acquisition closed, and over the following eighteen months the thing you owned was quietly, methodically converted into a thing you rent. This is the fourth episode of In the Net, a series on the documented mechanics of vendor lock-in. The premise has not changed. Every platform tells you how to come in. The architecture, and increasingly the contract, tells you whether you can leave, and on whose terms.
This episode is slightly different from the first three. Adobe, LinkedIn and AWS each built their lock-in into the product. VMware's lock-in was built into a transaction. The product barely changed; the ownership of it did, and the terms followed.
The Promise
VMware made the hypervisor boring, and that was the whole point. A hypervisor is the layer that lets one physical server pretend to be many; VMware's ESXi and vSphere did this so reliably that an entire generation of data centres was built on the assumption that it simply worked. vMotion moved a running virtual machine from one host to another without dropping a packet. Distributed Resource Scheduler balanced load without anyone watching. For roughly two decades, vSphere was the quiet floor the enterprise data centre stood on.
Crucially, it was sold as a perpetual licence. You paid once for a version, and you owned the right to run it indefinitely. Support and upgrades were a separate, optional subscription. The distinction mattered: the software you ran was yours, and the relationship with the vendor was something you chose to maintain, not something you were compelled to renew. That distinction is the thing this episode is about, because it is the thing that was removed.
The Hooks
Broadcom announced its intent to acquire VMware in May 2022 and completed the acquisition on 22 November 2023, after some eighteen months of regulatory review by the US FTC, the UK Competition and Markets Authority and the European Commission. The deal was valued at around 69 billion US dollars, including roughly 8 billion in assumed debt. VMware was delisted from the New York Stock Exchange. None of the customers running vSphere were consulted, which is unremarkable: that is how acquisitions work. What followed is the part worth documenting.
Perpetual licences ended. Broadcom moved VMware to a subscription-only model. Existing perpetual licences do not technically expire, but they no longer receive support or security updates, and at renewal the customer is moved onto an annual subscription. In practical terms, for any estate that needs patched, supported software, ownership became tenancy.
The free tier was removed, then quietly returned. The free edition of ESXi, long the entry point for labs, edge nodes and small deployments, was withdrawn in early 2024. In April 2025 it returned, as ESXi 8.0 Update 3e, without a press release: the news appeared in the product release notes, which described it as an entry-level hypervisor for non-production use, with no Broadcom support, and unable to connect to vCenter. A capability removed with a transition plan and reinstated in a footnote tells you something about who the communication is for.
Per-CPU pricing became per-core, sold in bundles. Licensing shifted from a per-socket model to a per-core model, and the products were repackaged into a small number of bundles: VMware vSphere Foundation (VVF) and the larger VMware Cloud Foundation (VCF). These bundles include components such as NSX (networking) and vSAN (storage) whether or not the customer runs them. A customer who wanted vSphere and nothing else now buys a suite, and the suite is priced per core.
Each of these, on its own, is a defensible commercial decision. Taken together, over eighteen months, they convert a one-time purchase into a recurring obligation whose size the vendor sets.
The Standing
VMware entered this period holding roughly 70 per cent of the server-virtualisation market on Gartner's figures for 2024. Gartner has since projected that share falling to around 40 per cent by 2029, a consequence the firm attributes directly to the post-acquisition strategy. A vendor does not shed thirty points of a market it dominates because the technology got worse; it sheds them because the terms changed faster than the customers could be locked down.
The clearest signal came from the distribution channel rather than the customers. In December 2024, Ingram Micro, one of the largest technology distributors in the world, announced it would end its Broadcom and VMware distribution relationship in select regions from early January 2025. Distributors do not walk away from a 70-per-cent-market-share product over a disagreement about colour schemes. When the people whose business is selling the product decide it is not worth selling, the contract on offer was not a friendly one.
How the User Is Treated
The dignity dimension this series tracks has, in VMware's case, an unusually clean piece of evidence: a court filing.
In August 2024, AT&T filed suit against Broadcom in the New York State Supreme Court. AT&T held perpetual VMware licences and, critically, had signed an amended support agreement with pre-acquisition VMware in 2022, running through 8 September 2024, with an option to renew for two further years. After the acquisition, AT&T alleged, Broadcom declined to honour that renewal option unless AT&T purchased VMware's new subscription bundles under a three-year commitment. AT&T sought injunctive relief; by October 2024 the parties were reported to be moving towards a settlement.
Set aside the legal merits, which the settlement leaves unresolved in public. The structural fact is the one that matters for this series. AT&T is one of the largest telecommunications companies on earth, with a procurement department the size of a mid-tier vendor, and it concluded that the only way to obtain support it believed it had already contracted for was to sue. Between AT&T's courtroom and the small cluster paying for cores it cannot use sits every customer too large to ignore the bill and too small to file in the New York State Supreme Court.
The Exit That Isn't
The exit problem here is not technical in the way AWS' was; a virtual machine is far more portable than an IAM policy. The exit problem is contractual and informational, and the clearest illustration is the 72-core episode.
In March 2025, reports emerged that Broadcom would raise the minimum core count per VMware order from 16 to 72, effective 10 April 2025. For a customer with a 32-core cluster, this meant buying 72 cores and paying for 40 they could not use: a 125 per cent increase for no additional capability. After a sharp industry backlash, the requirement was withdrawn, with distributors confirming the minimum stayed at 16. When asked, a Broadcom spokesperson stated that the company had "never announced a price change".
This is worth sitting with, because it is the exit problem in miniature. A pricing rule that can appear, reshape a customer's renewal maths for several weeks, and then be denied as never having existed is a pricing rule the customer cannot plan against, cannot cite in a negotiation, and cannot appeal. The lock-in is not only that migration takes time; it is that the terms you are migrating away from will not hold still long enough to be measured.
This is lock-in by design, in the precise sense this series uses the phrase. Not a boardroom conspiracy to trap users, but an arrangement in which the architecture and the contract together produce the outcome that customers stay and pay even when, on their own honest accounting, they would rather not. Ownership was converted to tenancy by a transaction the customer was never party to, and the terms of the tenancy are set, revised and occasionally un-set by the landlord.
The Price
The price of staying is the renewal, and the renewals have been steep. Reported increases vary enormously by customer, tier and how much grandfathering survived the transition: figures from 150 per cent to several hundred per cent are common in trade reporting, with some large or specialised estates citing four-figure percentage increases. These are reported numbers from customer accounts and analyst write-ups rather than published list prices, and should be read as a range, not a constant; the precise figure depends on the bundle, the core count and the negotiation.
The price of leaving is a hypervisor migration, which is real work but bounded work. A virtual machine has a portable representation: the Open Virtualization Format (OVF) for the metadata, and disk images that convert cleanly to qcow2 for KVM-based platforms. The migration cost is in testing, in re-tooling the operational layer (backup, monitoring, automation that assumed vCenter), and in retraining. It is measured in months for a large estate, not years, and it is a one-time cost against a recurring one. That asymmetry, a bounded one-time exit cost against an unbounded recurring stay cost set by the vendor, is exactly the calculation that shifts when the renewal letter arrives.
The Escape Route
The escape route from VMware is unusually mature, for a reason worth naming: the alternatives reached production quality at almost exactly the moment the pricing crisis hit. The timing was not planned by anyone, but it changed the negotiation.
Proxmox VE is the headline alternative. It is an open-source virtualisation platform built on KVM and QEMU for full virtual machines and LXC for containers, with clustering, live migration, software-defined storage (including Ceph) and a web interface. It is free to use with no feature gates; a support subscription is optional and starts at around 115 euro per CPU socket per year. For the small and mid-sized estates that felt the 72-core maths most acutely, the cost delta against a VVF or VCF renewal is not subtle.
XCP-ng with Xen Orchestra is the Xen-based alternative, with a strong following in environments that prefer the Xen hypervisor and want a polished management layer. Nutanix AHV is the hyperconverged option for customers who want an integrated appliance-style platform. Microsoft Hyper-V and Azure Stack HCI exist for Windows-centric shops, with their own lock-in characteristics to weigh. OpenStack remains the option at genuine cloud scale, with genuine operational weight to match.
And the FreeBSD stack is the quietly elegant choice for those who want their virtualisation layer to be a small, auditable part of a coherent operating system rather than a product in its own right: bhyve for virtual machines, jails for containers, ZFS for storage, all in the base system and none of it for sale to an acquirer. This is the point worth holding onto, because it is the structural answer to the whole episode: a perpetual licence can be revoked at the next renewal, but a base-system component is not a product anyone can buy out from under you.
For those who miss the point-and-click of a Proxmox dashboard, Sylve puts bhyve and jails under one Proxmox-style web interface, written in Go and SvelteKit, with clustering and scheduled backups via zelta. Sylve made its first release in early 2026 and sits at v0.2.x at the time of writing; it requires FreeBSD 15.0 or later, and it is honestly young: not yet a drop-in for an enterprise estate that needs support contracts and proven HA at scale, but already productive for homelabs and FreeBSD-first shops whose operators are comfortable on the v0.x curve. The substance is the stack, which is mature; Sylve is the convenience layer, which is new.
The migration path itself is the open one: export from VMware to OVF, convert disks to qcow2, import to the target. The Proxmox project ships a VMware importer that automates the bulk of this for the common case. The work is in the operational tail, not the disk images.
Coda
VMware did not get worse. vSphere still moves a live machine without dropping a packet, and vMotion is still a small marvel of engineering. The product is not the issue. The issue is that the product was sold as something you own, and an acquisition you had no part in converted it into something you rent, with the rent set by a landlord who reserves the right to revise the terms and then deny having done so.
The lesson generalises well past virtualisation. A perpetual licence is a promise about the future, and a promise about the future is only as durable as the entity that made it. The moment a layer of your architecture stops being replaceable, its price stops being negotiable. The defence is not loyalty to a vendor or hostility to one; it is keeping the replaceable layers replaceable, so that the renewal letter is an invitation rather than a summons.
You owned a hypervisor. You were handed a subscription. The hypervisor, it turns out, was never the expensive part.