Windows 365 | Deployment guide

Windows 365 Enterprise

Deploy Cloud PCs the cloud-native way

Seven decisions stand between you and a Windows 365 deployment. Work through each one, record where your organization needs an exception, and generate a decision record you can take to a design authority.

Your answers stay in this browser. Nothing is uploaded.

7
deployment decisions
30
components you can keep out of the failure path
0
Azure infrastructure to build first
Overview At a glance The seven decisions Supportability Resources Your decision record
Community resource — not an official Microsoft product. This guide and the plans it generates are provided under the MIT licence with no warranty. Review every plan before you run it, and check Microsoft Learn for authoritative guidance.

Overview

Deploying Windows 365 involves seven decisions. Each has a recommended option that aligns to cloud-native and Zero Trust principles, and a traditional alternative that exists for organizations with requirements the recommended option can't meet.

This guide takes a position. Microsoft recommends the cloud-native model as the preferred starting point for most organizations because it aligns with Zero Trust principles and reduces operational complexity. Your task as a decision maker isn't to choose freely between two equal options — it's to validate the recommended option against your requirements first, and to document any deviation with the reasoning behind it.

Note

Cloud-native deployment options can still provide access to on-premises resources when required. In this context, cloud-native doesn't mean cloud-only.

How to use this guide

  1. Read the recommended option and the case for it for each of the seven decisions.
  2. In the decision builder, record whether you're adopting the recommended option or taking an exception.
  3. Where you take an exception, complete the justification — affected users, the requirement, trade-offs accepted, owner, and exit criteria.
  4. Generate your decision record and export it for review.

Cloud-native and Zero Trust principles

Zero Trust is a security strategy rather than a product or service. It's an approach to designing and implementing a set of security principles: verify explicitly, use least privilege access, and assume breach.

Rather than relying on implicit trust based on network location, this model prioritizes identity, device health, application access, and data protection controls to verify access continuously and reduce security risk. Each recommended option below moves a control away from network location and towards identity and device posture.

Zero Trust overview

Decisions at a glance

For each decision, start by evaluating the recommended option against your organization's requirements. Use a traditional alternative only where a documented dependency exists and can't be met by the cloud-native approach — for example technical, regulatory, compliance, security, application, networking, or operational.

Important

Where an alternative may be required, aim to limit its scope only to the affected users or scenarios. The decision should be documented along with the affected users or workloads, the business or technical requirement being addressed, the associated trade-offs, and any future opportunities to move toward the recommended cloud-native model.

Organizations should seek to minimize the scope of exceptions wherever possible. This ensures that the broader deployment can retain the benefits of a cloud-native, Zero Trust approach.

Benefits of the recommended model

Less configuration

Each recommended default requires minimal or no configuration. You can run a cloud-native pilot without an Azure network, an image pipeline, or a directory-sync dependency.

Fewer on-premises dependencies

Removing directory sync, custom-image capture, and network configuration shortens both provisioning time and sign-in time.

Reduced operational scope

Microsoft manages the network, the base image, patch orchestration, and the underlying infrastructure. Your team manages policy and apps.

Identity-based access control

Access is based on identity, device posture, workload, and data signals rather than a trusted network location.

Scaling without infrastructure changes

Microsoft-managed capacity and service-managed images let you add or move Cloud PCs without first expanding infrastructure.

Fewer components to troubleshoot

Fewer customer-owned components in the failure path narrows the scope of most troubleshooting.

Before you begin

Confirm the following before working through the decisions. These apply regardless of which options you choose.

    This checklist is for your own tracking and is saved in this browser only.

    The seven decisions

    Each decision sets out the recommended cloud-native option and the traditional alternative. Expand Why this is recommended, and when to deviate for the full case, the trade-offs and the adoption path. Where you choose an alternative, complete the justification so the exception is documented rather than implicit.

    0/7
    Cloud-native decisions
    0
    Exceptions taken
    0
    Components re-introduced
    0
    Exceptions fully documented

    Make a selection for each decision to build your record.

    Your decision record

    This record summarizes your selections and the justification for any exception. Export it as Markdown for a design document, or print it for review.

    Generate a deployment plan

    Three of the seven decisions — identity, network and image — determine what actually gets created in your tenant. Those compile into a w365plan/1 file you review, then apply yourself with PowerShell. The other four decisions stay in your record as guidance; the plan never writes them.

    This page never signs in to your tenant. The plan is compiled in your browser session and handed back to you. Sign-in happens on your own workstation, delegated, when you run the plan — not here.

    Must be a concrete region. The plan adds a check that validates it against supportedRegions before anything is written.

    Azure Network Connection

    The domain join password is never entered here and never stored in the plan. The runner prompts for it when it needs it.

    How to run the plan
    # 1. The only module the runner needs
    Install-Module Microsoft.Graph.Authentication -Scope CurrentUser
    
    # 2. Preview. GET requests only — writes nothing, not even a manifest.
    ./Invoke-W365Plan.ps1 -PlanPath ./plan.json -TenantId <tenant> -WhatIf
    
    # 3. Apply. Re-running resumes rather than duplicating.
    ./Invoke-W365Plan.ps1 -PlanPath ./plan.json -TenantId <tenant>
    
    # Undo — deletes only what this run created, newest first.
    ./Invoke-W365Plan.ps1 -Rollback ./plan.manifest.json

    Sign-in is delegated and requests only the scopes the plan lists. Run it against a test tenant before production.

    Supportability benefits

    Supportability describes how easily a deployment can be operated, diagnosed, and repaired once it's in production. It's a measurable outcome of the decisions above, not a separate workstream added afterwards.

    Every component you don't own is a component you never have to troubleshoot

    Adopting the recommended option for all seven decisions keeps 30 customer-managed components out of the failure path — from domain controllers and virtual networks to image pipelines and profile shares.

    Each one removed is something that can't fail, doesn't need checking during triage, and never has to be explained to anyone else.

    Three questions determine how supportable a deployment is:

    • Product quality — does the platform avoid causing the issue in the first place, and when something does fail, does it fail in a way that's clear and actionable?
    • Customer enablement — can the administrator or user understand and resolve the issue themselves, without opening a case?
    • Support efficacy — when a case is opened, is there enough evidence to establish cause quickly, and can it be resolved at the first tier?

    Components in the failure path

    The most direct supportability benefit comes from reducing the number of components involved. Every customer-managed component is a component that can fail, must be checked during triage, and must be described accurately before anyone else can help. Removing one removes all three costs.

    Components removed from the customer's estate when the recommended option is used.
    DecisionComponents removed from the failure path

    Why the effect compounds

    Components don't add to triage effort one at a time, because a failure in any of them can surface as the same symptom.

    Consider a Cloud PC that's Microsoft Entra hybrid joined, provisioned over an Azure Network Connection, from a custom image. The reported symptom is a single line: provisioning failed. That one symptom has candidate causes in every layer — the image failed validation, the virtual network has no available addresses, a route or firewall rule blocks a required endpoint, DNS can't resolve the domain, no domain controller is reachable from the subnet, the service account lacks permission on the target organizational unit, or the computer object hasn't yet synchronized to Microsoft Entra ID.

    Each has to be eliminated in turn, and the error alone rarely identifies which one applies. The layers also depend on each other, so a fault in one presents as a failure in another: a DNS or routing problem in the virtual network reports as a domain join failure, which sends the investigation to the identity team when the fault is in the network.

    What this means when a case is raised

    Even a well-designed deployment produces cases. The recommended model changes what happens when one is opened.

    Recommended modelTraditional alternative
    Scope of triageConstrained to Microsoft-managed services, Intune policy, and the connecting device's networkIncludes customer-managed identity, network, image, update, and file infrastructure
    Reproducing the issueStandard baseline — reproducible on any Cloud PC from the same provisioning policyDepends on which image version, policy set, and network path the device received
    Evidence at case creationIntune and Windows 365 reporting, Microsoft Entra sign-in logs, Autopatch reports, connection quality dataFragmented across multiple consoles and infrastructure owners
    Likely resolution tierHigher proportion resolved at first contactMore escalations, because establishing cause requires customer infrastructure knowledge
    Remediation optionsReset, reprovision, resize, or reapply policy without data lossConstrained where user data or configuration is coupled to the device or a share
    Coordination requiredAdministrator and Microsoft SupportAdministrator, network team, identity team, imaging team, and Microsoft Support
    Tip

    Cloud user data delivers the single largest supportability benefit. When data lives in Microsoft 365, reprovisioning becomes a routine repair rather than a data-loss event. A wide class of issues — profile corruption, disk pressure, unexplained configuration state, failed in-place remediation — converts from an investigation into a repair the administrator, and often the user, can perform directly.

    The supportability cost of an exception

    Traditional alternatives exist because some requirements genuinely can't be met another way. The supportability consideration isn't whether to allow exceptions, but how widely they're applied and how well they're recorded.

    Each exception reintroduces components for the users it covers, and moves responsibility for that part of the deployment back to your organization. An exception applied to a single persona affects the supportability of that persona. The same exception applied estate-wide as a default removes the benefit from the entire deployment.

    A documented exception is also a diagnostic asset. When a case is opened for an affected user, knowing that this population deviates from the standard model — and precisely how — is often the fastest route to cause.

    Validate the deployment

    Confirm each of the following from a pilot device before wider rollout. Capture the results as a baseline so a later change can be measured against a known-good starting point.

      Caution

      Create Conditional Access policies in Report-only mode, review sign-in logs for several days, and confirm a break-glass account is excluded before enforcing. These policies evaluate the connecting device, not the Cloud PC. An untested policy blocks users from reaching their Cloud PCs entirely.

      Monitor the deployment

      The reporting below indicates whether the deployment is behaving as designed, and provides the evidence needed if a case is opened.

      SignalWhere to find itWhat it tells you
      Provisioning success and failure reasonsIntune admin center > Devices > Windows 365Whether provisioning failures cluster around a specific policy, region, or network configuration
      Connection quality, including UDP percentageReports > Windows 365 > Connection qualityWhether RDP Shortpath is being used, and whether latency and bandwidth are within expectation per location
      Startup performance and application reliabilityReports > Endpoint analyticsWhether an issue originates on the Cloud PC or on the client side
      Update compliance and rollout progressWindows Autopatch reports in IntuneWhether an issue correlates with a recent update, and which ring is affected
      Policy and application deployment statusIntune device and policy reportingWhether a device received its intended configuration
      Sign-in and Conditional Access resultsMicrosoft Entra admin center > Sign-in logsWhether access failures are policy decisions rather than platform faults
      Note

      A low UDP percentage in the connection quality report means RDP Shortpath is being blocked somewhere in the path. This is the most common cause of poor Cloud PC performance on an otherwise healthy deployment, and it identifies the problem without requiring a network capture.

      Resources

      Was this guide helpful?