Picture a hypothetical company, pieced together from the kind of firm that crops up whenever this subject is discussed. It sells claims-handling software to insurers, employs a few hundred people across Leeds and Manchester, and moved its email, files and customer-facing web apps to the cloud years ago. What it didn’t move was the core: a cluster of VMware hosts in a colocation facility, running a tangle of Windows and Linux servers, a large SQL Server estate and one ageing application that three people understand and nobody wants to touch.
For a long time that arrangement was fine. Then the VMware renewal quote arrived. Since Broadcom completed its acquisition of VMware in late 2023, it has ended the sale of perpetual licences, moved customers onto subscriptions and bundled products that used to be bought separately. Plenty of organisations found their renewal cost far more than they’d budgeted for, and our imaginary insurer is one of them. Its finance director now wants to know whether staying put still makes sense, and its CTO has a separate problem: the product team wants to add AI features that need the core data somewhere a modern AI service can actually reach it.
Those two pressures together are driving what some in the industry call a second wave of cloud migration. The first wave took the easy workloads. This one is about the difficult ones left behind, which is why firms without spare engineering capacity are increasingly looking at migration partners who handle the move in phases rather than attempting one big cutover.
Three options, none of them free
The company has broadly three routes. It can renew with Broadcom, grumble, and buy time. It can move its virtual machines more or less as they are into the cloud, using something like Azure VMware Solution, which runs the VMware stack on dedicated hosts inside Azure, or rehost them as native Azure virtual machines. Or it can use the moment to modernise at least some applications, moving databases to managed services and rewriting the parts that need it.
Most businesses in this position end up with a mix, and the order matters more than the destination. The work breaks naturally into stages, and each stage has a different risk profile. A rehost of stateless web servers is low risk and quick. Moving a busy SQL Server estate is not, because of the cutover window, the dependencies nobody documented and the reports that someone in finance runs every quarter from a server under a desk.
Renewing looks cheapest in the short term and often is, but it only postpones the question, and it does nothing for the AI ambitions. A straight lift into Azure VMware Solution gets the company out of its colocation contract quickly and keeps its operations team on familiar tools, but it means paying for VMware licensing in a new place. Going native on Azure removes that dependency and opens up managed databases and platform services, at the cost of a longer project and a real learning curve for the people who’ll run it.
Why AI changes the maths
The AI angle is what separates this wave from the last one. A company can run a perfectly sensible application on a well-maintained virtual machine for years. What it can’t easily do is point modern AI services at data locked inside that machine. Retrieval-based AI features need data that is reachable, reasonably clean and governed well enough that the security team will sign off on a model reading it. That usually means moving data into managed databases or a proper analytics platform, with identity handled through Entra ID and access controlled by policy rather than by who happens to know the server password.
For our insurer, that tips the decision. Lifting everything as-is into the cloud would fix the licensing problem but leave the claims data exactly as hard to use as before. Modernising everything at once would stall the business for a year. The sensible plan, and a common one for organisations in this spot, is to rehost the bulk quickly to escape the renewal, then modernise the handful of systems that matter for AI, starting with the data they hold.
There are trade-offs even there. Rehosting first means paying twice for a while, once for the old estate winding down and once for the new one ramping up. Modernising the data layer means dealing with that ageing application, and the three people who understand it will have opinions. Cloud costs also need watching from day one, because a rehosted estate that was sized for on-premises hardware is often oversized for the cloud.
The part of the plan that tends to get least attention is the application nobody wants to touch. Every estate has one. Whether it’s rewritten, wrapped in an API or left running on a single VM until the business retires it, somebody has to make that call explicitly. Leaving it undecided is how second migrations turn into third ones.