Virtual desktop was evaluated across a decade, both major platforms were piloted, and the position I held was that the enterprise should not buy it. It never was.
Piloted properly, and closed properly
A Citrix XenDesktop environment was built in 2011 on a dedicated pilot subnet, with firewall policy staged so the whole thing could be backed out cleanly, and instrumented with per-policy counters so it could be measured as well as removed. A VMware View environment was piloted alongside it. Both were decommissioned by change order inside eighteen months.
Closing a pilot deliberately is what separates an evaluation program from an abandoned project. The difference shows up years later, when somebody asks what happened to it and the answer is a change record rather than a shrug.
Three objections, one sentence
When virtual desktop reappeared on the technology executive committee’s slate in 2015 — carrying visibility but no owner, no scope and no written justification — the objection was that it had no defined constituency, a prohibitive cost case, and a load the existing infrastructure would have to absorb.
That held for eight years. The technology recurred in hyperconverged and vendor pitches through 2023 and was never funded as an enterprise program.
The judiciary had reached the same conclusion separately and on different grounds: courtrooms had moved to file-less proceedings, which made a network partition a courtroom-stopping event. Local disk and local images were a resilience requirement a thin client could not meet.
The textbook case was already solved
Several hundred public-access machines across the library branches are exactly what virtual desktop is sold for: untrusted users, no persistent state wanted, a guarantee that every session starts clean, and central fleet management.
Those branches already ran reboot-to-baseline image control — discarding every change a patron made and reverting to a known-good image on restart, on a nightly thaw window with a deliberately small persistent volume, and used as the distribution channel for anything that had to reach a public machine.
It delivered the same guarantee with no data center compute, no storage tier sized for hundreds of concurrent desktops, no brokering layer and no additional branch capacity, because the compute stayed on hardware already bought. Virtual desktop would have moved all of it inland across the metropolitan links to buy a property the libraries already had. The approach ran unbroken to 2023 and was extended to the community center estate rather than replaced.
A note on the evidence
This is recorded as the position the evidence supports rather than as a written trade study. A formal cost comparison between the two does not survive in the record, and saying so is more useful than implying one exists.
Outcome
Never funded as an enterprise program. The strongest internal case was already solved more cheaply by control the estate already had.