Speaker 1: One edition of Windows is built to put a whole team on a single machine, and it's called multi-session Windows. You might know that Windows Server can host several people at once, but this works differently, because it's a client version of Windows that looks and feels like the desktop your users already know. It only runs in production inside Azure Virtual Desktop, and it's licensed per user, not per device. The confusing part is that the easy-looking decisions, the ones you make before you deploy anything, are the ones that decide whether it runs smoothly or turns into a mess. We work in these environments every day, and the trouble almost always comes down to four things. The join type, the profile setup, the host size, and the session limit. Let's start with the plain version. Windows 10 Enterprise Multisession and Windows 11 Enterprise Multisession are special editions of Windows that let several people run their own desktop session on one virtual machine at the same time. Each person signs in with their own account, gets their own desktop, and opens their own apps, and they never see anyone else's files or Windows. The reason these editions exist is simple. Before them, only Windows Server could host several interactive sessions at once. A shared virtual desktop meant handing users a server desktop that looked and behaved differently from the Windows your company actually runs. Multisession changed that because it brings the familiar client Windows experience onto a machine that many people can share. What they share is the machine underneath, which means the processor, the memory, and the storage. That shared resource model is the whole reason multisession lowers your cost per user. Instead of paying for a separate virtual machine for every person, you pay for one machine and divide it up. Picture five co-workers logged into the same machine at once, each one convinced they have a computer to themselves, and from their side that's exactly what it feels like. One quirk is worth knowing early, because it explains a lot of strange behavior later. The operating system reports itself internally as Windows Server. Some older application installers look at that and refuse to install Windows, while others install a server build of the app without warning you. You won't always see that as an error message, so test your critical applications before you roll them out to a shared pool. That server disguise exists on purpose because it keeps multi-session compatible with older remote desktop tools that expect a server. It also keeps applications that were written for shared sessions working the way they should. The trade-off is that a handful of installers will misread the machine and the only reliable way to find them is to test before you deploy. This is the part people get wrong, so it's worth saying clearly. In production, multi-session Windows is only allowed to run inside Azure Virtual Desktop, and the machines have to sit in Azure or in Azure Local. Microsoft doesn't allow you to run it in production outside the AVD service. It also won't activate against an on-premises key management service, so you can't stand it up on your own hardware and call it a day. A newer option called Azure Virtual Desktop Hybrid lets you run cloud-managed desktops on your own hardware. Multi-session isn't supported there yet, so if you're planning around that, keep your pooled multi-session hosts and your hybrid work in separate host pools for now. You also can't take an existing Windows Pro or Enterprise Virtual Machine and upgrade it into multi-session. It doesn't work that way and you can't switch the edition with a product key either. If you get this wrong, the only fix is to redeploy the machine from the right image. So pick the multi-session image from the start. When you go looking for that image in the Microsoft Marketplace, filter for the publisher Microsoft Windows Desktop, the offer Windows 11, and a SKU that ends in AVD. That suffix is what marks the multi-session editions apart from the ordinary single-session Windows 11 images, and getting it right saves you a rebuild later. Now you might be thinking about Windows 365 because cloud PCs sound like the same idea. They're not the same thing at all. A Windows 365 cloud PC is built for one user. Even the frontline shared mode, which lets a license be shared across up to three people, never puts those people on the machine at the same time. Multisession means simultaneous, and that's the line between the two products. Now for the question your finance team will ask, which is what it actually costs to be allowed in. Multi-session Windows is licensed per user, not per device, and per device licenses don't qualify. If you already own Microsoft 365, E3, E5, A3, A5, F3, or Business Premium, you likely already have the access rights you need. A Windows Enterprise E3 or E5 license qualifies too, and so do Windows Education A3 and A5 and Windows VDA per user. Here's the part that surprises admins coming from the server world. You don't need remote desktop services client access licenses for this. The old shared desktop model was a server product, and it charged a client access license for every person who connected. Multisession flips that around because it writes on the per-user Windows or Microsoft 365 license you may already be paying for. That is the cost argument in one sentence, since you stop buying separate server access rights and start using licenses you already own. Before you budget anything new, check what's already assigned in your tenant. Because buying a second license for a user who already qualifies is one of the easiest ways to waste money. A second model matters here because it comes up with managed service providers and anyone building for customers. If you're providing desktops to people outside your own organization for commercial purposes, you use per-user access pricing through an enrolled Azure subscription instead. That's a pay-as-you-go meter. It only charges for users who actually connected that month. and it covers access rights in FSLogix rather than Office or Defender. Whichever path you take, the Azure infrastructure underneath is still consumption-based, so you pay for compute, storage, and networking on top of the license. The license gets you in the door, and your operations decide the rest of the bill. So let's look at what those operations actually involve. Let's get into the mechanics, because this is where the money is made or lost. In Azure Virtual Desktop, the group of shared machines is called a host pool. And every machine inside that pool counts as a session host. Your users connect to the pool and the service decides which session host they land on. There are two ways to spread people across those machines and the choice changes your bill. Breadth First spreads new sessions evenly across all the hosts, which keeps performance steady because no single machine gets crowded. That first does the opposite because it fills one host to its limit before it starts using the next one. which means fewer machines run at once and you spend less on compute. So this one setting decides whether you're paying for machines nobody is using. Depth first needs a maximum session limit, and that's the number that tells the service when a host is full. Breadth first doesn't require one, and here's a detail worth remembering. When you create a pool with breadth first, default maximum session limit is 999,000, which is also the highest value the setting accepts. Leave it alone and you never really cap a machine, which is a common way a shared pool gets overloaded. A quick example makes it concrete. Say you have nine session hosts. With breadth first, a new person lands on one of the five emptiest hosts at random, so sessions stay roughly even across the pool. With depth first, that same person lands on the fullest host that still has room, and the empty machines stay empty until you actually need them, then they're sizing, And this is where honesty matters. Microsoft recommends a minimum of four vCPUs per session host with a supported range of four to 24. How many people actually fit depends on what they do. Light work like email browsing and basic documents packs in at higher density, while video editing CAD or data-heavy applications need far more per person and fit far fewer. No magic users per host number exists that you can copy from a blog because the machine only knows the workload you give it. So let's validate that in your own environment before you commit a whole department to it. Spin up one host, load it with your real applications, and watch what the processor and memory actually do under load. The last piece of the mechanics is what happens when demand drops. Autoscaling can shut down idle hosts, and it pairs naturally with depth first because you only need the machines you're actually using. one warning here, and it's a costly one, because a powered-off Azure virtual machine still builds. Autoscaling only saves money when it fully deallocates idle hosts instead of just stopping them. So let's talk about when multisession makes sense and when it's the wrong tool. The clearest fit is task workers, which means people who live in email, a browser, and a couple of line of business apps. Call centers, back-office teams, and shift-based roles fit the pooled model well because their work is light and their sessions look a lot alike. Students are another good fit because they need a standard desktop with standard apps and they don't need a personal machine running between classes. Contractors and seasonal staff fit too since you can hand someone a desktop for a few weeks without buying a machine for each person. It also suits teams that don't need a dedicated machine to survive between sessions. If nobody cares which physical host they land on, Pooling is the cheaper design because the shared model is exactly what drives the cost down. The wrong fit matters just as much. Developers, designers, and anyone running heavy tools want a machine with more horsepower and they often want to install their own software, which conflicts with the shared model. Give a power user a small shared host and they'll feel it inside the first hour. Users who need admin rights are a poor fit for a shared machine because on a pooled host, their installs disappear when they sign out. The same goes for anyone who likes to install a lot of their own tools. When someone genuinely needs a dedicated, persistent desktop that keeps running while they're away, a personal or single session desktop is the honest answer. The rule of thumb is simple, pool the predictable standard work and give the demanding or highly individual work its own machine. That split is where multi-session pays off. without turning into a support burden. Once machines are shared, the next question is what happens to a person's files and settings when they land on a different machine tomorrow. Pooled hosts are shared and they're often non-persistent, so a user can sign in on one machine today and a completely different one tomorrow. Without something to carry their profile along, their desktop layout, their bookmarks, and their mail setup would all reset. FSLogix solves that. It packages a user's profile into a virtual disk container, and that container attaches to whichever host the user connects to at login. The profile follows the person instead of the machine. FSLogix comes pre-installed on every multi-session image from Microsoft, and Azure Virtual Desktop users can use it at no extra cost, though you still have to configure where those containers live. A simple picture helps here. Sign in on host one on Monday and host four on Tuesday and your desktop, your bookmarks and your mail profile are exactly where you left them. The container writes along to whichever machine you land on. Most teams store those containers on Azure Files because it works cleanly with both cloud identities and on-premises accounts. Then there's the rule that catches almost everyone. Applications installed in the user context don't persist on a shared host. When the person logs off, those installs disappear because the profile that held them gets cleaned up. Apps have to be deployed at the device or system level so every user on that machine gets them. In practice, that changes how you work. You stop letting people add their own applications and you usually disable the Microsoft Store to keep the shared image clean and consistent. It's a real trade-off because it's less freedom for the user and it's the price of a shared machine that behaves the same way for everyone. Now that the pieces are in place, let's talk about running this day to day. You manage these hosts with Configuration Manager, where version 19.06 and later support multisession or with Microsoft Intune. Both work and the choice usually follows whatever you already use for your other Windows devices. If you want Intune to manage them, there's a hard rule you have to respect. The hosts must be either Microsoft Entra joined or Microsoft Entra Hybrid joined. Hosts joined to Microsoft Entra domain services aren't supported, and that's a constraint you can't fix after the fact. You have to decide the join type before you build the host pool. Once they're enrolled, think in two scopes. Device scope covers the shared machine itself, which means operating system settings, system-level applications, and security policy. User scope covers the individual session, which means per-user settings and user certificates. Most of your configuration belongs at the device scope because it needs to be the same for everyone who lands on that machine. A couple of things to note about what doesn't work. Security baselines, Windows update rings, and most remote actions like wipe, fresh start, and BitLocker key rotation aren't supported on pooled multi-session hosts. For security policy, build the settings manually in the settings catalog instead of applying a baseline. And for updates, you use Windows Update for Business instead of update rings. That sounds like a step back, and in some ways it's because you're trading convenience for a machine that serves many people at once. The year-grade-out buttons aren't a misconfiguration you can troubleshoot away, since they're expected behavior, so plan your security and update strategy around the tools that do work. One more rule follows from the shared design. Applications must be assigned to devices rather than users, and they must install in the system context, because there's no per-user install on a pooled host. Compliance policies and conditional access do work here, so you can still require a healthy device and enforce multi-factor sign-in. Image management is the ongoing core. You build a golden image stored in Azure Compute Gallery, so you get versioning and replication across regions, and then re-image hosts to roll out changes. One warning applies here, since you shouldn't clone a host that's already enrolled, because the clone's enrollment and sync will break. Build new images from scratch instead. Let's put the traps in one place, because these are the ones that cost real money. Start with the maximum session limit, because leaving it at its default is the easiest mistake to make. On a breadth-first pool, the default is the highest value the setting accepts, so nothing caps a host, and one machine can take on more users than it can carry. Sizing comes next, and undersizing is the fastest way to a bad first impression. Four vCPUs is the floor, not a target, and power users need far more room per person, so the first session of the day often sets the tone for everything after it. Then there's the Windows 365 mix-up. Cloud PCs serve one user, and frontline shared mode shares a license without putting people on the machine at the same time. Design for the wrong one, and your licensing and your costs both come out wrong. Forgetting FSLogix causes its own kind of pain. because users then lose their settings and their applications vanish at log off. The same thing happens when apps are installed for a single user instead of for the whole device. Join type is another one that bites later. If you go with intra-domain services, Intune won't manage your hosts and you can't repair that without rebuilding. Treating a multi-session machine like an ordinary Windows virtual machine is a related trap. You can't upgrade a pro or enterprise machine into multi-session. and switching editions with a product he isn't supported either, so a wrong image means a rebuild rather than a quick fix. The last trap is about the bill. Stopping a virtual machine is not the same as deallocating it, and a powered-off machine still costs you money. Autoscaling only saves what it's supposed to save when it actually deallocates idle hosts, which means the setting has to be configured rather than assumed. That's the honest limit worth repeating. Pooling lowers your per-user cost, but only with active management, because left alone, idle hosts keep running and the savings disappear. That is how a good design turns into an expensive one. So let's pull it together. Multisession Windows is a client edition of Windows that puts many users on one Azure virtual machine, and it only runs in production inside Azure Virtual Desktop. It's licensed per user, and it leans on the Microsoft 365 or Windows license you may already own. Four decisions decide whether it works, the join type, the FSLogix profile setup, the host size and the session limit. Here's my question for you. How many people are sharing a desktop in your environment today and what decided that number? Tell me in the comments because the answer usually says more about the design than the hardware does. If this made the topic feel less intimidating, subscribe to M365FM to master the tools and the mindset shaping the future of work