Picking Homelab Hardware Without Wasting Money
The used-enterprise trap, the mini-PC sweet spot, and why your first server should cost less than your monitor.
Homelab From Zero02 / 08TL;DR
Buy one used mini-PC, run it for three months, and let the workloads you actually run — not the ones you imagine — write the second shopping list.
The trap: used enterprise gear
The classic advice is a used rack server, because the price-per-spec looks unbeatable. The specs are real. So is everything the listing does not say: it idles at the power draw of a space heater, it sounds like one too, and it teaches you server hardware — which is probably not what you set out to learn.
You wanted to learn virtualisation, networking, and services. A machine that is expensive to leave on is a machine you turn off, and a lab that is off teaches nothing.
The sweet spot: one used mini-PC
An office-surplus mini-PC — the small business desktops sold by the thousand off corporate leases — runs a real hypervisor and a dozen light services, idles under a light bulb, and sits silently on a shelf. That is the whole starter list:
- CPU: whatever the mini-PC came with. It is enough.
- RAM: the one spec worth maxing. Virtualisation eats RAM long before CPU. 32GB if the price is close.
- Storage: one SSD. Redundancy comes later, when something is worth protecting.
- Network: the port on the back. Managed switches are a part-five problem, not a part-one problem.
What about GPUs?
Only if local AI is the actual goal — it is the one workload that genuinely needs one, and it changes the build completely: a mini-PC has no room, no power, and no cooling for it. That fork in the road gets its own part later in this series. Do not buy the GPU first and let it dictate everything else unless you already know that is the point.
FAILURE MODES
- Speccing for imagined workloads. The Kubernetes cluster you might run someday keeps not existing while the RAM you bought idles.
- Buying two of everything for redundancy. A lab that is allowed to break does not need a failover pair. That is restaurant thinking.
- The empty rack. Buying the rack before the machines is furniture shopping. The shelf was fine.
WHAT I LEARNED
Three months of real usage produces a better hardware plan than any amount of research, because it replaces guesses about workloads with a process list you can read.