VMware
VMware per core licensing. The real math.
The meter counts physical cores with a 16 core floor per socket. Host shape, density, and a defensible count decide the invoice.
The meter counts physical cores with a 16 core floor per socket. Host shape, density, and a defensible count decide the invoice.
VMware subscriptions bill per physical core with a 16 core minimum per CPU, so the bill is set by how you count, configure, and consolidate hosts rather than by what you deploy.
VMware licensing bills per physical processor core across every host in the licensed estate, sold as a subscription to VMware Cloud Foundation or vSphere Foundation. Perpetual licenses and standalone SKUs are gone; the core meter is the whole model, as Broadcom's VMware portfolio page makes explicit.
The meter ignores utilization. A host at 20 percent load bills the same as one at 90 percent, which is why density and host shape, not deployment choices, decide the invoice.
Per socket licensing ended with the subscription transition. Estates designed for socket economics, many small two socket hosts, now carry the worst shape for core economics.
The 16 core minimum bills the gap between your actual cores and 16 per CPU, so an estate of 8 and 12 core processors pays for cores that do not exist. The penalty is structural and compounds across every socket in the estate.
The 16 core minimum in practice
| CPU configuration | Actual cores | Billed cores | Penalty |
|---|---|---|---|
| 2x 8 core | 16 | 32 | 100 percent |
| 2x 12 core | 24 | 32 | 33 percent |
| 2x 16 core | 32 | 32 | None |
| 2x 32 core | 64 | 64 | None, best density |
Branch and edge estates with small dual socket hosts feel it worst. A retail or manufacturing footprint of dozens of 8 core hosts can carry a 40 to 100 percent billing penalty against physical reality.
Count correctly by building the inventory from the hypervisor layer itself, then subtracting hosts that should not be licensed: decommissioned clusters, powered off capacity, and workloads scheduled for migration. The defensible count is almost always lower than the quote's count.
Decommissioning records, migration project plans with dates, and cluster level utilization exports. An assertion that hosts are going away is not evidence; a dated plan is.
Three moves reduce billable cores: consolidating workloads onto fewer, denser hosts, refreshing hardware toward 16 plus cores per socket, and shrinking the licensed estate by moving portable workloads to another platform. All three work because the meter is physical, not virtual.
Consolidation programs in our 2024 to 2025 benchmarks cut billable cores 20 to 35 percent at constant workload. The savings recur every term, which makes the hardware case self funding.
The standard advice is to accept the core count in the renewal quote and negotiate the rate per core. We disagree. In roughly 20 of the 30 plus VMware estates Morten Andersen benchmarked in 2024 to 2025, the count itself was 10 to 25 percent too high, and no achievable rate discount matched the value of correcting it. The buyer side move is to rebuild the count from vCenter, evidence the exclusions, and only then discuss rate. Negotiating the rate on an inflated count is paying a discount on cores that do not exist.
Three cuts of our advisory engagement file frame the size of the opportunity.
Source: Redress Compliance advisory engagement file, 2024 to 2025.
Five moves turn this analysis into a lower invoice on the next renewal.
White Paper · Broadcom / VMware
Broadcom VMware Renewal Survival 2026
The 2026 buyer side reference on Broadcom VMware renewals. Read it free.
VMware subscriptions bill per physical core across every licensed host, with a 16 core minimum per CPU socket. Virtual CPUs and VM counts are irrelevant to the meter; only the physical inventory matters.
Every processor socket bills at least 16 cores even if the chip has fewer. A dual 8 core host is billed as 32 cores, a 100 percent penalty, which is why small host estates structurally overpay.
Yes. Any host running the licensed software counts, including management clusters, DR capacity, and idle hosts. Unlicensed cold standby is possible only when the software is genuinely not installed and running.
No, and that is the lever. More VMs per host cost nothing extra, so consolidation onto fewer, denser hosts directly strands hosts for decommissioning and cuts billable cores 20 to 35 percent in typical programs.
Yes, and it should be. First pass counts ran 10 to 25 percent above the defensible number in our benchmarks. Rebuild the count from vCenter, document exclusions with dated evidence, and table your number first.
Fewer sockets with 16 or more cores each. Refreshing four small two socket hosts into two dense ones holds capacity constant while halving exposure to the per socket minimum.
The core counting workbook, minimum penalty model, and consolidation math from 25 plus estates.
Used across more than five hundred enterprise engagements. Independent. Buyer side. Built for procurement leaders running the next renewal cycle.
Negotiating the rate on an inflated count is paying a discount on cores that do not exist. Fix the count first.
500+ enterprise clients. 11 vendor practices. Industry recognized. One conversation can change what you pay for the next three years.
One buyer side briefing a week. Pricing moves, audit signals, and the levers that work. No vendor spin.
