Check your VM size now. Azure VM families move through four stages—Current, Extended, End of Life and Retired. Current and Extended sizes remain supported; an End of Life size has a published retirement date and must be migrated before that date; a Retired size can no longer run, be created or be selected for resize.
For the immediate planning case, Dv3, Dsv3, Ev3 and Esv3 remain supported until November 15, 2029, but Microsoft began restricting capacity growth for them in July 2026. Inventorying affected workloads, selecting a supported replacement, testing it and scheduling a rollback-ready production change is safer than waiting for the deadline.
As an Amazon Associate I earn from qualifying purchases.
How Azure VM size lifecycle stages work
A lifecycle label applies to a VM size series, not necessarily to every individual VM configuration. Microsoft defines four stages:
| Stage | Meaning | What you should do |
|---|---|---|
| Current | The actively recommended generation for new deployments. | Use it for new workloads and consider it first when replacing an older series. |
| Extended | Still fully supported while customers evaluate and adopt newer generations. | Plan modernization, but you are not facing an announced retirement deadline yet. |
| End of Life | An announced retirement date exists. The series remains usable until that date, subject to any capacity or regional restrictions. | Start migration planning, testing and scheduling immediately. |
| Retired | The series is no longer available to create, resize into or run. | Use a supported replacement. If a VM was deallocated at retirement, resize it and restart it on a supported size. |
The lifecycle state is separate from a guest operating system’s support policy. A supported Windows or Linux release does not make a retired Azure VM size available again, and a current VM size does not extend an operating system’s own support window.
#1 Best Overall
Why a retirement date is a migration deadline
Microsoft describes retirement as a point by which customers must move away from the affected feature to avoid service disruption. A retirement notice is therefore not a suggestion to upgrade at some convenient future date. It is the last date on which the old size is expected to be supported.
Existing VMs can remain operational during the End of Life period, but capacity is not guaranteed to remain equally easy to obtain. Microsoft began capacity growth restrictions for Dv3, Dsv3, Ev3 and Esv3 in July 2026. A stopped VM, a scale-set increase, a zone move or a new deployment may encounter subscription, quota or regional limitations before November 15, 2029.
Keep lifecycle notices in your change-management calendar, assign an owner for each affected workload and recheck the notice immediately before production execution. Dates, replacement availability and regional support can change.
Azure VM series with announced or completed retirement dates
| Series | Lifecycle event | Planning implication |
|---|---|---|
| Dv3, Dsv3, Ev3 and Esv3 | Retire November 15, 2029. Capacity growth restrictions began July 2026. | Existing instances are supported until the retirement date, but check capacity and eligibility now. |
| HC-series | Retire May 31, 2027. | Evaluate newer high-performance-computing families such as HBv5, HBv4, HX or HBv3. |
| Msv2 and Mdsv2 isolated medium-memory sizes | Retire March 31, 2027. | Plan migration to Mv3 medium-memory sizes. |
| DCas_cc_v5, DCads_cc_v5, ECas_cc_v5 and ECads_cc_v5 | Retired September 1, 2026. | These sizes are already past retirement; use a supported confidential-computing replacement before attempting recovery. |
The dates above are Azure lifecycle dates published for 2026 planning, not performance benchmarks or market statistics. Your subscription and region can have additional restrictions, so confirm eligibility in the Azure portal and the current retirement notice.
What happens if you do nothing before retirement?
After a size is retired, Azure can deallocate the VM. A deallocated VM stops incurring compute charges, but it is no longer covered by the VM size’s SLA or support. The operating system’s in-memory state and data on the temporary disk are lost during deallocation. Data on attached managed disks is preserved.
The documented recovery path is to resize the VM to a supported size and restart it. Recovery is not the same as a seamless continuation: applications may need to rebuild caches, replay transactions, reattach services or re-establish network sessions. If the retired size is unavailable in the region or subscription, you may need to select another family, region or zone after checking dependencies.
Managed disks protect persisted disk contents, but they do not replace application-level backups. Confirm that databases, configuration stores, secrets, certificates and external data services have their own recoverable copies before changing production.
A migration plan that minimizes downtime
1. Build a complete inventory
List every standalone VM, virtual machine scale set instance and dependent service using an End of Life or retiring series. Capture resource group, subscription, region, availability zone or set, size, image, operating system, managed disks, network interfaces, public endpoints, extensions, boot diagnostics, monitoring agents and backup policy.
Include stopped and deallocated machines. A machine that is not running today may still be needed for disaster recovery, scheduled processing or a scale-out event, and capacity restrictions can affect its next start.
2. Read the specific retirement notice and check constraints
For each resource, verify the announced date, any interim capacity restriction and whether your subscription or region is eligible for the proposed replacement. Check vCPU quota, family quota, reservation or savings-plan treatment, availability-zone support, availability-set behavior and regional capacity.
Rank #3
Record specialized requirements before choosing a target: accelerated networking, GPU or FPGA support, confidential-computing features, InfiniBand, premium storage, temporary-disk size, encryption mode, host-based capabilities and extension compatibility.
Quick wins for a faster PC:
Fix the driver behind crashes, sound loss and screen glitchesFind Drivers →Clear out junk files and repair common Windows errorsFree Scan →3. Choose a Current or Extended replacement
For Dv3 or Dsv3 workloads, Microsoft lists Dv5, Dsv5, Ddv5, Ddsv5, Dasv5 and Dadsv5 as families to evaluate. For Ev3 or Esv3 workloads, evaluate Ev5, Esv5, Edv5, Edsv5, Easv5 and Eadsv5. These are family-level alternatives, not an automatic one-to-one size mapping; confirm the exact SKU, vCPU count, memory, disk characteristics and regional availability.
4. Resize a representative test workload
Clone or use a non-production instance with comparable traffic and data shape. Resize it to the candidate SKU, restart it and exercise normal and peak workflows. Test both an online resize, when supported, and the planned stopped/deallocated procedure if production will require one.
5. Validate before approving production
Compare application latency, throughput, CPU and memory pressure, disk IOPS and throughput, network bandwidth, packet behavior and host-level monitoring with the old size. Verify that boot diagnostics, agents, certificates, custom scripts, load-balancer probes, backup jobs and security controls still work.
Test the failure behavior you depend on: reboot, host maintenance, zone-aware deployment, scale-out, failover and restoration from backup. A replacement that is faster in a single test but unavailable in your required zone is not operationally equivalent.
6. Schedule a rollback-ready production change
Choose a maintenance window based on application restart time and data synchronization, not only on the Azure resize operation. Take or verify current backups, document the original size and settings, and define the rollback trigger. Keep the old configuration recorded, but do not assume that resizing back will be possible after the old series reaches retirement or loses regional capacity.
After the change, monitor health probes, application logs, resource metrics, billing, backup completion and user-facing error rates. Close the migration only after a normal operating cycle and a tested recovery point.
How to select among replacement families
| Decision axis | Questions to answer |
|---|---|
| CPU and memory | Does the workload need more cores, more memory per core, or the same ratio? Measure sustained and burst utilization rather than copying the old SKU name. |
| Storage | Are managed-disk limits, local temporary storage, caching modes or required IOPS changing? |
| Networking | Is accelerated networking enabled? Does the candidate provide the required bandwidth, IP behavior and load-balancing compatibility? |
| Specialized features | Does the workload require GPU, confidential computing, InfiniBand, nested virtualization or a particular extension? |
| Placement | Is the exact size offered in the required region, availability zone or availability set, with sufficient quota? |
| Economics | Compare expected compute, disk and network cost, then check whether reservations or savings plans apply to the new family. |
| Support horizon | Prefer a Current or Extended family with a clear support runway, while keeping the next modernization cycle in your roadmap. |
| Change risk | How much will a new CPU generation, memory layout, disk controller or temporary-disk behavior alter application performance? |
Do not infer equivalence from a shared letter. A D-series replacement is generally aimed at general-purpose workloads, while an E-series replacement emphasizes memory. The correct target is the SKU that satisfies measured workload and platform requirements in your region.
Validation checklist for applications and platform services
- Performance: replay representative transactions and record latency, throughput, CPU, memory and garbage-collection behavior.
- Storage: verify disk attachment order, caching, encryption, IOPS, throughput, temporary-disk use and database recovery.
- Networking: confirm private and public connectivity, DNS, accelerated networking, firewall rules, routes, health probes and required bandwidth.
- Extensions and agents: check monitoring, security, configuration-management, backup and custom script extensions after restart.
- Boot and identity: verify boot diagnostics, cloud-init or startup scripts, managed identity, certificates and key access.
- Availability: test the required zone or set placement, scale-set model, failover and re-provisioning behavior.
- Operations: confirm alerts, dashboards, runbooks, patching, backup schedules and cost reporting identify the new SKU.
Common planning mistakes and fixes
Waiting for the final date
Waiting can leave you competing for scarce capacity or discovering that the desired replacement is unavailable in your zone. Start with a test migration while the old VM is still supported.
Do these 3 things before closing this tab:
1Scan for outdated or missing drivers - takes under a minute2Repair Windows errors before they cause bigger problems3Fix the driver behind crashes, sound loss and screen glitchesAssuming managed disks preserve everything
Managed disks preserve their persisted data, not RAM, temporary-disk contents or unsaved application state. Move caches and scratch data to rebuildable locations and verify application-consistent backups.
Choosing by name instead of measurement
A newer family may alter memory ratios, network limits or local storage. Use workload telemetry and a representative test rather than selecting the first SKU with a similar label.
Ignoring quota and placement
A SKU can exist in a region yet fail deployment because your subscription lacks family quota or the required zone has no capacity. Request quota early and test the exact zone and deployment model.
Forgetting dormant disaster-recovery VMs
Include deallocated replicas, templates and scale-set configurations in the inventory. They must also be changed before a recovery event depends on a retired size.
Free tools Windows power users keep installed
One-click scans. No signup required.
A practical timeline for the affected families
- Now: inventory Dv3, Dsv3, Ev3 and Esv3 resources and identify July 2026 capacity restrictions affecting growth.
- During testing: validate one or more listed Dv5/Dsv5/Ddv5/Ddsv5/Dasv5/Dadsv5 or Ev5/Esv5/Edv5/Edsv5/Easv5/Eadsv5 candidates in the required region and placement.
- Before production: obtain quota, confirm backups, document rollback and complete application and platform validation.
- Before November 15, 2029: finish production migration and remove retired-size references from scale sets, templates, automation and disaster-recovery plans.
- For earlier deadlines: prioritize HC-series migrations before May 31, 2027 and Msv2/Mdsv2 isolated medium-memory migrations before March 31, 2027. Treat the September 1, 2026 DCas_cc_v5, DCads_cc_v5, ECas_cc_v5 and ECads_cc_v5 retirement as already effective.
What to do if a retired VM is already deallocated
Locate the preserved managed disks and confirm the required data is present. Select a supported size that is available in the region and meets the workload’s networking, storage and security requirements. Resize the VM, start it, and then run the full validation checklist rather than assuming the guest will resume unchanged.
If the first candidate cannot be allocated, check quota, zone capacity and regional availability before changing application architecture. A different supported family or placement may be required. Once service is restored, investigate why the retirement was missed and update inventory, alerts and ownership so other resources do not reach the same state.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
