Deciding on infrastructure: VMware, migration, data center
Three questions, one set of figures: what your current estate really costs, what an already optimized target will cost, what moving from one to the other costs. Provider tools deliver the expected format; we replace their assumptions with your real costs.
Three questions, one set of figures
- Optimized status quo: the current estate after reduction, renegotiation and rationalization, the honest reference for any comparison
- Rebalanced hybrid: what gains from moving moves, the rest is optimized on site
- Full migration: every workload to the target, wave by wave
- Explicit assumptions: dual running, Windows, SQL Server, Oracle and VMware licenses, egress, managed services
- Migration credits and commitment discounts as separate lines, to be confirmed by the provider; the scenario without credits is always shown
Eight steps, usually six to eight weeks
Durations are indicative and set at scoping according to the size of the estate; several steps overlap.
Frame the decision
In a workshop with your CIO office and your finance department: the question to settle (full exit, datacenter only, mixed, VMware renewal), scope and explicit exclusions, targets, currency, horizon, decision rules (discount rate, capex and opex, cash or depreciation). For a VMware exit, this week of scoping includes reading the renewal quote, the RVTools inventory and the choice of scenarios. One owner named per source.
Collect
Exports and contracts as they are, with no reworking on your side: hypervisor inventory, CMDB of physical servers, 95th percentile of CPU and memory over 30 days, databases, application-to-server mapping; first the hypervisor renewal quote, then fixed assets, maintenance, hosting, licenses, backup.
Qualify every machine
Sources matched on normalized machine name, count differences explained before choosing a reference source. Each machine receives a disposition: migrate, retain on site, replace with a managed service, retire, exclude or to be qualified.
Rebuild the current cost
Full cost by platform, through cost pools allocated to machines, then the do-nothing trajectory: indexed opex, hypervisor renewal at its date, hardware refresh by cohort, end of support. Your own costs, when you provide them, take precedence.
Cost an optimized target
Each machine sized on its measured peaks, in the family that matches its profile; non-production costed on usage; license included or brought according to actual eligibility; databases to managed services; storage on used space, never on provisioned space.
Cost the transition
Waves consistent with your integrator's plan, dual running recomputed line by line until decommissioning, refresh frozen as soon as a segment migrates, migration cost per server, per database and as fixed program costs.
Scenarios and reconciliation
A series of scenarios and their sensitivities, bounded by a favorable and an unfavorable combination, then a table that explains the gap with every estimate in circulation: provider tools, integrator reports, benchmarks.
Read out and decide
Four or five key figures and three messages to your CIO office and finance department: the dominant variable of the current estate, the FinOps levers of the target, program discipline.
The calculation, explained simply
- Stay = indexed run cost (about 3% a year by default) + hardware renewal by purchase cohort + hypervisor at its renewal price, from its date.
- Migrate = steady-state target × wave ramp-up + support + migration cost + dual running (what each segment still costs until it is decommissioned) + retained estate, minus credits and discounts.
- Saving per period = stay minus migrate. We track the cumulative saving, the net saving over the horizon and the steady-state saving; ROI = net saving / migration cost net of credits; NPV at your finance department's rate; payback period: the first one in which the cumulative total becomes durably positive.
- The target: required resource = current resource × 95th percentile peak / 0.8, rounded to instance sizes; storage = used space + 20%. Without a measurement, a documented default peak is flagged as an assumption to lift. Public prices of the target region, excluding tax, 3-year commitment with no upfront payment as the base; 1-year commitment and pay-as-you-go as scenarios; cloud price changes set to zero for prudence.
- The same VM model on bare metal: the number of nodes is set by the strongest of three constraints. CPU: vCPUs to host compared with the node's cores × vCPU-to-core ratio, minus the resources the platform reserves on each node; this ratio is set per environment, more cautious in production, and validated with the vendor. Memory: after that reservation and the target utilization rate. Storage: consumed capacity after data reduction and replication factor. A resilience node is added in production.
- The current estate: cost pools (annualized hardware and maintenance, hypervisor, hosting and energy, storage, network, licenses, backup) allocated to machines pro rata to vCPUs, with a check: the sum of the allocations equals the total. Staff costs stay outside the base infrastructure TCO, so that the comparison remains indisputable; a full-cost section is offered as an option.
- The renewal is counted in licensable cores per host, with the minimums per processor and the other contract minimums applied, as the actual quote sets them, after removing stopped VMs and useless replicas.
- Hardware is shown under two conventions, cash at refresh dates or smoothed depreciation, with the effect of each. Savings are broken down by component: compute, storage, databases, network, support.
A distribution, not a figure. Each scenario is the same model with a modified set of parameters: hypervisor renewal, refresh cycle, renewal freeze, decommissioning lead time, commitment model, target without right-sizing, license brought, scope, no credits, full cost with staff.
The criteria presented to the committee: steady-state saving, net saving and ROI, NPV, payback period, gap between the favorable and unfavorable combinations, share of the saving that still rests on benchmarks.
Five outcomes, a reference and their combination
Same scope, same 5 to 6-year horizon, same full cost, wave by wave and dual running included: each path is a set of parameters of the same model.
Optimized status quo
Same platform, base reduced before the renewal: right-sizing, host consolidation, removal of stopped VMs and useless replicas. Always computed: without it, any alternative wins on paper.
- To model: the same items as the renewal, on the reduced base, plus the capacity management effort
- Decisive data: CPU and memory metrics, achievable consolidation rate, VMs to retire
Stay and renegotiate
Nothing changes for operations; the contract moves to a per-core subscription, VCF or VVF edition.
- To model: subscription per physical core, with the minimums per processor and the other contract minimums checked on the actual quote; imposed edition and bundles; hardware, hosting, storage, licenses, backup and network
- Decisive data: actual renewal quote, licensable cores per host, contract and refresh expiry dates
- Usually holds: recent and dense estate, distant exit horizon, few skills on the alternatives
- Watch out for: a long commitment on an estate due to shrink, an edition richer than the need, late renewal penalties
Change hypervisor on site
Same datacenter, VM conversion, new operations tooling: for example Nutanix, Hyper-V or Azure Local, Proxmox VE, OpenShift Virtualization.
- To model: subscriptions per core or per node, hardware reusable or not, conversion and testing, training, compatible backup, monitoring and recovery, dual running during conversion
- Decisive data: compatibility of hardware and guest operating systems, support of applications certified by their vendors, in-house skills
- Usually holds: datacenter kept for other reasons (latency, industrial IT, long hosting contract), recent hardware
- Watch out for: conversion and requalification cost, maturity of the backup and recovery ecosystem, vendor support
Host the same VM model at a hyperscaler
Exit from the datacenter without redesigning the VMs, operations tools kept: for example Azure VMware Solution, Google Cloud VMware Engine or Amazon EVS with portable VCF licensing, or an HCI solution on bare metal instances.
- To model: nodes sized by the strongest of three constraints (CPU, memory, storage), with resilience; hypervisor licenses; pay-as-you-go, 1-year or 3-year commitment compared per environment; network, dedicated links and backup
- Decisive data: VM inventory and consumed storage, hourly profile of each environment, regional prices and quotes
- Usually holds: constraining hosting deadline, transitional step before a modernization
- Watch out for: licenses that can exceed the infrastructure under a long commitment, a 24/7 commitment applied to non-production, instance availability in the region
Migrate to the public cloud
VMs converted into native instances sized on usage (rehost, replatform), databases to managed services.
- To model: optimized target (measured peaks, suitable families, Start & Stop), licenses included or brought, storage on used space, backup, recovery, network, support; transition (waves, dual running, migration, credits)
- Decisive data: utilization metrics, license eligibility, wave plan, integrator quote
- Usually holds: estate due for renewal, large share of non-production, will to modernize, optimized and committed target
- Watch out for: a like-for-like or on-demand target more expensive than the current estate, storage and databases often more expensive, underestimated dual running
Decommission or replace with a service
Stopped VMs, replicas and tooling retired; directory, files or software packages moved to a managed service or SaaS. Always to handle first: it reduces the base of every other path.
- To model: avoided cost, service subscription, data migration
- Decisive data: inventory dispositions, vendor roadmaps
- Watch out for: a scope moved to SaaS appears neither in the do-nothing cost nor in the target
Partial or mixed retention
Some segments stay on site (industrial IT, remote sites, appliances, Unix), the rest follows another path.
- To model: unavoidable costs kept identical in every scenario, fixed costs spread over a smaller base
- Decisive data: segments where the target costs more than the allocated current cost, latency and compliance constraints
- Usually holds: small depreciated sites, segments not worth moving
- Watch out for: residual fixed costs to reassign, lasting dual operations tooling
The destinations named are examples, compared as scenarios on the same criteria. Already quantified in the field: a migration to the public cloud and an HCI solution hosted on bare metal instances; Proxmox, OpenShift Virtualization, hosted VMware offers and sovereign clouds are compared as scenarios, with no field quantification to date. The alternative on-site hypervisor is costed on a generic model, not yet proven in the field.
Financial first, then qualitative
The reading is done on these criteria, with your teams; the qualitative criteria and their weight are set with you at scoping. A segment where the target costs more than the allocated current cost is presented as such: retention, consolidation or a simplification decision outside the TCO.
| Criterion | How we measure it | Reading |
|---|---|---|
| Steady-state saving | Annual cost of the path once the transition is over, compared with the do-nothing trajectory | Robustness of the target |
| Net saving over the horizon and NPV | Cumulative cash-flow gaps over 5 to 6 years, discounted at your finance department's rate | Depends mostly on the speed of the exit |
| Payback period | First period in which the cumulative total becomes durably positive | To compare with contract expiry dates |
| Robustness | Number of scenarios that stay positive, gap between the favorable and unfavorable combinations | A saving that only survives the favorable scenario is not one |
| Maturity of the figures | Share of the saving that still rests on benchmarks rather than on your contracts | Sets the order of the data request |
| Calendar | Compatibility with the renewal date, notice periods and migration capacity | Conditions the length of dual running |
| Migration risk | Volume and complexity of VMs, certified applications, dependencies, switchover windows | Qualitative, criterion set with you at scoping |
| Skills and operations | Gap between the skills in place and those required, tooling to replace | Qualitative, criterion set with you at scoping |
| Reversibility and lock-in | Commitment term, portability of licenses and VMs, exit cost | Qualitative, criterion set with you at scoping |
| Sovereignty and compliance | Data location, legal regime of the provider, regulatory requirements | Possible knock-out criterion, to set at scoping |
The qualitative criteria are a method proposal: their list and their weight are set with you at scoping.
One set of figures, several readers
Consolidated inventory
One row per machine (sources, characteristics, disposition, target, costs) and the funnel from total to active to in-scope inventory.
Full cost of the current estate
By platform, by application and by machine, with the do-nothing trajectory over the horizon.
Configurable model
Recomputable in your spreadsheet, with its frozen scenarios, its reconciliation table and the register of assumptions ranked by impact on the saving.
Readout in the expected format
For your committee and for the cloud provider, fed with your real costs rather than default assumptions.
CIO and CFO note
The decision in a few pages, with the two-page data request and the assumptions still to be lifted.
Model on your actual costs
A short iteration replaces the benchmarks with your invoices and contracts, in a workshop with your procurement and financial control: base scenario and negotiation scenario.
Options: a full-cost section with staff costs, vendor negotiation, tracking of savings by wave after the decision.
What KeeFinops does, what your teams do
| Topic | KeeFinops | Your teams |
|---|---|---|
| Scoping | Proposes the scenarios and the data request | One sponsor in IT and one in finance, one owner per source, the list of applications and constraints |
| Collection | Checks, completes and flags the gaps | Inventory exports, contracts, invoices, as they are |
| Base reduction | Analyzes usage and quantifies right-sizing | Application owners validate |
| Modeling | Builds the TCO and documents every assumption | Validate the business assumptions: criticality, grouping into waves |
| Comparison | Quantifies the outcomes at the same level | Validate criticality, constraints and switch windows |
| Readout | Presents the note and the business case | The committee decides |
| After the decision | Tracking of savings by wave and guided execution, as options | Run the migration with the integrator |
What skews a comparison
- Comparing the alternatives with a status quo that is not optimized: they all win on paper.
- An inventory several months old: machines may have been created or retired since, and stopped machines and templates remain to be confirmed.
- Host, core and machine counts that differ between tools: they must be reconciled, never added up.
- Costing the renewal from a public summary or an unchecked licensing rule, instead of the actual quote, or counting licensable cores without applying the minimums per processor and the other contract minimums, or without first removing stopped VMs and useless replicas.
- Owned licenses valued at subscription price, or databases embedded in a software package counted as direct licenses.
- Third-party estimates that do not license SQL Server, compare different instance generations or regions, or sometimes include staff.
- HCI and local disks counted a second time as storage, or storage valued on provisioned rather than used space.
- A like-for-like or on-demand target, often more expensive than the current estate: without right-sizing, commitments and suitable families, a mass migration is not justified; likewise, a commitment billed 24/7 applied to non-production environments switched off at night and at weekends, for which pay-as-you-go is often cheaper.
- A long hypervisor commitment signed during the migration, on an estate due to shrink: better to negotiate one year or a transitional extension. On cloud bare metal, hypervisor licenses can exceed the infrastructure cost under a long commitment.
- Dual running that weighs more than the migration itself, underestimated if the hardware refresh is not frozen in migrating segments or if decommissioning slips; conversion, application requalification, training and compatible tooling (backup, monitoring, recovery) are often underestimated too.
- Segments that are not worth moving (small depreciated sites, infrastructure already paid for), to be presented as a retention or simplification decision, not as a TCO gain; their unavoidable costs, such as the local network of the sites kept, appear identically in every path.
- A cost line annotated or excluded without explanation in an extract: to be clarified with its author, because it can reverse an item.
Frequently asked questions
How long does it take?
Usually 6 to 8 weeks for a first reasoned decision, depending on the size of the estate and the availability of data, then 3 to 4 weeks to replace the benchmarks with your invoices and contracts. The exact duration is set at scoping.
What do you do with the migration credits offered by the provider?
They appear as separate lines, to be confirmed by the provider, and the scenario without credits is always shown: a business case that only holds thanks to credits is not one.
Does your engine read VMware?
No. The KeeFinops engine reads public clouds; for VMware, the tool is the engagement full cost model, fed with RVTools, your contracts and your invoices.
Do you compare Proxmox, OpenShift Virtualization and sovereign clouds?
Yes, as comparison scenarios, on the same criteria as the other outcomes: licenses, hardware, skills, migration cost, and sovereignty can become a knock-out criterion set at scoping. Only a migration to the public cloud and an HCI solution hosted on bare metal instances have already been quantified in the field; Proxmox, OpenShift Virtualization, hosted VMware offers and sovereign clouds have no field quantification yet.
Your VMware quote deserves a second opinion
Thirty minutes to scope the perimeter, the size of the estate and the decision deadline.