The cost of a virtual desktop is a function of how your workforce operates, the infrastructure you run, the control you need to maintain, and how long you intend to run the platform. This page evaluates the total per-user cost of four delivery models, including foundation setup fees amortized across your enrollment.
Azure Virtual Desktop
When compute cost should track demand precisely. Dedicated session hosts within your environment scale in real time and charge only for active usage. Every performance tier is available on demand, from standard workstations to GPU-equipped machines, with native integration across Microsoft 365, Entra ID, and Intune. A higher foundation investment reflects the engineering depth required to deploy and orchestrate the platform correctly from day one.
Windows 365
When a flat monthly rate per user is the priority. Each user receives a dedicated Cloud PC, available on any device, with no infrastructure to size or manage. At lower enrollment counts and longer commitment horizons, no other model is more economical. For workforces with shift workers, contractors, or seasonal staff, Windows 365 Flex extends the model on a concurrent basis, licensing for active sessions rather than every enrolled user.
Self-hosted HCI — Proxmox
When the public cloud is not cost-effective, not suitable, or not yet viable. Each user runs their own dedicated virtual machine with exclusively allocated CPU, memory, and storage, on non-shared hardware. Proxmox is open-source, fully auditable, runs on commodity hardware, and carries no virtualization licensing fees.
Self-hosted HCI — Nutanix
When the organization requires a fully integrated, enterprise HCI stack with a single vendor accountable for hardware, software, and lifecycle management.
The Infrastructure Cost Comparison
The Architectural Math & Assumptions
Our financial model relies on conservative, defensible metrics. The projections are a direct function of the following hardware, operational, and setup constraints:
- Foundation Fees: Windows 365 deployments carry a $1,500 one-time foundation fee. Azure Virtual Desktop and all self-hosted options carry a $4,500 foundation fee reflecting the substantially greater orchestration and engineering involved. Both are amortized across the enrolled user base over the period selected by the above slider. At small scale and short horizons, this differential is the single most consequential number on the page.
- Cluster CapEx: $60,000 for a 3-node Proxmox HCI cluster (dual-socket nodes, ~256 GB RAM each, NVMe Ceph, redundant 25 GbE). Fully specced to carry 75 isolated desktops alongside existing server workloads with high-availability headroom. At 75 users, this translates to a dedicated per-user allocation of approximately 4 vCPU, 8 GB RAM, and 80 GB of NVMe-backed storage. These vCPUs are drawn from a dedicated physical core pool with a controlled oversubscription ratio, making each allocated core a more consistent compute unit than its cloud equivalent, where vCPUs are hyperthreads on shared multi-tenant hardware. On a performance-equivalent basis, this allocation compares more closely to the Windows 365 4 vCPU / 16 GB tier at $53 than to the 2 vCPU / 8 GB tier at $33, with no exact cloud equivalent at this price point. By comparison, the AVD multi-session baseline approaches $29/user at scale by hosting multiple users on a shared session host, typically a 4 to 8 vCPU / 16 to 32 GB Azure VM serving 6 to 10 concurrent users. Each user receives a session within a shared operating system instance rather than dedicated resources, with actual available compute varying with the load other users place on the same host.
- Investment Duration: 60 months (5 years), straight-line amortization, yielding $1,000/month for the entire cluster.
- VDI Allocation: 50% of the cluster is attributed to desktops ($500/month hardware, $1,500/month maintenance), while the remaining 50% supports core servers and structural headroom.
- Cluster Maintenance: $3,000/month for competent, proactive HCI stewardship covering hypervisor health, Ceph storage management, patching, backup validation, and incident response. The desktop workload carries half of that cost.
- Fixed & Variable Ops: Both self-hosted Proxmox paths carry approximately $300/month of fixed platform cost: the optional Proxmox subscription at roughly $185/month and a small allowance for monitoring tooling and platform overhead. Backup is handled by Proxmox Backup Server, which is open-source and carries no licensing cost. Variable expenses run $13/user, covering Windows E3/VDA licensing and per-user storage overhead on Ceph.
Notes:
The AVD Line: AVD combined with our orchestration approaches $29/user at scale but is not a fixed line. At low user counts, the minimum infrastructure required for availability, two always-on session hosts regardless of headcount, creates a fixed cost floor that makes AVD more expensive per user than Windows 365. The $3,000 setup fee differential adds further weight at low counts and short horizons. As headcount and tenure grow, both pressures dissolve and AVD's multi-session efficiency takes over. Crucially, AVD achieves its per-user efficiency by sacrificing resource isolation. Multiple users share a single operating system instance. AVD vCPUs are hyperthreads on shared Azure infrastructure; the per-user compute allocation on a multi-session host varies with concurrent demand.
The Windows 365 Line: The Windows 365 line reflects a 2 vCPU / 8 GB Cloud PC at $33 per user per month, the standard configuration for a knowledge worker on Microsoft 365 apps, browser, Teams, and Outlook. For that workload, the experience is fully adequate and will not feel like a compromise. Under sustained CPU pressure, such as large Excel models, heavy PDF processing, or multiple resource-intensive apps running simultaneously, this tier begins to show its limits, and the 4 vCPU / 16 GB configuration at approximately $53 per user is the more appropriate baseline. For full workstation-class requirements, an 8 vCPU / 32 GB Cloud PC runs $123 per user.
The Nutanix Line: Nutanix AOS Pro licensing adds approximately $2,250/month in software subscription for a 3-node cluster, on top of a hardware capex of roughly $80,000 and the same maintenance allocation. This raises the fully-loaded VDI fixed base to approximately $4,100/month. The exact crossover thresholds depend on the amortization period selected and are reflected dynamically in the chart and slider above.
Setup Fee Amortization: Foundation fees are spread across the enrolled user base over the selected commitment horizon, appearing as a declining per-user contribution that diminishes as headcount and tenure grow. At 5 users over 12 months, the Windows 365 setup fee adds $25/user/month and the AVD fee adds $75/user/month. At 50 users over 36 months, those contributions fall to under $1 and $2.50 respectively. This is why the commitment horizon matters so much at small scale, and why the slider above exists.
The Windows Enterprise Gate: Microsoft 365 Business Premium does not grant Virtual Desktop Access (VDA) rights for on-premises deployments. Operating sovereign desktops on private HCI mandates an upgrade to Microsoft 365 E3 or standalone VDA licensing, introducing a structural ~$10–$15 variable premium per user factored into the TCO math.
IT Operating Models: Where Does Your Investment Go?
The thresholds below describe the general shape of the economics calibrated to a 36-month horizon. At 12 months every crossover shifts rightward by 15 to 25 users. At 48 months they shift leftward by a similar margin. On a performance-equivalent basis, comparing self-hosted VMs against the 4 vCPU / 16 GB Cloud PC at $53, every self-hosted crossover shifts further leftward still. Use the slider above as your primary decision tool.
| User Count | The Strategic Verdict |
|---|---|
| Under ~20 Users | Windows 365 is the clear choice on every axis. Its $1,500 foundation fee is three times lower than any alternative, and at small headcounts that differential dominates all other considerations. On a performance-equivalent basis, self-hosted on an existing cluster becomes competitive from as few as 10 users. |
| ~20 to ~54 Users | The Tipping Point Band. Where the cluster already exists and its costs are shared across other workloads, Proxmox self-hosted dedicated VMs become more economical than Windows 365 from around 19 to 28 users depending on the commitment horizon. For cloud-only deployments, Windows 365 continues to beat AVD until approximately 51 to 72 users. This band is where the commitment horizon matters most: a 12-month view favors Windows 365 longer, while a 36 to 48-month view pulls both crossovers leftward. |
| ~54 to ~119 Users | AVD and Self-hosted Converge. AVD has crossed below Windows 365 and is now the more economical cloud option for workloads where resource isolation is not a requirement. For isolated desktops, Proxmox self-hosted on an existing cluster remains cheaper still. On a performance-equivalent basis, a new fully-loaded Proxmox cluster also crosses below Windows 365 in this band, from around 59 to 64 users. |
| Beyond ~119 Users | Sovereign Scale. Proxmox self-hosted infrastructure, fully loaded with hardware amortization, cluster maintenance, and setup fees, costs less per user than Windows 365 on any accounting basis and at any commitment horizon. Beyond approximately 144 users it undercuts the AVD baseline as well, delivering fully isolated desktops on dedicated hardware for less than the cost of shared cloud compute. Nutanix reaches those same thresholds at roughly 30 to 40 users later, reflecting the premium its software licensing carries. Each user's virtual machine runs on exclusively reserved CPU, memory, and storage. Zero resource contention. |
The Bottom Line: At small scale and short horizons, the setup fee differential alone makes Windows 365 the decisive choice before a single recurring cost is considered. As scale and tenure grow, that differential dissolves and the structural economics of each platform take over. The performance advantage of dedicated hardware, where each vCPU is drawn from a physical core pool under your control rather than a shared hyperthread on multi-tenant infrastructure, compounds the cost advantage at scale. The economics of self-hosted HCI are driven by fixed costs that must be shared broadly to justify the model. Where a cluster already serves multiple workloads, the desktop cost falls quickly and the crossover arrives early. Where the cluster exists for desktops alone, that crossover requires scale. In either case, the resource isolation, data sovereignty, and performance consistency of private infrastructure remain constants that cloud cannot replicate.