Do All Your VDI Users Really Need Their Own Virtual Machine?

I have watched a lot of VDI costs grow steadily over my career. The environment works, users are productive, and the invoice keeps climbing, and nobody can point to the line that is might be considered waste. More often than not, one unexamined assumption is doing the damage: every user needs their own dedicated virtual machine.

That assumption is worth questioning, because a good number of your users probably do not. While some absolutely do, figuring out which is which is where the savings lives, and it is a far more useful question than a "single-session versus multi-session" conversation as a matter of principle. Here is how I have thought it through with organizations in the past:

Why does one user per machine cost more than it needs to?

A one-to-one setup costs more because each dedicated machine spends most of its life underused, even while someone is actively working on it. Each user gets a full VM sized to cover their busiest moment of the day, which means it is oversized for almost every other moment. The engineers I have worked with over the years describe it the same way every time, which is that personal desktops are comfortable and expensive, and the expense is mostly capacity nobody is touching. Nerdio, a Microsoft AVD tooling partner, backs the size of that gap: they put typical utilization on a one-to-one desktop at only 20 to 50 percent on average, which means half or more of those resources go unused while someone still pays the full cost.

What I have seen on the customer side is that this cost hides well. It does not show up as a failure or an outage, it shows up as a number that is a little higher than it should be, month after month, until someone finally asks why. By then it is baked into the baseline and nobody wants to touch it. The encouraging part is that a large share of those users never needed a dedicated machine to begin with.

Where does multi-session actually save money?

Multi-session saves money when several users can comfortably share one machine without stepping on each other. Instead of a VM per person, you put a group of users on a single, larger host and let them draw from a shared pool of resources. The technical people I trust on this put the practical density at a handful of users per VM for standard workloads, and the same Nerdio analysis puts average utilization on a well-packed host up near 80 percent. That jump from half-used to mostly-used is the whole savings story in one sentence.

The customers this fits are the ones I would call predictable. Task and knowledge workers living in email, the Office suite, a browser, and a couple of line-of-business apps rarely need a whole machine to themselves. I have watched teams like that run beautifully on shared hosts, and the finance side stops flinching at the bill. The pattern is consistent enough that it should be the default question for any standardized, high-headcount group: is there a reason these users each need their own VM, or have we just never asked? Some of them will have a real reason, and those are the users worth protecting.

Where does single-session still earn its keep?

Single-session still earns its keep whenever one user's work would degrade everyone else's on a shared machine. That is not a rare exception, and it is not a fallback, it is a real design choice for a real set of users. When I checked this against what the engineers I have worked with see in the field, the list was consistent: power users and developers who install and uninstall constantly, graphics and GPU workloads, applications that simply are not built to run multi-session, users who need admin rights that would spill onto their neighbors, and anyone under isolation or compliance requirements that call for a dedicated environment.

I have watched a single heavy user drag a shared host to a crawl for everyone else on it, and it is a bad afternoon for the IT team that gets those tickets. Giving that person their own machine is not waste, it is the right call. The mistake is not using single-session, it is using it by default for people who never needed it. So the real skill is not choosing a side, it is knowing how to tell one user from the next.

What should actually decide the model for a given workload?

The workload should decide the model, not a company-wide preference for one or the other. When I sit with this, the factors that actually matter are the workload profile, application compatibility, licensing, the constraints of your specific cloud and platform, and any security or isolation requirements. Those five will tell you far more than any rule of thumb.

The one I would flag hardest, because it trips people up in real evaluations, is that multi-session support is not uniform across platforms and clouds. I wanted to be sure before I put this in writing, so I ran it past the technical team, and the short version is that what is possible, how it is licensed, and how dense you can safely go all shift depending on the stack you are standing on. So before you assume the savings are there, the honest first step is knowing what your own environment actually allows. That is less exciting than a blanket recommendation, and it is the part that saves people from a nasty surprise later. Run that same sorting across an entire organization, and a familiar pattern tends to appear.

Why do most real environments end up mixed?

Because once you match the model to the workload instead of the other way around, you almost always end up running both. Nearly every environment I have looked at lands on a blend: multi-session carrying the large, standardized population, single-session reserved for the users and workloads that genuinely need it. That is not a compromise, it is the point. The savings come from right-sizing each group, not from forcing everyone into one model and hoping it fits.

The work, and the reason this is worth doing carefully, is in the sorting. Someone has to look at your actual users, map them to the right model, and account for what your platform and licensing will and will not support. Done well, you stop paying for idle machines without asking a single person to accept a worse day.