Are my own servers worth it?

I've spent a while explaining how my machines are set up, and the question I get most isn't technical: it's whether all this is worth it. Time to do the maths, in money and in hours, and say frankly what I'd do if I started today with half the budget.

The videos sell it like this: "set up your AI server at home and forget about paying for subscriptions". You buy a machine, install a couple of programs and live for free forever. If only.

The honest question is a different one: is it worth having my own servers, at home and outside, or would I have been better off with just one option?

In case you're just arriving, a quick summary. In Why I have eight servers at home and six outside I explained how I ended up with two networks. In How I set up a new VPS, the steps I follow every time I rent one. In Linux or Windows on my servers, why I choose one system or the other. And in What runs on each of my machines, how the tasks are split. This is the balance sheet.

I have an internal network of 8 servers and an external network of 6 VPS. A VPS, for newcomers, is a slice of a large server that you rent from a provider and that behaves like a whole computer for you, on day and night and connected to the internet.

The split, without repeating what I covered in What runs on each of my machines: at home are the local models, the backups, client data that mustn't leave and the testing; outside are the websites, the agents that have to respond at any hour and the services that serve clients over the internet. The underlying rule is simple: whatever has to be always reachable goes outside; whatever is heavy or sensitive stays inside.

These figures come from my invoices, not from catalogue prices. As of August 2026, the 6 VPS cost me: two small machines at €6, three medium ones at €14 and a large one at €38; four at Hetzner and two at IONOS, all in data centres in the European Union. In total, about €92 a month including tax. Reviewed in September 2026: if you're reading this later, assume the prices have moved.

It hasn't grown in a straight line. I started in 2023 with a single €5 VPS; by 2024 I had four and in 2025 I reached six. The first four are at Hetzner; I put the last two at IONOS because I already had domains and email there. And there's a less-told side of renting: someone else decides the price.

What my contract includes and what it doesn't, to be checked in each provider's terms: the fixed IP is included; traffic has a generous monthly limit I've never come near; I have the provider's automatic backups on all six machines and they're included in that €92. The other layer of backups I do myself, towards home. And that's where the other network comes in.

Time to take apart the most widespread idea: your own server isn't free once you've bought it. It looks free because no invoice arrives with its name on it, but the money goes out anyway through other doors.

First, the purchase. Across the 8 machines I must have spent about €5,800; almost four in every ten euros, about €2,300, are the server with the graphics card, card included. Spread over five years of useful life, that comes to about €97 a month. That's depreciation: spreading the purchase price over the months you'll use the machine.

Second, electricity. I measured consumption with a plug-in meter for a month: the internal network uses on average about 11 kWh a day, more when the local models are working hard. At my rate, about €50 a month.

Third, breakdowns and spare parts. In three years: two dead drives, a power supply that gave me quite a scare and a fan that sounded like a light aircraft. About €400 in total, which spread out comes to about €11 a month.

Fourth, the connection. I paid for a fibre upgrade with a fixed IP that I didn't need before: about €15 more a month.

And fifth, what never appears on any invoice: noise and heat. In summer the server room reaches 31 degrees and I've ended up moving them twice. I don't add it up, but it's there.

Putting it all together, including what these sums usually leave out:

Almost double the VPS. Not bad for "forgetting about subscriptions".

I leave the hours out on purpose: they don't come out of the bank, they come out of what I could be doing instead. If you want to put a price on them, multiply them by what you charge per hour.

The cost that weighs most isn't money, it's time. My honest estimate, logged over the last six months: about 6 hours a month on the home network and about 3 on the VPS. Updates, checking that backups actually ran, looking at logs and dealing with scares.

The scare that sums it up best: one Sunday in March 2026 the power went out in the neighbourhood for twenty minutes. When it came back, five servers booted fine and three didn't. One had a damaged system drive and another had lost its network configuration. I spent the whole afternoon recovering them, laptop on my knees, with a backup that, thank goodness, had actually run on the Friday. Four and a half hours of Sunday. Since then I have an uninterruptible power supply, another expense that wasn't in the plan.

The comparison with the VPS is interesting. A VPS takes the hardware off my hands: if a drive breaks in the data centre, I never find out. But it doesn't take security off my hands. As soon as a machine is on the internet it starts receiving break-in attempts, and updating, closing ports and reviewing access is still my job. Fewer hours, yes; zero, never.

Here's what justifies having both networks, and why I wouldn't change it.

The home network gives me three things. Data that never leaves the building, which for some clients is a condition for working together. Graphics power to work locally with Ollama, LM Studio, llama.cpp and Unsloth, with which I've tested more than 50 models without paying per query. And independence from a provider changing its terms. An example: for a client in the healthcare sector I process internal documents with a local model; those documents have never left my network, and that was precisely what convinced them.

The VPS give me the opposite: permanent availability and exposure to the internet without opening up my home network. An example: the agent that handles enquiries on a client's website lives on a VPS; if my power goes out at home tomorrow, it keeps answering.

Usage and availability, in numbers: the graphics server works hard for about three or four hours a day on average. So far in 2026, the home network has had two outages I noticed, the March power cut and a fibre outage lasting half a morning; the VPS, none longer than a few minutes.

For the graphics side, the fair comparison is renting a server with a similar card. I looked into it in September 2026: by the hour it came out about the same as what that machine costs me in electricity and depreciation; by the full month, quite a lot more. And neither option solved the data that mustn't leave home.

With just one option I'd have to give up one of those two things or force it a lot.

None of this means you need 14 machines. I'm thinking of three profiles.

The freelancer who needs a website or an agent that's always available: I'd recommend a VPS, or two at most, and nothing at home. Their problem is being reachable at any hour, not power.

The small business with client data it doesn't want to take out of the office: a modest server of its own inside, for that data and maybe a local model, and a VPS outside for whatever faces the internet. And someone with hours assigned to maintaining it, even if only a few.

Someone tinkering to learn: one machine at home, the one they already have if possible, and a cheap VPS to practise the internet side. You learn a lot and spend little.

To decide, I find four questions useful: what data you'll handle and where it can live, how much you can pay once you count electricity and hours, whether it has to be available all the time and whether it will be exposed to the internet. And one clarification: it isn't all or nothing. Combining is the norm.

I'd first set up two VPS, one for websites and another for agents, and a single machine at home with a good graphics card for the local models and the backups. I'd leave out the test servers that are switched off most of the time. And I'd spare myself my worst mistake: buying hardware before knowing what was going to run on it.

With all this I've learnt that the question isn't where, but what and with which data. If you want to see how I ended up with two networks, it's in Why I have eight servers at home and six outside.

Verdict: it's worth it if you know why you have each machine; if not, it's an expensive hobby with fans.