Proxmox VE 8.4 stops getting support in August. 9.2 finally adds the automatic load balancing VMware refugees kept asking for — and I think the upgrade is worth doing for that alone, deadline or not.
If you run Proxmox at home or in a small business, put a date in your calendar: Proxmox VE 8.4 reaches end of support in August 2026. That isn't a nudge to upgrade eventually — it's the point where your hypervisor stops getting fixes. The good news is that 9.2 is a genuinely strong release rather than a box-ticking one, and it closes the gap people leaving VMware have been complaining about since the Broadcom licensing exodus started.
The DRS-shaped hole finally gets filled
The Dynamic Load Balancer is the reason to care. Proxmox's Cluster Resource Scheduler now uses real-time node and guest utilization to automatically migrate HA-managed guests between nodes. If you moved off VMware in the last couple of years and have been manually shuffling VMs whenever a node gets hot, this is the feature that makes the migration feel finished rather than merely done. Proxmox's own roadmap has the full change list, and it's worth reading before you plan the window.
I'd argue this is the single most under-discussed release in the homelab world this year, precisely because 'automatic load balancing' sounds like an enterprise checkbox. In practice it removes a recurring chore. Nobody enjoys noticing at 11pm that one node is pinned while another idles.
The SDN work in 9.2 matters most if you've outgrown a flat network — which happens sooner than most homelabbers expect. · Unsplash
What else landed
SDN got serious: WireGuard and BGP fabric connectivity, route maps and prefix lists for BGP/EVPN filtering, OSPF route redistribution, and an IPv6 underlay for EVPN.
New arm-ha / disarm-ha commands let you run planned cluster-wide maintenance without triggering node fencing. If you've ever had a maintenance window turn into an incident, this one's for you.
Custom CPU models are now creatable and editable in the web UI, with cluster-wide compatibility checks — no more hand-editing config to keep migration working.
Ceph Tentacle 20.2.1 is production-supported for fresh installs, alongside Squid 19.2.3.
How I'd actually approach the upgrade
Proxmox major upgrades are usually uneventful, and I don't want to make this sound scarier than it is. But 9.2 crosses a Debian release — to 13.5 'Trixie' — and bumps the kernel to 7.0, so it deserves more care than a routine apt upgrade on a Tuesday evening.
Read the official upgrade guide for your exact version path rather than a forum post — Proxmox documents these properly and the edge cases are version-specific. If you're running Ceph, check its upgrade ordering separately; storage and hypervisor upgrades want to be sequenced, not simultaneous.
So my honest take: do this one on the deadline rather than after it, and do it because the load balancer is genuinely useful rather than because support is ending. The deadline is just what gets it onto your list.
Common questions
What happens if I stay on 8.4 after August 2026?
It keeps running — nothing switches off. But you stop receiving fixes, including security fixes, which is a poor position for a hypervisor holding everything else you run.
Is the upgrade risky?
It's a major version that crosses a Debian release and a kernel bump, so it deserves a maintenance window and a tested backup. It is not generally considered a hazardous upgrade, but it's not a routine patch either.
Do I need a cluster to benefit?
The Dynamic Load Balancer specifically needs multiple nodes with HA-managed guests. Single-node users still get the Debian 13.5 base, kernel 7.0, ZFS 2.4 and the UI improvements.
Can I skip 9.x and wait for the next release?
I wouldn't. 8.4 support ends in August, and skipping major versions makes the eventual upgrade path longer and more fragile, not shorter.
Is Proxmox still a genuine VMware replacement?
With automatic load balancing landing, the most-cited functional gap against DRS is closed. Whether it replaces VMware for you depends on what else in that stack you rely on — but the core hypervisor case is considerably stronger than it was two years ago.
