Proxmox vs. ESXi in 2026 for the home operator

I ran ESXi at home from 2019 through early 2024. It worked. It was quiet. Its management UI was stable in a way that homelab software rarely is. Then Broadcom bought VMware, the free edition evaporated, and the writing on the wall turned into a neon sign. I gave Proxmox a weekend in late 2024. I haven't been back.

This is the full report. Every number is from the rig you've seen on this site — Minisforum MS-01, i9-13900H, 64 GB DDR5, ZFS mirror — and every command is what I actually typed.

What changed with Broadcom, in one paragraph

ESXi Free — the single-socket, no-vCenter, no-support SKU that thousands of homelabbers ran — is gone. Broadcom removed the standalone "Free" edition from their download portal in February 2024. You can still run ESXi 7 if you hoarded an installer and a license key, but VMware's release cadence has visibly slowed, the community forums went quiet, and every "how to download ESXi 8 free in 2026" blog post now ends with "see how to switch to Proxmox." The writing was on the wall for any home operator whose long-term plan was "stay on ESXi for another decade."

Proxmox in three sentences

Proxmox is a Debian-based hypervisor that gives you KVM virtual machines and LXC containers in one web UI. It's free and open-source; the paid "Subscription" is enterprise support, not feature-gating. Version 8.2 is what I'm running, Debian 12 underneath, kernel 6.8, and it's been quietly stable for nine months.

Migration plan: one service at a time

I did not do a clean cutover. I ran ESXi and Proxmox in parallel for six weeks on two machines, moved services one-by-one, and only powered ESXi down once everything stopped depending on it. The order that worked:

1. Home Assistant first, because if HA is down my partner notices within ten minutes.
2. Jellyfin next, because I wanted to validate hardware transcoding on the new host.
3. Tailscale relay + DNS third, because those services need to be up for everything else to reach each other.
4. The *arr stack, then Immich, then Nextcloud, then everything else.

For each service I did the same dance: export the VM/LXC config on ESXi, recreate it as a Proxmox LXC or KVM guest, restore from Borg backup, verify, flip DNS in Tailscale. Cutover for Home Assistant took 40 minutes. Jellyfin took longer because I reinstalled from scratch to get the native LXC setup instead of the VM I'd been running.

LXC is the reason I'd recommend Proxmox even without Broadcom

ESXi does not do OS-level containers. Everything is a full VM. That means every service carries a full OS overhead — kernel, init, systemd, the works. On a 64 GB host running twelve services, that overhead adds up fast.

Proxmox gives you LXC as a first-class citizen. An LXC container is essentially a chrooted Debian install sharing the host kernel. It starts in under a second, eats under 100 MB of RAM at idle, and snapshots via ZFS behave like git tags — atomic, fast, free.

The workflow I use before any risky upgrade:

# 1. Snapshot the container's ZFS dataset
zfs snapshot rpool/data/subvol-101-disk-0@before-ha-upgrade

# 2. Do the scary thing (dist-upgrade, add-on install, config rewrite)
pct enter 101
apt update && apt full-upgrade -y

# 3a. If it works, drop the snapshot
zfs destroy rpool/data/subvol-101-disk-0@before-ha-upgrade

# 3b. If it broke, roll back — takes ~5 seconds
pct stop 101
zfs rollback rpool/data/subvol-101-disk-0@before-ha-upgrade
pct start 101

I have used step 3b more times than I'll admit — broken Home Assistant updates, a Jellyfin database migration that truncated the watch history, a Proxmox kernel that wouldn't boot on the i9. Every time, the fix was under ninety seconds.

KVM is still there when I need it

A handful of services can't live in an LXC container — Windows VMs for testing, anything that wants to load a custom kernel module, anything that wants PCI passthrough. Proxmox does KVM exactly as well as ESXi did, without any of ESXi's licensing friction. My one KVM guest right now is a Windows 11 VM I spin up twice a month to run an accountant's tax tool.

The parts of ESXi I miss

Honest accounting. There are three things ESXi did better:

• vSwitch is more polished than Linux bridges. Proxmox network config, if you touch it outside the web UI, is raw /etc/network/interfaces with all the joy that implies. I still occasionally make it yell at me.
• vMotion on free ESXi. Wait — that wasn't actually free. Never mind.
• The web console was faster. Proxmox's web UI is Perl + JavaScript and feels older. Functionally fine; cosmetically dated.

Nothing on that list is worth the migration back.

What I'd tell a homelab operator sitting on ESXi today

If you're on ESXi 7 and it's still working, there's no fire. You can coast another year. But every month you stay is a month further from the community, a month with a shrinking pool of tutorials, and a month closer to a hardware change that won't have ESXi drivers.

Move when you have a weekend free. Take one service over at a time. Keep both hypervisors running in parallel until every dependency has migrated. The transition is not dramatic — it's just work. And when it's done, you'll have LXC, you'll have free ZFS snapshots, and you won't have Broadcom invoices in your future.

Next in this series: Tailscale subnet routers and ACLs in anger, which is the networking layer that made the ESXi→Proxmox migration feel genuinely trivial on the DNS side. And for the full stack, start at The self-hosting stack I actually use in 2026.