The Real Cost of Self-Hosting: Electricity, Time, and What You Actually Save

There is a story self-hosters tell themselves, usually at about 1am while staring at a rack of blinking lights, and it goes like this: “I’m saving so much money by not paying for all these subscriptions.” It is a lovely story. It is also, in my experience, mostly fiction — or at least a great deal more complicated than the version we like to repeat at dinner parties to justify the noise coming from the cupboard.

I self-host a ridiculous amount of my life. I am not here to talk anyone out of it. But I am here to do the sums honestly, because the case for self-hosting is real and it does not need the inflated numbers we usually attach to it.

Electricity is the bill you forget

The hardware cost is a one-off you notice. The electricity is the cost you don’t, because it dribbles out a few pence at a time. So let’s actually measure it.

A box that idles at 40 watts and occasionally spikes higher will average, generously, around 50W over a day. The arithmetic is the same every time:

1
2
3
4

power (kW) x hours x days x price = annual cost

0.050 kW x 24 h x 365 days = 438 kWh/year
438 kWh x £0.28/kWh         = £122.64/year

That is one modest always-on machine. Add a NAS spinning eight drives, a second mini-PC, a managed switch and a UPS quietly wasting a few watts of its own, and a hundred-watt average is easy. At UK prices that is roughly £245 a year, every year, before you have hosted a single useful thing. A Raspberry Pi sips perhaps 4W and changes this maths entirely — which is rather the point of measuring rather than guessing.

And you must measure, because nameplate figures lie in both directions. A power supply rated at 550W does not draw 550W; it draws whatever the components ask for, which at idle might be a tenth of that. Meanwhile a “low-power” NAS with the drives spun up, a couple of Docker containers doing real work, and a transcode running for the family film night can briefly triple its idle figure. The only honest number comes from a cheap plug-in energy monitor left in place for a week, capturing the real duty cycle rather than the moment you happened to look. Mine told me my “efficient” mini-PC was averaging nearly double what the spec sheet implied, entirely because of a background indexer I had forgotten I was running.

There is also a seasonal wrinkle worth a line. In a British winter the heat a server dumps into a room is not entirely wasted — it is a few watts you are not paying the heating for. In summer, in a warm cupboard, it is a few watts you might end up paying to cool. Neither effect is large, but if someone tries to tell you the electricity is “basically free because it heats the room”, that is true for roughly four months of the year and quietly false for the rest.

Your time has a price even when you don’t charge it

The line item nobody puts in the spreadsheet is the hours. The initial build is fun, so we don’t count it. But self-hosting is not a purchase, it is a relationship, and relationships need maintenance.

Realistically, expect to spend time on:

• OS and container updates, and the occasional one that breaks something
• Certificate renewals that silently fail at the worst moment
• Backups, and the far rarer but more important act of testing a restore
• A drive that dies, a power cut that corrupts a database, an ISP that changes your IP

Even a tidy setup eats an hour or two a month once you account for the 3am “why is the reverse proxy returning 502” sessions. Call it twenty hours a year. You needn’t bill yourself consultant rates, but pretending the number is zero is how people end up resenting their own homelab.

The distribution of those hours matters as much as the total. Maintenance time is not smooth; it arrives in ambushes. Eleven months of near-nothing, then a weekend swallowed whole by a failed distribution upgrade, a database that will not come back cleanly, or a certificate authority that changed something upstream and broke every renewal at once. The average might be twenty hours, but the experience is a quiet hum punctuated by the occasional emergency, and the emergencies have a talent for landing when you are least free to deal with them. I have spent this at length before, because it is the single most under-counted cost in the hobby: self-hosting is not free once you honestly account for your own time, and the people who burn out are almost always the ones who budgeted zero.

There is a way to shrink the number, and it is worth stating because it changes the sums. Automation front-loads the effort: unattended upgrades, automated certificate renewal via ACME, a monitoring stack that pages you before the disk fills rather than after, and configuration kept in version control so a rebuild is a script rather than an archaeology dig. That work costs you hours up front and repays them for years. A homelab you can rebuild from a git repository in an afternoon is a fundamentally cheaper thing to own than one that lives only in your memory and a pile of undocumented tweaks.

So what do you actually save?

Now the other side of the ledger, honestly. Compare like for like against the SaaS you would otherwise pay for:

Stack a few of those and you are displacing perhaps £200–£350 a year of subscriptions. Against £245 of electricity and twenty hours of your life, the pure financial case is often a wash. If you are hosting one app to save a fiver a month, you are losing money and should feel no shame in admitting it.

But that table hides two effects that push the sums in opposite directions, and honest accounting has to name both.

The first is the upfront capital, which the running-cost view conveniently forgets. A second-hand mini-PC, a NAS, four drives, a UPS and a decent switch is not nothing — realistically £400 to £800 to start, plus the drives you will replace as they age. Amortise that over the three or four years the kit will last and you are adding another £150–£250 a year to the true cost, at least early on. The financial case does not turn positive until the hardware is paid off, which is precisely the point at which people tend to buy more hardware.

The second effect runs the other way and is the strongest quiet argument for the whole exercise. A password manager for a family costs a fixed subscription; so does photo storage; so does a notes service. Each one, hosted, is a marginal load on a box you are already running. This is the same structural logic that makes subscriptions creep in the first place — per-seat and per-service pricing scales linearly while your own fixed costs barely move — and it is exactly why self-hosting gets better value the more you consolidate onto one machine. The tenth service on an existing box is nearly free; the first service on a box bought specifically for it is the most expensive thing you will ever self-host.

Where the case genuinely holds up

The honest argument for self-hosting was never really about money, and the moment you stop pretending it is, the decision gets clearer. You self-host because:

• You’re already paying the fixed cost. Once the box is on for one thing, the marginal cost of the tenth service is nearly zero. Economies of consolidation are real.
• You own the data. No vendor deciding your photos now cost more, no service shutting down with ninety days’ notice, no quiet training of a model on your files.
• You learn things that have actual market value. The skills are transferable in a way a Netflix subscription is not.
• You enjoy it. This is allowed to be a reason. A workshop does not have to turn a profit.

The hidden costs that ambush the spreadsheet

A few line items never make it into the “look how much I save” story, and they are the ones that turn a modest win into a loss if you ignore them.

• Failed drives. Spinning disks die; it is a question of when. Budget for a replacement every couple of years across an array, and budget for the fact that a proper setup means buying the redundant drive you hope never to need. A degraded array being rebuilt at 2am is the maintenance ambush in its purest form.
• Backups you actually test. Off-site backup is a real cost, whether that is a second machine, a cheap cloud bucket for the encrypted archive, or the electricity and drives for a mirror. An untested backup is a liability wearing the costume of an asset; the day you discover the restore does not work is the day the whole ledger goes negative.
• The upgrade trap. The urge to buy faster kit for a workload that was fine is a genuine cash leak, and one I have argued against at length — good enough is very often good enough. Every “while I’m at it” upgrade resets your amortisation clock and pushes the break-even date further out.
• Bandwidth and IP oddities. A residential connection that throttles uploads, an ISP that rotates your IP, or a data cap can quietly cost you either money or a workaround’s worth of hours.

None of these are dealbreakers. All of them are real, and all of them are missing from the napkin maths people wave around at 1am.

The honest verdict

Self-hosting one or two services to save money is a bad trade once you price in power and time. Self-hosting a dozen, on hardware you already run, for the data ownership and the learning and the sheer pleasure of it — that is a fantastic trade, and the electricity bill is a hobby cost, not a saving.

So measure your watts, be honest about your hours, and then choose with your eyes open. Put the plug meter on the box for a week, write down the number of evenings you actually lost this year, add the hardware you have quietly bought, and set that against the subscriptions you have genuinely cancelled rather than the ones you imagine you might one day. Do that and the decision becomes clear and, more often than not, still worth making — just for reasons that survive contact with a spreadsheet. Just stop telling yourself it’s free. It isn’t, and it doesn’t need to be.