The Enterprise VDI Landscape: What IT Leaders Are Prioritizing Today

Virtual desktops are no longer just about cutting costs or supporting remote work. IT leaders are also focused on predictable spending, reliable user experience, and better control over AI tools and agents operating inside user sessions.

At the same time, many VDI and DaaS vendors are pushing AI-driven management, while IT teams may have more immediate concerns. The bigger questions are where VDI still delivers clear value, where it does not, and whether it can become a useful control point for AI governance.

Ace Cloud Hosting spoke with David Marshall, founder of VMblog.com, which he launched in 2004. The publication covers virtualization, cloud computing, end-user computing, AI, cybersecurity, Kubernetes, DevOps, and other modern data center technologies.

David also co-authored two books on server virtualization and served as technical editor for Virtualization for Dummies and VMware VI3 for Dummies.

In this Expert Opinion Q&A, David shares where VDI still earns its place, what cost models often overlook, and why AI governance is becoming an important part of the virtual desktop conversation.

1. You’ve been covering this space since 2004. When you talk to IT leaders today, what’s genuinely at the top of their VDI priority list, and where is that furthest from what vendors are currently selling?

Honestly, the conversation has gotten a lot more grounded than it was even three years ago. When I talk to IT directors and VPs of infrastructure today, three things come up over and over: predictable cost, a tolerable user experience on home broadband and coffee-shop Wi-Fi, and (this is newer) some kind of handle on what employees are doing with AI tools inside their sessions. Nobody’s asking me about the next flashy feature. They’re asking how to stop getting surprised by the bill.

Here’s where it gets a little uncomfortable for vendors. A lot of the messaging coming out of the traditional VDI vendors and the DaaS crowd right now leans hard into AI-infused management, copilots for the admin console, and “agentic” everything. That’s nothing.

Some of it genuinely helps with image management and troubleshooting. But it’s not what’s keeping IT leaders up at night. They want fewer moving parts, not more intelligent moving parts. There’s a real mismatch between “Buy our AI layer” and “please just make renewal season less painful this year.”

2. Cost is where most VDI business cases get challenged internally. In your experience, what do initial projections consistently underestimate?

Almost every VDI cost model I’ve seen over the years gets the “sticker price” roughly right and everything else roughly wrong. Licensing is the classic offender — teams price out the hypervisor and the broker, then get blindsided by the Windows or Microsoft 365 licensing layer, a third-party profile management tool, and a monitoring add-on that quietly becomes a permanent line item.

A couple of other items that reliably get underestimated:

• Peak concurrency, not average load. Teams size for a typical Tuesday, then get walloped when the whole call center logs in at 8 a.m. sharp.
• The human cost of care and feeding. Someone has to patch golden images, tune GPU allocation, and answer “why is my session slow” tickets — that’s headcount, not software.

And now there’s a new one creeping into every conversation: GPU capacity for AI-augmented desktops. Organizations bolting Copilot or local inference onto virtual sessions are discovering that GPU-backed instances cost meaningfully more than the CPU-only desktops they budgeted for. That gap rarely makes it into the first-year projection.

3. AI arrived on employee desktops well ahead of IT approval, and agents are starting to act with real permissions inside user sessions. Does the virtual desktop become a natural control point for governing that, or just another blind spot?

Genuinely, it could go either way, and right now it’s trending toward a blind spot more than a control point, mostly because organizations haven’t updated their thinking yet.

Here’s the case for VDI as a control point: sessions are already brokered, monitored, and in the non-persistent case, wiped clean on logout. That’s a decent foundation for containing an AI agent that goes rogue or gets tricked into doing something it shouldn’t. If IT treats the session boundary as an identity boundary, scoping what an agent can read, call, or move data to, the same way you’d scope a new hire’s access, the virtual desktop becomes a genuinely useful choke point.

The problem is that most shops still treat VDI as a delivery mechanism, not a governance layer. An employee installs a coding assistant or a browser-based agent inside their session, and it can read files, call external APIs, and move data out through channels that traditional endpoint tools were never built to watch.

Security researchers have been blunt about this: legacy monitoring catches a file upload, but it’s largely deaf to an agent quietly exfiltrating data through what looks like normal API traffic. Unless IT explicitly builds AI governance into the session — not bolts it on after a scare — the desktop just becomes one more place shadow AI operates undetected.

4. Where do you see virtual desktops delivering the clearest value today, and where are organizations forcing them onto use cases that don’t need them?

The clearest wins haven’t changed much in twenty years, honestly. VDI still earns its keep wherever data can’t leave the building (or the country), and the workforce is transient or distributed. Think regulated healthcare and financial services, contractor and M&A carve-out scenarios, and call centers running standardized, non-persistent images, and disaster recovery scenarios where you need to stand up desktops fast without shipping laptops.

Where it goes sideways is when organizations reach for VDI to fix a problem that isn’t really a VDI problem. Forcing developers or creative teams — anyone doing latency-sensitive, GPU-hungry, highly customized work — onto shared virtual infrastructure usually just trades one headache (endpoint management) for another (session lag and unhappy engineers).

I’ve also seen VDI used as a substitute for actually fixing sloppy device management or weak identity practices, which never works. If your real issue is that laptops aren’t patched consistently, virtual desktops won’t solve that; they’ll just move the mess somewhere less visible.

5. If you were advising an IT leader building their 2027 roadmap, what deserves more attention than it’s getting, and what’s getting more than it deserves?

More attention, without question: agentic AI governance built into the session layer, not layered on top after something goes wrong. Most roadmaps I’ve reviewed treat AI oversight as a policy document. It needs to be an architecture decision — how permissions are scoped, how agent activity gets logged, how data movement out of the session gets watched.

I’d also put real cost accounting for GPU-enabled desktops on that list; too many teams are still estimating AI capacity needs by gut feel rather than actual workload data.

What’s getting more attention than it deserves? That’s a tough one. I’d gently push back on the instinct to over-provision GPU capacity across the board “just in case” — that’s how you end up with expensive silicon sitting idle while the finance team asks pointed questions at renewal time. Instead, spend the planning energy on governance and real usage data first. The flashy stuff can wait for a budget cycle.

Get the Basics Right Before Adding AI

David Marshall’s main takeaway is simple: get the basics right before chasing the latest AI features. For IT teams, the smartest move isn’t adding more tools. It’s deciding what AI can access inside a work session and tracking what it does. Do that well, and the virtual desktop becomes a great place to keep AI under control. Ignore it, and it turns into the one spot where unapproved tools run in the background without anyone noticing.

Cost planning deserves equal attention. Teams that budget only for typical usage can be caught off guard by peak demand, ongoing management, licensing, and GPU requirements. Accounting for these costs early helps avoid unexpected spending later.

This is exactly where Ace Cloud Hosting can help IT leaders modernize desktop infrastructure without adding operational complexity. Our secure, centrally managed virtual desktops scale across users, locations, and workloads, while GPU-powered options support demanding CAD, BIM, engineering, 3D design, and other graphics-intensive applications without the need for continuous investment in high-end physical workstations.

With centralized controls, scalable CPU and GPU resources, predictable pricing, and 24/7 support, IT teams can improve performance and standardize desktop delivery while retaining flexibility for different business needs. Explore our virtual desktop solutions and start with a free trial or speak with our experts today.