Subscription per node, billed annually, with every capability included. No control-plane edition, no per-vCPU meter, no charge for the workloads you run. Below this page there is a cost model with every assumption exposed so you can rebuild the argument in your own finance meeting.
Node counts refer to production hypervisor nodes, not virtual machines. There is no cap on workload count at any tier — the fabric does not meter what you run on it.
Virtualisation, live migration, the AI fabric, the full MCP endpoint, the network and backup appliances, the console, the CLI, and the ability to run commercial workloads. No edition gates.
Where most of your estate is GPU nodes, the AI fabric is packaged per accelerator node at $940/node/year. This is a packaging choice — the capability is identical, it is simply metered against the nodes that carry it.
Business hours is included from Standard. 24×7 with a 4-hour response adds $620/node/year; 24×7 with a 1-hour response adds $1,280/node/year.
Professional services are quoted separately — typically $1,850 per engineer-day. A first deployment is usually 8–15 days depending on node count and integration complexity. This is not a margin grab; it is the difference between a working cluster and a demo that falls over on day two.
Every assumption behind this model is stated below the calculator. Drag the inputs. If our arithmetic is wrong, the numbers will show it — and we would rather you found that here than in a procurement review.
This is a model, not a quote. It assumes a converged stack (Kubernetes + virtualisation) or a commercial hypervisor as the alternative. If your current platform is already cheaper to run than these assumptions, the conclusion changes — and we would rather you tell us that than discover it after signing. Send us your actuals and we will redo the model with your numbers, including the ones that make us look bad.
So you can challenge any line of it. These are conservative planning figures, not published list prices, and they should be replaced with your procurement data.
| Assumption | Value used | Basis |
|---|---|---|
| Control-plane RAM per host | ~2 GB fixed | Does not scale with fleet size — the structural advantage. |
| Alternative per-vCPU licence | $145 / vCPU / year | Mid-band enterprise hypervisor list pricing with support. Enterprise-plus suites sit materially higher. |
| Alternative per-host charge | $900 / node / year | Typical management-plane infrastructure component. |
| Alternative control-plane hardware | 3 nodes · 8 vCPU · 32 GB · $8,600 each | Quorum control plane, amortised over a 3-year refresh within your horizon. |
| Control-plane maintenance | 18% of capex per year | Hardware maintenance only. Platform headcount is counted once, in the staffing line below — never twice. |
| Fully-loaded platform engineer | $175,000 / year | Salary, benefits and overhead. |
| Platform staffing — stateful alternative | 1 engineer per 250 VMs, plus 1 per 40 AI workloads | Effort scales with workload count, because each workload can fail independently and needs managing. |
| Platform staffing — Unimatrix0 | 1 engineer per 24 bare-metal nodes, plus 1 per 10 GPU nodes, rounded up to a minimum of 0.5 FTE | Effort scales with node count instead. Failover is automatic, nodes are disposable, and there is no control-plane database to repair. |
| Power released | $0.19 / kW-month delivery + $0.11 cooling | Reserved memory and idle management nodes are what stop drawing power. |
| One-time services | 8 + 0.9 days per node, at $1,850/day | Design, first node, validation. Capped at 60 days. |
| Migration effort | 5 hours per node, rebuild + verify | Estimated per node at $165/hour fully loaded. |
| GPU fabric packaging | $940 / accelerator node / year | Charged against the nodes that carry the AI fabric. |
Hardware capital cost, because you own the hardware either way. Contract exit costs from your current vendor. Migration of data at volume. Opportunity cost of your engineers' attention during the change.
Very small estates with low workload counts may not see a positive return inside one year. If that is your situation, we will say so rather than sell you a three-year story.
GPU-dense inference estates, where memory recovery and fast weight delivery compound. Regulated deployments where the audit and residency story changes the conversation. Consolidating a second platform onto one fabric.
Indicative pricing. Figures shown are planning-grade and subject to a signed order form. Currency is USD, exclusive of tax. Multi-year terms and volume beyond the Enterprise tier are quoted directly.
Node count, core count, workload mix, your current licence spend, your current platform team. We will build the model with your figures — including where it argues against us.