I’d upgrade node-by-node (Cluster OS Rolling Upgrade) rather than doing “big-bang” in-place upgrades across the board. It’s the safest way to get to Windows Server/Hyper-V 2025 with essentially zero VM downtime, a clean rollback path, and no mass host rebuilds. If you have any standalone/non-clustered Hyper-V hosts, an in-place upgrade is fine when health checks pass, and you have a tested backup/rollback.
If you’re clustered (S2D or shared storage)
Use Cluster OS Rolling Upgrade (CORU): drain one node, upgrade that node’s OS to 2025, patch/driver it, bring it back, rinse and repeat. Keep the cluster in mixed OS mode until every node is 2025, then bump the Cluster Functional Level. This avoids stopping workloads.
Vendor gear? Grab the 2022 to 2025 S2D guidance (Dell AX example shows both in-place and rolling/clean options and gotchas).
If you’re not clustered (standalone Hyper-V)
Microsoft explicitly supports in-place upgrades to 2025, and starting with Server 2025 you can jump up to four versions on non-clustered systems. Still: run health checks (SFC/DISM, drivers/firmware), snapshot config, and ensure image-level backups first.
Clusters: I’d not build an entirely new cluster unless you’re also doing a hardware refresh or major redesign. CORU exists to make this easy and safe.
Standalone hosts: If hardware/firmware are current and the role is simple, in-place is efficient. If the host is “crusty” (ancient drivers, LBFO teams you’ve meant to retire, old agents), I’d clean-install.
None of these are show-stoppers today, but they’re worth a quick check before you cut over:
Cumulative-update regressions seen mid-2025
– Older VM configuration versions (e.g., v8/12.0) failed first boot after certain July/Aug updates until patched; update hosts and/or raise guest config version if affected.
– GPU-P passthrough quirks were reported when guest Windows 11 installed certain CUs (hosts were fine). Verify your GPU-P workloads after patching guests.
Release-health items (now resolved)
– WinRE USB input issue and Smartcard auth issue from Oct 2025 Patch Tuesday were acknowledged and resolved—be fully current before/after upgrade.
Security patch watch-outs
– Late-Oct 2025 WSUS critical RCE needed an out-of-band fix. If your hosts run WSUS (or you patch via a WSUS box), make sure that system is fully remediated.
What’s new that you’ll want
– Big scalability bumps (host up to 4 PB RAM/2,048 logical processors; Gen2 VMs up to 240 TB RAM/2,048 vCPUs). Plan capacity/placement with those ceilings in mind.
– Live Migration/CPU compatibility improvements (dynamic compatibility) and GPU partitioning enhancements are commonly cited—great if you’re mixing CPU stepping's or doing GPU-heavy VMs.
Backups & rollback: Full VM backups + host bare-metal (or image) backups; confirm restore. (No citation needed, best practice.)
Pause + drain a node; evict or keep in cluster (per your process).
Upgrade that node’s OS to Windows Server 2025, install OEM drivers/firmware, patch fully.
Rejoin/bring back into cluster; fail a few non-critical VMs over and validate Live Migration/storage/SET/GPU-P.
Repeat for each node.
When all nodes are 2025 and happy, run Update-ClusterFunctionalLevel. This is one-way—do it last.