The biggest AVD bill-shock comes from running every session host 24×7 when your users work 8×5. Autoscale scaling plans fix that by starting and stopping hosts on a schedule and on demand — often cutting compute cost by half or more without users noticing.
Where AVD money actually goes
- Compute — session host VM hours. By far the largest and most controllable lever.
- Storage — profile and image storage; smaller, steadier.
- Licensing — typically covered by eligible M365/Windows licences.
- Egress/extras — usually minor.
Because compute dominates and is the easiest to flex, autoscale is the highest-leverage optimisation you can apply.
Scaling plans in plain terms
A scaling plan attaches to one or more pooled host pools and divides the day into phases. In each phase AVD decides how many hosts to keep on based on a load threshold and a capacity buffer:
| Phase | What it does |
|---|---|
| Ramp-up | Morning: bring hosts online ahead of demand so logons are fast |
| Peak | Business hours: keep enough capacity for full load |
| Ramp-down | Evening: drain and shut down hosts as users leave |
| Off-peak | Overnight/weekend: run a minimal floor of hosts |
Two key dials
Capacity threshold (the load % that triggers more hosts) and minimum percentage of hosts on (your floor). Tune these per phase to trade responsiveness against cost.
Graceful drain, not a hard stop
During ramp-down, autoscale doesn’t kill active sessions. It puts hosts into drain mode (no new sessions), optionally notifies signed-in users, waits a grace period, then deallocates empty hosts. You control the forced-logoff behaviour for stragglers.
Right-sizing the hosts themselves
Autoscale decides how many hosts run; right-sizing decides how efficient each one is. Two questions to revisit with real telemetry:
- Is the VM SKU correct? Watch CPU, memory and disk via Azure Monitor / AVD Insights. Persistently low utilisation means you’re paying for headroom you don’t use.
- Is users-per-host tuned? Too many users per host hurts experience; too few wastes money. Adjust the max-session-limit and SKU together.
Levers beyond autoscale
- Reserved Instances / savings plans for the always-on floor of hosts you know you’ll run.
- Ephemeral OS disks for stateless pooled hosts — cheaper and faster to rebuild.
- Right-tier storage — Premium where profiles need it, Standard where they don’t.
- Image hygiene — smaller, well-maintained images deallocate and start faster.
Measure before and after
Capture a week of cost and utilisation before enabling a scaling plan, then again after. It both proves the saving and shows whether your thresholds are too aggressive (slow logons) or too generous (wasted hosts).
A pragmatic first scaling plan
Start conservative: a modest ramp-up before business hours, a peak phase with a comfortable buffer, a ramp-down an hour after typical finish with user notifications, and an off-peak floor of one or two hosts for late workers. Watch logon times for a week, then tighten.
Autoscale is the rare optimisation that improves both the bill and the platform — fewer idle hosts, fresher hosts each morning, and a clear cost story for the business. Set it up early and revisit the thresholds with real data.
Need a hand with your AVD platform? 🚀
I help organisations design, migrate and optimise Azure Virtual Desktop. If you’re planning or troubleshooting a deployment, get in touch.
