Home Server Compute: Mini PCs, Used Enterprise Gear, and GPUs

The compute layer is where your services, virtual machines and models actually run. Every other pillar on this site assumes you have already made the decision this one is about.

Which is why this hub starts somewhere unusual for a buying guide: the purchase price is the smallest of the three numbers that matter, and for a lot of people the right answer is not to buy at all.

The three numbers

What it costs to buy. The one every guide leads with, and the one that stops mattering after the first month.

What it costs to own. Electricity for as long as it is switched on, and the idle hours dominate that: one watt left running for a year is 8.77 kWh, so a machine idling at 40 W costs about $105 a year before it does anything at all. Add storage, networking, a UPS, and the heat you may pay again to remove.

What the alternative costs. Renting the same work. This is the comparison that decides it, and it is the one nobody runs before ordering.

Cloud GPU vs owning hardware puts all three together, and I will spoil the finding because it is the honest one: at light usage, the monthly electricity on a machine you own exceeds the API calls it would replace. There is no crossover to wait for. Three of the nine cells in its table say “never”.

I earn affiliate commission on hardware, which is exactly why that article leads with the case against buying.

When buying is right anyway

Cost is frequently not the deciding factor, and none of these appear in any of the arithmetic:

• Data that cannot leave the building, for policy or law
• Work that has to continue without a network
• Latency a round trip cannot meet
• A model nobody offers as a service, or one you have fine-tuned
• Wanting to, which is a perfectly good reason to buy a computer

What the arithmetic gives you is an honest price for those reasons, rather than a story about saving money that does not survive a spreadsheet.

What belongs in this pillar

• The buy-or-rent decision, with the electricity and the break-even worked through rather than asserted
• Buying guides: mini PCs, tower builds, rack servers and GPU picks, by budget
• Platform choices: bare metal against virtual machines against containers
• Starter builds: complete parts lists at fixed budgets
• The cost of ownership: idle draw, noise, and the rest of the lab a new machine drags in with it

Buy for idle, not for peak

If there is one habit this pillar wants to install, it is that one.

A machine’s specification sheet advertises what it can do flat out. Your bill is decided by what it draws while nothing is happening, because that is most of its life. Dropping 80 W of idle to 20 W saves around $145 a year at 30 cents per kWh, and no amount of tuning the working hours comes close to that.

The corollary is that a used enterprise server is cheap to buy and expensive to own. It is loud, it draws more at rest than a small modern machine, and both of those are permanent. That is a real trade rather than a reason to avoid one, and the way to make it is to measure rather than guess: put a plug meter on the thing and use the power cost calculator.

The tools

• Local vs cloud break-even: the buy-or- rent question, which says “never” in as many words when it means never
• Power cost: what a candidate machine costs to own, and what share of that is idle
• Model fit: whether a machine you are considering can actually hold the models you want, before you buy it

What is not here yet

Every hardware review and every “best machine” list, because those need those machines on the desk and measurements from them. The first published run is Spark inference, not a mini-PC review. This pillar currently gives you the framework for a decision and declines to make the decision for you.

FAQ

Should I start with a mini PC or a used server?

Start with a mini PC unless you know you need drive bays or PCIe slots. Small modern machines are quiet and cheap to keep running, which matters more than their purchase price because it recurs. A used rack server is cheap to buy and expensive to own: it is loud, it draws more at rest, and both are permanent. Measure the idle draw of anything you are considering rather than trusting a specification.

Do I need a GPU in my home server?

Only for specific workloads: hardware video transcoding, local AI inference, or remote desktops. For everything else the money is better spent on memory and storage. If it is for local AI, check first whether the models you want actually fit the card you are considering, since a machine that cannot hold the model is not cheap at any price.

Is it cheaper to run AI locally than to pay an API?

At high volume, usually yes. At low volume, usually no, and often never. A machine idling at 40 W costs about $8.77 a month whether you use it or not, and light usage against a cheap model does not generate enough API spend to repay that, let alone the hardware. The crossover depends entirely on how many tokens you actually produce in a day.

How much does a home server cost to run per month?

Multiply its standing draw in watts by 0.73 to get monthly kWh, then by your unit rate. A 20 W device is about 14.6 kWh a month, so roughly $4.38 at 30 cents per kWh. For a machine that also does heavy work, split it: idle watts for the hours it rests, load watts for the hours it works.

Is it cheaper to leave a server on or turn it off between uses?

Off, and usually by more than people expect, because idle is most of the bill. A machine idling at 40 W costs $105 a year to leave switched on before it does a single unit of work. If you use it an hour a day, that is $105 spent on the twenty-three hours you did not. The counter-argument is convenience rather than cost, and wake-on-LAN answers most of it.