What is the Broadcom VMware price increase?

It is the effect of Broadcom's restructuring of VMware licensing after the acquisition: the product catalog was collapsed into a small number of bundles, perpetual licences were eliminated in favour of subscription, and a core minimum applies per CPU socket. Those three mechanics, not a single published percentage, are what move a renewal quote. We do not publish a multiple, because there isn't an honest one — the increase a given mid-market estate sees depends on how many sockets it runs, how many cores are in each of them, and how much of the new bundle it will actually use. An estate on older, lower-core-count CPUs typically absorbs a much larger increase than a newer one, because the per-socket core minimum charges it for capacity it does not have. The number that settles it is your own renewal quote set against your real consumption, and that comparison takes about an hour to build.

Why does memory pricing belong in a VMware exit business case?

Because a VMware exit usually means new hypervisor hosts, and a host is mostly memory. Memory is the largest single line in a typical host bill of materials, storage sits directly beside it, and both are priced off the same DRAM and NAND markets that moved sharply through 2026. If your business case models the licensing change and treats the hardware as a flat, known number carried over from a quote written a year ago, the hardware line is the one that will be wrong. This is not an argument for or against any particular destination platform — it is an argument for repricing the hardware at the same time you reprice the licences, rather than assuming last year's per-host figure still holds.

Are server memory price increases real, or is that a consumer PC story?

Both, and the distinction matters because the reporting blurs them. On the server side specifically, TrendForce forecast on July 9, 2026 that server DRAM contract prices would rise 13–18% quarter-on-quarter in 3Q26. That is a forecast. The reported actuals behind it are larger: TrendForce reported on June 1, 2026 that 1Q26 DRAM industry revenue rose roughly 81% quarter-on-quarter to about $97 billion, with conventional DRAM contract prices up 93–98% quarter-on-quarter, and on August 18, 2026 that combined revenue of the top five NAND Flash brands rose 77% quarter-on-quarter to US$68.87 billion in 2Q26. Those are reported results, not projections, which means the increase is already in the channel rather than approaching it. Consumer PC coverage is a separate strand: IDC forecast on June 2, 2026 that 2026 global PC shipments would fall 11.3% while average selling prices rose about 17%. IDC does not split commercial from consumer in that figure, so it should not be read as a business-PC number.

Should we defer a hardware refresh until memory prices come down?

Deferral is a bet, and it is currently a bet against the only published forecast either major analyst house has issued. IDC forecast on June 2, 2026 that the memory shortage persists through 2027, with Q4 2026 PC shipments down about 20% year-on-year. If that forecast is right, waiting eighteen months buys higher prices and less runway, not relief. There are still good reasons to sequence a refresh — a support date, a lease boundary, a project you cannot staff this quarter — but 'prices will be better later' is not one you can currently point at evidence for. Plan the timing around your own constraints and treat any assumed price relief as unfunded.

Does the memory market make staying on VMware the cheaper option?

No, and it is worth being explicit about that, because it is the wrong conclusion to draw from the same facts. Hardware eventually has to be replaced on either path — an estate that stays on VMware still refreshes hosts and still buys memory at the prevailing market price, and it does so while carrying the restructured subscription licensing as well. What the memory market changes is the timing question, not the destination question: it argues for pricing the hardware now rather than assuming a stale figure, and for deciding whether the refresh and the renewal should land in the same budget cycle or be deliberately separated. Both options carry the memory cost. Only one of them also carries the renewal.

What is changing in Microsoft licensing on October 1, 2026?

Microsoft announced on August 12, 2026 that starting October 1, 2026 it applies a 5% cost of capital uplift for Cloud Solution Provider (CSP) software subscriptions with annual-term commitments billed monthly. The examples Microsoft names are SQL Server, Windows Server, Client Access Licenses and System Center — server and infrastructure licensing rather than per-user Microsoft 365 suites, which changed separately on July 1, 2026. For an existing annual-term subscription billed monthly, the uplift takes effect at renewal on or after October 1, 2026, so an in-term subscription is not repriced mid-cycle. Microsoft also published a correction alongside the announcement stating that an earlier communication carried an incorrect effective date; October 1, 2026 is the correct one, so a quote citing a different date predates the correction. Source: Microsoft Partner Center announcements, August 2026.

Does the 5% CSP uplift apply to annual billing or month-to-month subscriptions?

No. Microsoft states there is no change to annual billing or to month-to-month subscriptions. The 5% cost of capital uplift effective October 1, 2026 applies specifically to CSP software subscriptions that carry an annual-term commitment billed monthly — the shape where the customer commits for a year but pays in twelve instalments. That is a narrow trigger, and it is worth checking rather than assuming: plenty of organisations that believe they are on monthly billing are actually on an annual term paid monthly, which is exactly the case the uplift covers. Pro IT NW is not a CSP and does not resell Microsoft licensing; check the billing shape on your own agreement or ask whoever issues your licence invoices.

When does VMware vSphere 8 support end?

vSphere 8 reaches End of General Support on 11 October 2027. That is the date worth putting in a refresh plan, because it is the point after which running the platform becomes a risk decision rather than a budget decision. Working backwards from it is a useful discipline: a mid-market migration of any size needs discovery, a destination decision, a procurement window for hardware, and a cutover sequence with hypercare after it — and the procurement window is the part that has become least predictable, because it depends on a memory market that moved sharply through 2026.

In what order should we make the licensing, hardware and Microsoft licensing decisions?

Model the licensing metric first, price the hardware second, and check the Microsoft renewal boundary third — then decide whether all three should land in one budget cycle. The licensing metric comes first because it is what usually moved the bill and it is knowable from your existing quote and core counts. Hardware comes second because the destination platform changes what you buy, and because a per-host figure older than a few months should be treated as expired rather than approximate. The Microsoft renewal boundary comes third because it is a date, not a decision — for CSP software subscriptions on an annual term billed monthly, the 5% uplift lands at renewal on or after October 1, 2026, and knowing where that boundary falls tells you whether it belongs in this year's plan or next year's.

Field notes · 9 min read · August 19, 2026

What a 2026 infrastructure refresh actually costs

Two dates bracket this decision. Microsoft applies a 5% cost of capital uplift to CSP software subscriptions with annual-term commitments billed monthly — Windows Server, SQL Server, CALs and System Center — effective October 1, 2026, at renewal on or after that date. And VMware vSphere 8 reaches End of General Support on 11 October 2027. The memory market sits between them: TrendForce forecast on July 9, 2026 that server DRAM contract prices would rise 13–18% quarter-on-quarter in 3Q26.

If you are planning a VMware exit for the 2026–27 budget cycle, the number you are being asked for is usually a single one: what does the migration cost? The honest answer is that it is not one cost. It is three, and they are landing in the same cycle.

The first is licensing — the Broadcom VMware price increase, which is the one everybody names. The second is hardware, because a VMware exit generally means new hypervisor hosts, and a host is mostly memory. The third is Microsoft licensing, which most people are not tracking at all: a 5% cost of capital uplift on CSP software subscriptions effective October 1, 2026, covering Windows Server, SQL Server, Client Access Licenses and System Center.

The 60-second version: three separate cost movements are converging on the same budget line. None of them individually reframes a business case. Together they change which option wins, and they change it mostly by changing the timing question rather than the destination question. We are not going to multiply them together for you — nobody has published a figure for the combined effect, and inventing one would be worse than saying so.

Cost one: the Broadcom VMware price increase

This is the one that started the conversation, and it is the one that is hardest to give a number for honestly. Broadcom collapsed the VMware catalog into a small number of bundles, eliminated perpetual licences in favour of subscription, and applies a core minimum per CPU socket. Those three mechanics — not a headline percentage — are what decide where a given renewal lands.

The consequence is counterintuitive and it catches people out: a smaller, older estate frequently absorbs a larger increase than a bigger, newer one, because the per-socket core minimum charges it for cores it does not physically have. We walk through the mechanics, and what actually moves you within the range, in VMware after Broadcom: the six real exit destinations.

One date is worth writing down while you are here, because it is the only Broadcom lifecycle date we will state: vSphere 8 reaches End of General Support on 11 October 2027. That is the boundary you are planning against.

Cost two: the hosts — and a host is mostly memory

Here is the part most business cases carry over unchanged from a quote written a year ago. If your exit involves new hypervisor hosts, memory is the largest single line in the bill of materials, and the memory market has moved sharply.

Start with what has already happened, because these are reported results rather than projections. TrendForce reported on June 1, 2026 that 1Q26 DRAM industry revenue rose roughly 81% quarter-on-quarter to about $97 billion, with conventional DRAM contract prices up 93–98% quarter-on-quarter. On August 18, 2026 TrendForce reported that combined revenue of the top five NAND Flash brands rose 77% quarter-on-quarter to US$68.87 billion in 2Q26, with Micron up 99.2% quarter-on-quarter to $11.85 billion, taking third place from Kioxia. Those are actuals. The increase is already in the channel and in supplier results — it is not something arriving later.

Looking forward, TrendForce forecast on July 3, 2026 that 3Q26 conventional DRAM contract prices would rise 13–18% quarter-on-quarter and NAND Flash contract prices 10–15% quarter-on-quarter, moderating against a high base. In a separate note forecasting the server segment specifically, on July 9, 2026, TrendForce put server DRAM contract prices at +13–18% quarter-on-quarter in 3Q26. Both of those are forecasts, published on those dates, and we label them as such deliberately — anyone who actually buys memory will check.

The asymmetry is the argument, not the percentages. In that same July 9, 2026 note, TrendForce observed that multi-year long-term agreements signed by US cloud providers cap the increases for those customers. A hyperscaler signed a multi-year supply agreement. A mid-market firm buying three hosts signed nothing. That is why "the market is up, everyone is in the same boat" is not true, and why a per-host figure quoted to you last year should be treated as expired rather than approximate.

And storage sits right beside it. A hypervisor refresh is rarely just hosts — SAN or hyperconverged storage usually moves in the same project, and it is priced off the same DRAM and NAND markets. The NAND figures above are the storage half of the same story. If you were planning to reuse existing shared storage, that is now a materially more valuable decision than it looked twelve months ago; if you were planning vSAN-class parity on new hardware, price it fresh. We cover where that lands for one common destination in Azure Local as a VMware exit, and for the open-source route in what XCP-ng actually costs.

Deferral is not a strategy. IDC forecast on June 2, 2026 that the memory shortage persists through 2027 — alongside 2026 global PC shipments down 11.3% with average selling prices up about 17%, and Q4 2026 shipments around 20% lower year-on-year. (Those PC figures are IDC's, and IDC does not split commercial from consumer, so do not read a business-PC number out of them.) The point for a refresh plan is narrower: "we will wait for prices to come down" is a bet against the only published forecast either analyst house has issued.

For a sense of how far the mix has shifted, TrendForce published a modelled scenario on March 10, 2026 — five months old, and about a consumer notebook rather than a server — in which a mainstream US$900-MSRP notebook could rise about 40%, with memory plus CPU reaching roughly 58% of bill of materials from around 45%. Treat that as directional only. It is a model, not a quote, and it is not your host.

Cost three: Microsoft licensing on October 1, 2026

This is the one that is not on most refresh plans, and it lands on exactly the licences a server refresh touches.

Microsoft announced on August 12, 2026 that starting October 1, 2026 it applies a 5% cost of capital uplift for Cloud Solution Provider (CSP) software subscriptions — in Microsoft's wording, such as SQL Server, Windows Server, Client Access Licenses, and System Center — with annual-term commitments billed monthly. Microsoft states there is no change to annual billing or to month-to-month subscriptions, and that for existing annual-term subscriptions billed monthly the uplift takes effect at renewal on or after October 1, 2026.

Two practical notes. First, Microsoft published a correction alongside the announcement: an earlier communication gave the wrong effective date, and partners were told to disregard it. October 1, 2026 is the correct date, so a quote citing a different one predates the correction. Second, the trigger is narrow and frequently misread — it is the annual term billed monthly shape, not monthly billing generally. Plenty of organisations that believe they are month-to-month are on an annual commitment paid in instalments, which is precisely the case this covers.

Why it belongs in a refresh post: a hypervisor migration is usually also a Windows Server decision, and often a SQL Server one. We wrote up the wider Microsoft pricing picture, including the renewal-boundary rule, in the Microsoft 365 price increase and what the renewal boundary means; and the Windows Server end-of-support decision that frequently rides along with a hypervisor move in Windows Server 2016: ESU or modernize.

The three stack. We are not going to multiply them.

Here is the discipline we hold ourselves to, and you should hold your vendors to it as well. Licensing, hardware and Microsoft licensing are three independent movements arriving in one budget cycle. There is no published figure for their combined effect on a mid-market refresh, we have not seen one that survives scrutiny, and a number produced by multiplying three ranges together would be a sales artefact rather than an estimate.

What is defensible is the shape: neither the licensing change nor the hardware market reframes a business case on its own; together they change which option wins. A platform comparison built on a stale hardware number and a current licensing number is comparing two different years, and it will pick the wrong destination for a reason that has nothing to do with the platforms.

The one conclusion we will not let you draw: none of this is a reason to stay on VMware. Hardware gets replaced on either path — an estate that stays still refreshes hosts and still buys memory at the prevailing price, while also carrying the restructured subscription. Both options carry the memory cost. Only one of them also carries the renewal. What the memory market changes is when you buy, not what you move to.

How to sequence it

Model the licensing metric against your real core counts first. This is knowable today from your existing quote and an inventory, it is usually what moved the bill, and it costs about an hour. Do it before anyone builds a slide about platforms.
Reprice the hardware, do not carry it forward. Treat any per-host figure older than a couple of quarters as expired. Get current quotes for the memory configuration you actually intend to run, and get the storage quoted at the same time. Ask for the quote validity period in writing, and note it — in a market moving at this rate, how long a number stays good is part of the number, and a business case built on an expired quote is a business case that has to be rebuilt in front of your CFO.
Find your Microsoft renewal boundary. If any CSP software subscription is on an annual term billed monthly, the 5% uplift lands at its renewal on or after October 1, 2026. That tells you which budget year it belongs to.
Decide deliberately whether all three land together. Sometimes consolidating them into one cycle is right — one project, one procurement, one disruption. Sometimes separating the hardware refresh from the licensing decision is right. What is never right is discovering the answer after the quotes come in.
Work backwards from 11 October 2027. vSphere 8 End of General Support is the outer boundary, and the procurement window inside it is the least predictable part of the plan right now.

Sources

The 30-second version

A 2026 VMware exit is three costs, not one. The Broadcom VMware price increase is driven by bundling, the end of perpetual licences and a per-socket core minimum — which is why older, smaller estates often see the largest increases. New hosts are mostly memory, and memory moved: DRAM and NAND revenue figures for 1Q26 and 2Q26 are reported actuals, TrendForce forecasts a further 13–18% quarter-on-quarter rise in server DRAM contract prices for 3Q26, and hyperscalers capped their exposure with long-term agreements that a mid-market buyer does not have. IDC forecasts the shortage running through 2027, so deferral is a bet against the published view. And Microsoft applies a 5% cost of capital uplift to CSP software subscriptions with annual-term commitments billed monthly — Windows Server, SQL Server, CALs, System Center — at renewal on or after October 1, 2026. The three stack; we will not put a combined number on them. None of it is a reason to stay on VMware, because hardware gets replaced either way. It is a reason to price the hardware fresh and decide the timing on purpose.

If you want the core-count modelling, a current hardware picture and the destination decision done as one bounded piece of work, the project intake form takes about three minutes. Two-business-day response with scope and a fixed-fee range.

Pro IT NW handles VMware exit and server modernization for mid-market organizations in the Seattle area and across the Pacific Northwest. Vendor-neutral and labor-only: we do not resell hypervisors, hardware or Microsoft licensing, and we are not a CSP — so the recommendation in any specific case is not paying us anything extra either way. Pricing figures above are the analyst and vendor sources cited, not quotes we can hold open.

Questions we get asked

What is the Broadcom VMware price increase?: It is the effect of Broadcom's restructuring of VMware licensing after the acquisition: the product catalog was collapsed into a small number of bundles, perpetual licences were eliminated in favour of subscription, and a core minimum applies per CPU socket. Those three mechanics, not a single published percentage, are what move a renewal quote. We do not publish a multiple, because there isn't an honest one — the increase a given mid-market estate sees depends on how many sockets it runs, how many cores are in each of them, and how much of the new bundle it will actually use. An estate on older, lower-core-count CPUs typically absorbs a much larger increase than a newer one, because the per-socket core minimum charges it for capacity it does not have. The number that settles it is your own renewal quote set against your real consumption, and that comparison takes about an hour to build.
Why does memory pricing belong in a VMware exit business case?: Because a VMware exit usually means new hypervisor hosts, and a host is mostly memory. Memory is the largest single line in a typical host bill of materials, storage sits directly beside it, and both are priced off the same DRAM and NAND markets that moved sharply through 2026. If your business case models the licensing change and treats the hardware as a flat, known number carried over from a quote written a year ago, the hardware line is the one that will be wrong. This is not an argument for or against any particular destination platform — it is an argument for repricing the hardware at the same time you reprice the licences, rather than assuming last year's per-host figure still holds.
Are server memory price increases real, or is that a consumer PC story?: Both, and the distinction matters because the reporting blurs them. On the server side specifically, TrendForce forecast on July 9, 2026 that server DRAM contract prices would rise 13–18% quarter-on-quarter in 3Q26. That is a forecast. The reported actuals behind it are larger: TrendForce reported on June 1, 2026 that 1Q26 DRAM industry revenue rose roughly 81% quarter-on-quarter to about $97 billion, with conventional DRAM contract prices up 93–98% quarter-on-quarter, and on August 18, 2026 that combined revenue of the top five NAND Flash brands rose 77% quarter-on-quarter to US$68.87 billion in 2Q26. Those are reported results, not projections, which means the increase is already in the channel rather than approaching it. Consumer PC coverage is a separate strand: IDC forecast on June 2, 2026 that 2026 global PC shipments would fall 11.3% while average selling prices rose about 17%. IDC does not split commercial from consumer in that figure, so it should not be read as a business-PC number.
Should we defer a hardware refresh until memory prices come down?: Deferral is a bet, and it is currently a bet against the only published forecast either major analyst house has issued. IDC forecast on June 2, 2026 that the memory shortage persists through 2027, with Q4 2026 PC shipments down about 20% year-on-year. If that forecast is right, waiting eighteen months buys higher prices and less runway, not relief. There are still good reasons to sequence a refresh — a support date, a lease boundary, a project you cannot staff this quarter — but 'prices will be better later' is not one you can currently point at evidence for. Plan the timing around your own constraints and treat any assumed price relief as unfunded.
Does the memory market make staying on VMware the cheaper option?: No, and it is worth being explicit about that, because it is the wrong conclusion to draw from the same facts. Hardware eventually has to be replaced on either path — an estate that stays on VMware still refreshes hosts and still buys memory at the prevailing market price, and it does so while carrying the restructured subscription licensing as well. What the memory market changes is the timing question, not the destination question: it argues for pricing the hardware now rather than assuming a stale figure, and for deciding whether the refresh and the renewal should land in the same budget cycle or be deliberately separated. Both options carry the memory cost. Only one of them also carries the renewal.
What is changing in Microsoft licensing on October 1, 2026?: Microsoft announced on August 12, 2026 that starting October 1, 2026 it applies a 5% cost of capital uplift for Cloud Solution Provider (CSP) software subscriptions with annual-term commitments billed monthly. The examples Microsoft names are SQL Server, Windows Server, Client Access Licenses and System Center — server and infrastructure licensing rather than per-user Microsoft 365 suites, which changed separately on July 1, 2026. For an existing annual-term subscription billed monthly, the uplift takes effect at renewal on or after October 1, 2026, so an in-term subscription is not repriced mid-cycle. Microsoft also published a correction alongside the announcement stating that an earlier communication carried an incorrect effective date; October 1, 2026 is the correct one, so a quote citing a different date predates the correction. Source: Microsoft Partner Center announcements, August 2026.
Does the 5% CSP uplift apply to annual billing or month-to-month subscriptions?: No. Microsoft states there is no change to annual billing or to month-to-month subscriptions. The 5% cost of capital uplift effective October 1, 2026 applies specifically to CSP software subscriptions that carry an annual-term commitment billed monthly — the shape where the customer commits for a year but pays in twelve instalments. That is a narrow trigger, and it is worth checking rather than assuming: plenty of organisations that believe they are on monthly billing are actually on an annual term paid monthly, which is exactly the case the uplift covers. Pro IT NW is not a CSP and does not resell Microsoft licensing; check the billing shape on your own agreement or ask whoever issues your licence invoices.
When does VMware vSphere 8 support end?: vSphere 8 reaches End of General Support on 11 October 2027. That is the date worth putting in a refresh plan, because it is the point after which running the platform becomes a risk decision rather than a budget decision. Working backwards from it is a useful discipline: a mid-market migration of any size needs discovery, a destination decision, a procurement window for hardware, and a cutover sequence with hypercare after it — and the procurement window is the part that has become least predictable, because it depends on a memory market that moved sharply through 2026.
In what order should we make the licensing, hardware and Microsoft licensing decisions?: Model the licensing metric first, price the hardware second, and check the Microsoft renewal boundary third — then decide whether all three should land in one budget cycle. The licensing metric comes first because it is what usually moved the bill and it is knowable from your existing quote and core counts. Hardware comes second because the destination platform changes what you buy, and because a per-host figure older than a few months should be treated as expired rather than approximate. The Microsoft renewal boundary comes third because it is a date, not a decision — for CSP software subscriptions on an annual term billed monthly, the 5% uplift lands at renewal on or after October 1, 2026, and knowing where that boundary falls tells you whether it belongs in this year's plan or next year's.

Related service

VMware exit and server modernization

Written by the team at Pro IT NW · Senior-led Microsoft project consultancy · Seattle / USA-wide.

Share on LinkedIn Follow Pro IT NW