Broadcom Renewal vs Azure: How to Build an Honest VMware Exit Cost Model

An honest VMware exit cost model puts the full cost of staying next to the full cost of each Azure route, over the same years and for the same workloads, and then adds the one-off cost of moving. Staying means the Broadcom subscription counted per core, plus the hardware, support, power, space and people around it. Most bad comparisons fail because they set a licence quote against a cloud estimate and call it a business case.

This post gives you the method and the rules. It contains no prices, because prices change and yours come from your own quotes and agreements. It sits under our main guide, Leaving VMware After Broadcom, and pairs with choosing a route for each workload.

What should a VMware exit cost model compare?

The model compares scenarios, not products. One scenario is "renew and stay". The others are the routes you might take: Azure VMware Solution, Azure VMs, containers on AKS or on SUSE Rancher with SUSE Virtualization (Harvester), modernising the application, or replacing it with SaaS.

Few estates move as one block. A database server, a file server and an old ERP system rarely belong on the same route. We build the model per workload group, so the answer can be "AVS for this cluster, Azure VMs for these servers, and replace this application", rather than one total for the whole estate.

Three rules keep the comparison fair:

1. Use the same time horizon on both sides. We model five years, because the vSphere 8 support deadline, the Windows Server 2016 ESU years and most hardware refresh cycles all fall inside that window.
2. Use the same workloads on both sides. If you plan to retire some VMs, retire them in every scenario, including "stay".
3. Keep one-off costs separate from running costs, so you can see the payback point rather than a blended average.

How does Broadcom count cores for a renewal?

Broadcom licenses VMware Cloud Foundation (VCF) and VMware vSphere Foundation (VVF) subscriptions per physical core. Each CPU counts as at least 16 cores. Broadcom's own example is a host with two 8-core CPUs, which it licenses as 32 cores.

A few other rules shape the renewal side:

• Broadcom stopped selling perpetual licences and new support (SnS) renewals for them in December 2023. Perpetual licence holders keep the right to use the software but get support only until their current SnS contract ends.
• vSphere Standard and vSphere Enterprise Plus are still listed, but only up to vSphere 8.0 Update 3. There is no vSphere 9 Standard or Enterprise Plus.
• VMware told The Register in September 2026 that it is working on a new vSphere Standard edition. As of October 2026 we found no release date, version number or price for it, so it cannot go into a model yet.
• Distributors told partners in March 2025 that the minimum order would rise to 72 cores per product from 10 April 2025. The rule was withdrawn before it took effect, and Dutch IT Channel reported that Broadcom cancelled it for EMEA. Check the minimum order on your own quote.

A worked core-count example

Take a fictional estate of three host groups. The arithmetic uses only the per-core rule and the 16-core minimum per CPU.

The estate has 256 physical cores but needs a subscription for 304. The 48-core gap comes entirely from CPUs with fewer than 16 cores. The renewal cost is then the licensed cores multiplied by the per-core price on your quote, for each year of the term.

The lesson is that CPUs with fewer than 16 cores cost more per physical core than the spec sheet suggests, and consolidating onto fewer, denser hosts cuts the licensed count. Include every host you intend to license, standby and DR hosts among them, and confirm their treatment with your reseller.

Why not just add a percentage to last year's bill?

Because nobody outside your negotiation knows the percentage. AT&T alleged in a 2024 New York court filing that Broadcom proposed a 1,050% increase in its annual VMware bill; the two sides settled in November 2024 and did not publish the terms. CISPE reported that its members cited increases of 800% to 1,500%. Tesco's court filing claims licensing changes raised its prices by 237%. These are claims made by one side in a dispute, and none is a Broadcom figure. Use the number on your own quote.

What else belongs on the renewal side?

The subscription is usually the line people argue about, but it is rarely the only cost of staying.

Hardware refresh. List each host's age and when it would be replaced during the model period. vSphere 8, vSAN 8, NSX 4, VCF 5.2, HCX 4 and Aria 8.18 reach end of general support on 11 October 2027, and after that date Broadcom gives "no Technical Support and no new patches". Broadcom recommends moving to VCF 9 or VVF 9. Staying therefore includes an upgrade project before October 2027, plus any hardware that project needs.

Support contracts come next: hardware maintenance on servers and storage, plus software tied to the VMware platform.

Power, cooling and space belong on the sheet as well. If the data centre lease runs past the model period, note whether leaving VMware actually lets you shrink it, because a half-empty room still costs the same.

Staff time is the line people inflate most. Count only the hours on patching, upgrades and hardware faults that would change under each route, not salaries you will still pay.

What does each Azure route cost?

Every Azure route has the same three running cost families: compute, storage and networking. Licences, support and management tools sit on top. The routes differ in which of those dominates.

Azure VMs

Compute is priced by VM size and hours, storage by managed disk type and size, and networking by traffic leaving Azure plus connectivity back to your sites. The biggest lever is sizing: VMware allocations are often generous, and copying them one-for-one into Azure carries the waste across. Use the performance data from Azure Migrate to size each VM.

Azure VMware Solution

AVS is priced by node, not by VM. A cluster needs at least 3 hosts and can have up to 16. Since 1 November 2025, new AVS node purchases no longer include a VCF licence, so the AVS line has two parts: the Azure node cost, and a VCF subscription bought from Broadcom that covers every physical core of every node.

Compare that with the worked example above. Its licensed count on-premises was 304 cores. The AVS count depends only on the node type and number you deploy, whatever your VMs use. AVS Generation 2, which deploys private clouds directly into an Azure virtual network, supports only AV64 and is available in UK South and UK West.

Three more AVS rules belong in the model:

• Microsoft reports BYOL registrations to Broadcom monthly, and over-deployed or unregistered private clouds may be suspended. License the cores you actually run.
• Microsoft fixed the number of vDefend Firewall cores included with VCF on 15 October 2025. Use above that needs a Broadcom vDefend add-on.
• If you already run AVS on licence-included pay-as-you-go nodes, you must move to portable VCF by 31 October 2026. Licence-included reservations must be exchanged for VCF BYOL reservations by 30 August 2027, and a reservation that expires earlier loses the licence-included benefit at expiry.

Containers on AKS or on-premises

On AKS, node pools are VMs, so compute follows VM pricing. Windows workloads carry a structural cost: the first system node pool must be Linux, so a Windows-only application still needs a Linux pool. The larger cost is one-off, because containerising an application is a rebuild, not a copy.

If you run SUSE Rancher with SUSE Virtualization (Harvester) on your own hardware, the VMware subscription goes but the hardware, power, space and support lines stay. Add the platform subscription and support from your vendor quote. Our post on Rancher, Harvester and AKS covers that route in more detail.

How do Azure Hybrid Benefit and ESUs change the numbers?

Azure Hybrid Benefit (AHB) lets you use licences you already own in Azure. For Windows Server, you need core licences with active Software Assurance or qualifying subscription licences. Each VM needs at least 8 core licences, and a processor licence counts as 16 core licences.

The 8-core minimum matters for small VMs. Ten VMs with 4 vCPUs each need 80 core licences under AHB, not 40. Count per VM, not in total.

AHB for SQL Server covers Azure SQL Database (vCore tiers, not serverless), SQL Managed Instance and SQL Server on Azure VMs, and Microsoft's AHB page also lists AVS and Azure Local. Microsoft claims savings of "up to 80%" for Windows Server and "up to 85%" for SQL Server against pay-as-you-go rates, and says actual savings vary. Your model should use your own licence counts and your own rates.

ESUs used to be a simple line ("free on Azure"). They now differ by product:

Microsoft announced on 27 March 2026 that new ESU offerings from 1 April 2026 carry the same pricing in and outside Azure. Microsoft's AVS ESU page still says ESUs are free in Azure, but SQL Server 2016 shows that no longer holds for every product. We budget Windows Server 2016 ESUs as a paid line until Microsoft confirms otherwise.

Should you use savings plans or reservations?

Azure offers two savings plans, one for compute and one for databases. Each is a one-year or three-year hourly spend commitment. Unused commitment does not roll over, and you cannot cancel or refund a savings plan.

The compute savings plan covers Azure VMs, App Service, Functions Premium, Container Instances, Dedicated Host, Container Apps and Spring Apps Enterprise. It does not cover software, networking or storage. The database plan covers Azure SQL Database, SQL Managed Instance, PostgreSQL, MySQL, Cosmos DB and DocumentDB, plus hourly SQL Server licences on Azure VMs and Arc.

Reservations give deeper discounts for a specific resource, size and region and suit stable workloads. AVS has its own reserved instances. Our view is simple: reserve the steady base (AVS nodes and VMs whose size you trust), cover the moving part with a savings plan, and commit only after a few weeks of real usage data in Azure. A commitment made on pre-migration sizing locks in the waste.

What one-off costs do people forget?

The one-off column decides the payback date, and it is usually the thinnest column in the spreadsheet.

1. Discovery and assessment. The Azure Migrate appliance discovers servers through vCenter, and you can import an RVTools export (preview) for a quick first pass.
2. Migration effort. For VMware, Microsoft recommends agentless replication. One appliance supports up to 300 VMs replicating at once per vCenter, or 500 with a scale-out appliance. The classic agent-based replication experience retired on 30 September 2026, so older runbooks need updating.
3. Application remediation. Containerising or modernising an application is development work; price it per application.
4. Testing. Every business-critical journey needs testing before and after cutover. Our post on proving every app still works after a VMware migration sets out how.
5. Dual running. You pay for both environments while workloads move. Azure Hybrid Benefit allows 180 days in which licences can run on-premises and in Azure, so plan the overlap to fit. On the AVS route, the on-premises hosts and the AVS nodes both need VCF cores during the overlap unless your Broadcom agreement says otherwise.
6. Exit costs. Include decommissioning, data disposal and any early termination charges on hardware support or colocation contracts.

Can Microsoft funding offset the migration cost?

Possibly, and you should check before you finalise the model. Microsoft announced Azure Accelerate on 7 July 2025, combining Azure Migrate and Modernize, Azure Innovate and Cloud Accelerate Factory. For FY27 (1 July 2026 to 30 June 2027), Microsoft's page uses the name Frontier Accelerate for Azure. It lists Azure credits, partner engagement funding, funded assessments, a sandbox and Cloud Accelerate Factory, and it names VMware to AVS as a supported scenario.

Microsoft partners with the relevant specializations can access these programmes. Microsoft's relevant Azure specializations include Infra and Database Migration to Microsoft Azure, and Microsoft Azure VMware Solution. Microsoft has not published, on the public pages we checked, exactly which specialization each FY27 VMware engagement requires. Microsoft also introduced a VMware Rapid Migration Plan for AVS in 2024; we could not confirm its terms are still current, so do not count on them.

Put funding in its own row, and only once it is approved in writing. A model that only works with funding is a model that does not work.

How do Azure Migrate's business case and Reports help?

Azure Migrate does much of the Azure side for you. It builds a business case from appliance discovery, or from an RVTools import in preview. Its Reports feature (preview, March 2026) produces executive summaries, TCO and ROI views and security insights for Azure VM, PaaS or AVS targets.

Reserved tags for environment and migration intent (April 2026) change assessment and sizing, so tag dev and test servers before you run the business case. Network assessments for VMware and NSX (preview, September 2026) help price the networking line.

Treat the output as a strong draft, and replace its on-premises assumptions with your Broadcom quote and your own running costs.

A template you can fill in

Copy this table into a spreadsheet, one sheet per workload group. Fill each cell with a total for the whole model period, and put the source of every number in the last column.

What are the common mistakes?

1. Counting physical cores instead of licensed cores. The 16-core minimum per CPU changes the total.
2. Comparing a three-year renewal with a one-year Azure estimate, or the reverse.
3. Assuming AVS still includes the VMware licence. It has not for new purchases since 1 November 2025.
4. Pricing AVS by VM rather than by node. You pay for, and license, every core of every node.
5. Assuming every ESU is free on Azure. SQL Server 2016 ESUs are paid, and Windows Server 2012 ESUs have ended.
6. Applying Azure Hybrid Benefit without checking Software Assurance and the 8-core-per-VM minimum.
7. Buying reservations or savings plans before right-sizing. Savings plans cannot be cancelled.
8. Counting funding before it is approved.
9. Leaving out dual running and testing, which push the payback date back.
10. Forcing every workload onto one route because it makes the spreadsheet tidier.

How we build this model at Veriland

We start from your Broadcom quote and your vCenter data, count licensed cores per host group, and price each route per workload group using the rules above. The one-off column includes testing, because we baseline every business-critical journey before migration and re-test it after cutover. If the numbers say renewing is the right call for some workloads, the model says so. Our VMware exit service explains how the assessment works.