VM Platform Ledger Methodology: How Every Score Is Built

Part of the VM Platform Ledger (edition 2026-09-30). The results are in the overview and the platform reports.

Who this is for

The Ledger is written for the people who buy, run and resell virtualization platforms: IT teams choosing what comes after VMware, and the system integrators and service providers who build on these platforms for their customers. It is not written for vendors, and no vendor paid for, reviewed or was briefed on any score.

The analyst models most buyers know (Gartner's Magic Quadrant, the Forrester Wave, GigaOm's Radar, IDC MarketScape) score vision partly from vendor briefings, and none of them checks whether last year's vision actually shipped. The Ledger takes the opposite position. Every score comes from the public record: release notes, documentation, roadmaps, pricing pages, licence terms and source repositories. If a claim cannot be traced to a public page, it does not move a score.

What counts as a platform

A platform is in scope when it runs general-purpose server VMs on hardware the customer controls, with its own management plane, sold or supported for production. It also has to pass four tests:

At least one release in the last 12 months.
Public documentation or release notes, even if part of it sits behind a login.
Available outside one country or one hardware line. Platforms tied to one vendor's hardware are included and flagged.
Live migration, HA and a central manager are present.

Public cloud IaaS, desktop hypervisors, VDI-only products, container-only platforms and OpenStack distributions that add no VM features of their own are out of scope.

Engines and products

Some entries are open-source engines that vendors package into products. KubeVirt is the engine inside OpenShift Virtualization, SUSE Virtualization (Harvester) and Spectro Cloud VMO; oVirt sits inside Oracle OLVM; OpenStack inside Platform9; Xen inside XCP-ng and XenServer. Nobody buys a support contract for the engine itself, so engines are scored as projects: they get no Commercial risk score (shown as n/a), and each product report names its engine.

The dimensions

Each platform is scored on seven dimensions. They are kept apart on purpose: a mature platform with few headlines and a young one shipping fast would cancel each other out in a single blended number.

Dimension	Question it answers	Scale	Better
Capability	What can it do today?	0 to 100	Higher
Momentum	How fast has it improved since the end of 2022?	Points gained	Higher
Credibility	Does it ship what it announces, when it says it will?	0 to 25	Higher
Commercial risk	What can change in price, licence or ownership after you sign?	0 to 25	Lower
Technical lock-in	How hard is it to leave?	0 to 25	Lower
Integrator fit	How well can a partner build a business on it?	0 to 20	Higher
AI leverage	How well does it run AI workloads and use AI itself?	0 to 20	Higher

A further reading, Direction, describes where the engineering effort goes and what pulls it. It is never scored, because a vision-led roadmap is not worse than a customer-led one. It only becomes a problem when credibility is low.

Capability

Capability is scored in ten domains, each from 0 to 5, against fixed anchors. Half points are allowed. Only generally available features count; a technology preview counts as half a step at most.

Score	Meaning
0	Absent
1	Rudimentary or preview; not usable in production
2	Basic; works for simple cases, with clear gaps
3	Solid; covers what most mid-size deployments need
4	Strong; enterprise depth, few gaps
5	Reference; the implementation others are measured against that year

The domains are weighted to a total of 100. The anchors for a 3 and a 5:

Domain (weight)	3 means	5 means
Compute core (15)	Live migration, HA restart, hot-add CPU and RAM, snapshots, templates	Plus automated load balancing, fault tolerance or equivalent, NUMA control, large-VM limits, mixed CPU generations, Arm
Storage (12)	Shared storage over NFS, iSCSI or FC with snapshots, or built-in HCI storage	Both HCI and external arrays, storage live migration, policy-based management, stretched storage, NVMe/TCP, data reduction
Networking / SDN (10)	VLANs, a distributed or virtual switch, bonding	Overlay SDN, distributed firewall, VPCs, load balancer, BGP and EVPN routing
Data protection / DR (12)	Snapshots, backup with changed-block tracking, async replication	Plus sync or metro replication, orchestrated failover, broad backup-vendor support, immutability
Migration-in (8)	A supported import tool from at least VMware, cold or warm	In-place or near-zero-downtime conversion at scale from several sources, with network and storage mapping
GPU / AI (10)	GPU passthrough and vGPU	vGPU live migration, GPU scheduling, an integrated AI stack, DPU support
Kubernetes (8)	CSI and a cloud provider or Cluster API, so Kubernetes runs well on top	An integrated Kubernetes service; VMs and containers under one control plane
Security / confidential (8)	Secure Boot, vTPM, RBAC, encryption at rest	Confidential VMs (SEV-SNP or TDX), attestation, encrypted live migration, hardening profiles, live patching
Fleet / multi-site (9)	A multi-cluster view, two-node or stretched clusters	Central lifecycle for many clusters, cross-cluster live migration, edge-scale fleet tooling, disconnected operation
Automation / ecosystem (8)	A full REST API, Terraform and Ansible providers	Plus a self-service portal, multi-tenancy, a large integration catalogue, broad certified hardware

Capability total = the sum of (domain score ÷ 5 × weight).

Scored by year since 2015

Capability is scored for every year from 2015 to today, each change tied to the release that caused it. 2015 is when the modern field takes shape: Nutanix AHV ships, vSAN and NSX are established, and KubeVirt follows in 2016. Earlier years (2001 to 2014) appear in the Ledger as milestones only.

History is judged against today's fixed anchors, not against what was state of the art at the time. A line only rises when something ships, and it only falls when a capability is removed or deprecated.

Momentum is capability today minus capability at the end of 2022. The reports show the 2015 to 2022 change beside it for context.

Credibility

Credibility measures delivery: five criteria from 0 to 5, total 25.

Criterion	1	3	5
Cadence	Irregular, undated	Regular but unpublished	Published rhythm, kept
Roadmap transparency	Keynotes only, docs behind login	Public release notes, direction in blogs	Written public roadmap, open tracker
Say-do	Under 40% of announcements GA within 12 months	40 to 70%	Over 85%, dates hold
Velocity	Maintenance and rebases	One or two substantive capabilities per release	Several per release, sustained
Lifecycle stability	Short windows, forced migrations, renames	Predictable, some churn	Long support, clean in-place upgrades

Buyer risk, in two scores

Risk is split because the two halves call for different responses. Commercial risk is negotiated; lock-in is engineered around.

Commercial risk (0 to 25, higher is riskier):

Criterion	0	5
Price and licence volatility since 2023	No changes	Repeated or drastic changes to metric, minimums or bundles
Purchase constraints	Any quantity, any edition, perpetual or free option	Forced bundle, high minimums, subscription only
Channel and access	Open download, open partner programme	Invite-only partners, restricted access to patches
Owner stability	Stable independent owner or foundation	Recent acquisition, private-equity owner or strategy reversal
Cost of staying supported	Long windows, free updates	Forced upgrade cadence, paid features that used to be free

Technical lock-in (0 to 25, higher is more locked in):

Criterion	0	5
Formats	Open disk and config formats	Proprietary, conversion needed
Export path	Documented, tooling for many targets	None, or hostile
Stack coupling	Any storage, any network	Own storage and network stack required
Hardware and cloud coupling	Any x86 server, fully offline	Validated hardware only, mandatory cloud attach
Skills and tooling	Standard Linux and KVM skills, open APIs	Proprietary tooling, scarce skills

Integrator fit

Four criteria from 0 to 5, total 20: partner programme openness, multi-tenancy and self-service, API and infrastructure-as-code depth, and the availability of skills and certification.

AI leverage

AI leverage looks wider than the GPU / AI capability domain. It covers four strands, each from 0 to 5, scored by year like capability:

Strand	1	3	5
AI workload infrastructure	Whole-GPU passthrough only	vGPU or MIG sharing with scheduling awareness	vGPU live migration, fine-grained sharing, current accelerators validated, GPU-to-GPU RDMA
AI platform stack	Reference architectures only	An integrated model-serving or AI stack GA on the platform	A full private AI platform GA: model catalogue, inference, vector DB and RAG, agent tooling, multi-tenancy
AI-assisted operations	Rule-based analytics	ML forecasting or anomaly detection (2), or a GA assistant for operators (3)	An assistant plus closed-loop remediation GA, AI-assisted migration
Agent openness	A documented API that agents can drive (1 to 2)	An official MCP server or agent interface in preview	A GA MCP server or agent gateway with identity, guardrails and audit

Each report also lists the AI say-do record: the AI items announced from 2023 to 2026, and how many reached GA, stayed in preview, are pending or were dropped. It separates AI that ships from AI that is only announced.

The Ledger Index

For readers who want a single number, the Ledger Index combines the dimensions on a 0 to 100 scale. It is always shown next to its parts. The balanced weights:

Part	Weight	Normalised as
Capability	35	capability ÷ 100
Credibility	20	credibility ÷ 25
Buyer safety	20	1 minus (commercial risk + lock-in) ÷ 50; for engines, 1 minus lock-in ÷ 25
AI leverage	10	AI leverage ÷ 20
Integrator fit	10	integrator fit ÷ 20
Momentum	5	capability gained since 2022, capped at 20 points, ÷ 20

The interactive Ledger also offers other weightings (risk-averse buyer, innovation-first, integrator) and buyer lenses that re-weight the capability domains, for example for a mid-market VMware exit, edge sites or AI infrastructure.

Why some scores start in 2023

Capability and AI leverage can be scored for any year, because the question is simply what had shipped. Credibility, commercial risk, lock-in and integrator fit cannot: they depend on announcements, prices and partner terms that are not consistently on record for earlier years, and today's values would be wrong for 2018. So those four start in 2023, the year Broadcom's acquisition of VMware closed, and the Ledger Index is computed only from 2023 onward. Views of earlier years show capability and AI leverage alone.

Direction and archetypes

Every release headline and roadmap item from 2023 to 2026 is tagged with one theme (compute, storage, networking, data protection, migration, AI infrastructure, AI-assisted operations, Kubernetes, security, fleet, ecosystem, packaging) and one driver:

Driver	How it shows in the evidence
Vision	First seen in a keynote or press release; a new category name; ships as a preview before users asked
Customer need	Traceable to forum threads, issues or support cases; removes a named limit; restores something users had
Market window	Exists because of a competitor event, such as import tools and free-year offers after the Broadcom changes
Upstream	Inherited from Kubernetes, KubeVirt, Debian, oVirt, Xen or OpenStack, on the upstream's calendar
Portfolio	Serves the vendor's other products or revenue: bundling, tier gating, mandatory cloud attach

The share of each driver is the platform's driver mix. The idea-to-need index is vision ÷ (vision + customer need + market window): 0 means purely demand-pulled, 1 purely thesis-pushed.

Combining the dominant driver with credibility gives each platform an archetype, for example "Dependable builder" or "Rent extractor". Archetypes are labelled opinions built on this table, not scores:

Dominant driver	Credibility high	Credibility low
Vision	Real innovator	Keynote theatre
Customer need	Dependable builder	Slow follower
Market window	Sharp opportunist	Promo without product
Portfolio	Integrated stack	Rent extraction

Market share and fame are deliberately not scored. No reliable public share data exists, and a well-known platform should not outscore a better one.

Evidence grades

Every report carries an evidence grade:

A: read from a primary source (release notes, documentation, roadmap revision history, pricing pages).
B: a secondary source, such as a vendor blog or search result.
C: inferred or unverified.

A report's overall grade is the lowest grade covering more than a quarter of its scores. Vendor claims are marked as vendor claims, and each report ends with open questions: the items that are still previews, unclear or due for re-checking.

Refreshes and the Say-Do Ledger

The Ledger is refreshed twice a year, in March after the spring release wave and in October after the autumn conferences. Each refresh re-scores the previous edition's "what is next" list for every vendor as shipped, slipped or dropped. Over time, that record of promises against delivery is the part of the Ledger no analyst firm publishes.

Corrections are welcome from anyone, vendors included, as long as they come with a public source. Corrections are logged in the next edition.

Data

The scores, the yearly history and every score change with the release that triggered it are free to download from the overview, as CSV and JSON.