My Homelab in 2026: More Fiber, Fewer Regrets

Every January I tell myself I’m going to simplify the homelab, and every January I write one of these posts describing more hardware than last year. This is the 2026 edition of that tradition: what’s on the shelf, what’s running on it, and what I broke along the way.

If you’re new here, the short version is that my homelab started as “one old desktop running Plex” and has since metastasized into the thing that runs my house. When it goes down, the lights get dumber, the dashboards go dark, and I hear about it from everyone else who lives here. That’s the real SLA.

#The hardware: a shelf pretending to be a rack

I still don’t own a proper server rack, and honestly I’ve stopped pretending I’m going to buy one. The lab lives on a sturdy shelf setup in my office: a handful of small-form-factor machines, a NAS, and a UPS that has saved me from at least three brownouts this year.

The current lineup looks roughly like this:

A few mini PCs doing the heavy lifting as virtualization hosts. Low power, quiet, and shockingly capable for their size.
A multi-bay NAS handling bulk storage, mostly spinning rust with an SSD cache in front of it.
A UPS big enough to let everything shut down gracefully instead of face-planting when the power blips.
Networking gear, which got the biggest upgrade this year (more on that below).

The mini PC approach has been the best decision I’ve made in this hobby. I used to run a repurposed tower that doubled as a space heater. The minis idle at a fraction of the power, and I can add capacity by adding another node instead of forklifting the whole thing.

#Proxmox is still the boss

Everything runs under Proxmox, and 2026 is the year I stopped even window-shopping alternatives. The hosts are clustered, which mostly means I can migrate VMs off a node before I inevitably do something dumb to it.

My layout is deliberately boring:

A couple of VMs for the things that want to be VMs: a Linux box for general tinkering and a dedicated VM for Docker workloads.
LXC containers for small, single-purpose services where a full VM would be overkill.
One sacrificial “lab” VM where experiments go to live and, usually, die. Nothing graduates to the real stack until it survives a few weeks there.

The sacrificial VM rule exists because of past me, who once tested a storage driver on the same host that ran Home Assistant. Past me spent a Saturday explaining to the household why the lights wouldn’t turn off. Current me tests things in quarantine.

#The Docker stack that actually runs my life

This is the part people usually ask about. The service list shifts constantly, but the core has been stable for a while now:

Media: Jellyfin for streaming, with the *arr stack (Sonarr, Radarr, Prowlarr, and friends) handling the library management. Once you get the *arrs dialed in, the whole pipeline just quietly works, which is more than I can say for most software I touch.
Home Assistant: the single most important service in the lab. It runs the automations, the dashboards, and increasingly the thermostat schedule. It is also the reason homelab downtime is a family matter.
Reverse proxy: everything sits behind a reverse proxy with automatic certificates, so internal services get real HTTPS and I never have to look at a browser security warning on my own network again.
Monitoring: Uptime Kuma for “is it up,” plus a Grafana and Prometheus setup for the prettier “why is it slow” questions. I check the graphs more than I’d like to admit.
The long tail: a password manager, a recipe app, a document archive, and the usual pile of small self-hosted tools that each replace one subscription I got tired of paying for.

The theme this year was consolidation. I killed off several services I was running purely because a Reddit thread told me to, and the stack is better for it. If I haven’t opened a service’s UI in three months, it gets archived. That rule alone reclaimed a surprising amount of RAM and mental overhead.

#Networking: the year fiber changed everything

The biggest quality-of-life upgrade didn’t happen in the rack at all. In August 2025 I got T-Mobile Fiber installed, and going from my old connection to symmetrical fiber genuinely changed how I use the lab. Offsite backups that used to crawl overnight now finish before I’ve made coffee. Remote access stopped feeling like remote access.

The catch is that the fiber terminates downstairs, and my office, along with the entire homelab, is upstairs. Right now there’s a temporary link bridging the two floors that works fine but offends me on principle.

Fixing that properly is already my first project of 2026. I want an actual fiber run going upstairs, terminated cleanly, so the lab hangs directly off the fastest link in the house instead of sitting behind a compromise. I’ll write that one up separately once the drywall dust settles.

#Backups, or how I learned to stop trusting myself

I follow 3-2-1 like it’s a religion, because I’ve been the guy without backups before and I’m not doing that again. Three copies of anything that matters, on two different types of storage, with one copy offsite.

In practice that means nightly jobs from the Proxmox hosts to the NAS, a second local copy on separate disks, and an encrypted sync to cloud storage for the irreplaceable stuff. Photos, documents, and Home Assistant’s config get the full treatment. Media doesn’t, because the *arr stack can rebuild the library and my wedding photos cannot be re-downloaded.

The part most people skip: I actually test restores. Twice this year I pulled a random VM backup and brought it up on a different node just to prove I could. Both worked, and the peace of mind was worth the hour each time.

#Local AI: the experiment that stuck around

I spent a good chunk of the year playing with self-hosted AI, mostly running local models through Ollama with a web UI in front. I went in expecting a novelty and came out with something I use weekly.

The honest assessment: local models are not going to replace the frontier stuff for hard problems. But for quick summarization, home automation helpers, and anything involving documents I’d rather not upload to someone else’s cloud, they’re legitimately useful. The privacy angle is the killer feature, not the raw capability.

The limiting factor is hardware. Everything I run is modest-sized because the lab’s GPU situation is, to put it kindly, aspirational. A proper inference box is on the maybe list for this year, pending me winning an argument with my own budget.

#The power bill reality check

Because someone always asks: the whole lab, networking included, idles somewhere in the low hundreds of watts. That’s after the move to mini PCs, which cut the draw substantially from the old tower-based setup. At my electric rates it works out to less than most people spend on the streaming subscriptions this stack replaced, which is the math I use to justify all of this.

#What’s on the list for 2026

Three takeaways from this year, and where things go next:

Consolidation beat expansion. Fewer, better-maintained services made the lab more useful, not less.
The network is the foundation. Fiber made every other part of the lab feel faster, and finishing the upstairs run is priority one.
Local AI earned a permanent spot, and it’s the most likely reason I buy hardware this year.

If you’re running your own lab, I’d genuinely love to hear what your stack looks like in 2026, especially if you’ve solved the “fiber is on the wrong floor” problem more elegantly than I’m about to. And if you’re thinking about starting one: grab any old machine, install Proxmox, and break things. That’s all any of us are doing.