Most people keep their photos at Google, their mail at Gmail, their documents in whatever cloud came with the phone and their passwords in a browser built by an advertising company. Every byte sits on a server they don’t own, under terms they didn’t read, in a country they didn’t check. Nobody finds that strange anymore, because the alternative looks like a second job.
I want to talk about that alternative, and about why it sits behind a wall most people will never climb.
The deal underneath the free storage
The services are free because the business model is knowing you. Your mail gets scanned, your photos get analysed, your location history is a product with a price on it. Nobody hides this anymore, and most people have made a rough peace with it. The data is convenient where it is, and moving it sounds like more hassle than it’s worth. I get that. It still bugs me.
The technical crowd has an answer, and it’s a good one. Self-hosting. Run the services on your own hardware and the surveillance problem mostly goes away, because there’s nobody on the other end to do the surveilling.
I’ve run my own infrastructure for years. My own email, my own cloud storage, my own password vault, the lot. I love it, and I’ve written a fair bit about why on this blog. But I’m not going to pretend my parents could do the same. Self-hosting today is a privilege. It costs technical knowledge and a steady supply of evening hours, and most people have neither to spare. The barrier is real, and it’s high.
The other side of the wall
Picture it. You pay a monthly fee, somewhere between €15 and €30. For that you get a private cloud for your photos, documents and notes, your own email server, a password manager, calendar and contacts. The same things people rent from Big Tech today, with four differences underneath.
Your data stays yours, so no scanning, no profiling, no quiet resale to a data broker. It runs on commodity hardware, and the old PC in your closet qualifies. You can leave whenever you want, take your data and walk, no lock-in. And the operations are transparent, open source and auditable, no black boxes.
None of this is new. Nextcloud, Mailcow and a long list of mature open-source projects have covered the building blocks for years. What’s always been missing is the person who assembles them and then keeps them running. The whole idea is that this person stops being you.
The cost shows up after the weekend
Setup is a weekend. A determined person with a tutorial gets there. The cost comes afterwards, every week, for as long as the thing exists. Certificates expire. A disk fills up while you’re asleep. An upstream image ships a breaking change and the contacts app stops syncing on the day you need a phone number. A self-hosted setup is something you keep alive, week after week.
That maintenance load is why I’ve never handed a Raspberry Pi to a non-technical friend and called it a present. What they’d be getting is a pager.
This is the same friction that drives most of my technical choices. I want systems that fail gracefully and heal themselves, because otherwise I’m the single point of failure for someone else’s email, and that doesn’t hold up. It isn’t kind to either of us.
Self-hosting is good. The open question is who carries the operational weight.
One service cluster, many customer clusters
I’ve started building an answer. This is the shape of it.
A central service cluster manages a fleet of customer clusters. Each customer cluster runs on hardware the customer owns, whether that’s a mini PC at home, a server in an office or rented hardware in a datacenter.
The service cluster handles the parts that hurt. Deployment, GitOps-driven, declarative and reproducible. Monitoring, so something gets caught before the customer notices. Updates, automatic, tested and rolled back when they misbehave. Backups, encrypted, off-site and restored once in a while to prove they still work.
The customer cluster does two things. It runs the services, and it stores the data on hardware the customer can physically touch.
That split is the whole point. The service cluster never sees customer data. It manages the infrastructure, and the data stays on the customer’s side of the line. The babysitting is done by me and my tooling, and the person who just wanted their photos to stay private gets exactly that.
Running on old hardware
One thing I won’t give up. This runs on hardware people already own.
The old laptop, the retired office PCs, the Raspberry Pi on the shelf, all of it qualifies. K3s runs comfortably on machines Big Tech wrote off as obsolete years ago. That keeps usable hardware out of the bin and brings the cost of entry close to zero. It also makes the ownership literal. The box is in your house, and nobody in a datacenter holds a lease on it.
I’ve written before about sovereign infrastructure and why physical control is the part that counts. This is the same idea pushed one step further, out of the homelab and into the hands of people who’ll never build one.
The part I haven’t proven yet
Now the bit I find exciting, with the caveat that it’s the least tested.
Today I’m the operator. I read the alerts, I debug the failures, I make the calls. That works for friends and family. It doesn’t work for everyone, because my attention is the bottleneck and there’s only one of me.
Large language models are getting good at the judgement operations work asks for. Reading logs and metrics, diagnosing a common fault, proposing a fix, weighing context. Picture a system that notices a customer’s disk creeping towards full, looks at the usage pattern, recommends a cleanup or an expansion, applies whichever gets chosen, and remembers how it went for next time.
The building blocks for that exist today. The guardrails don’t, and that’s the part I’m careful about. An LLM with write access to someone’s email infrastructure is a serious thing, and getting the balance between autonomy and safety wrong helps nobody. So this stays speculative until it has run on my own cluster for a while. I’m not asking anyone to trust something I haven’t broken myself first.
If it works, my role shifts from operator to architect. I design the policies, the system handles the rest, and adding a customer stops adding a customer’s worth of pager duty.
What it costs
A vision that can’t pay for itself is a hobby with extra steps, so here are the numbers.
Individuals pay €15-20 a month for cloud storage, a password manager and photos. Families pay €25-35 a month for shared services plus individual accounts. Small businesses pay €50-100 a month for email, collaboration and CI/CD at a quality you’d run a company on.
Compare that with the going rate. Google Workspace is €5-18 per user per month, paid partly in your data. Microsoft 365 is €5-20 per user per month, same arrangement. Self-hosting it yourself is “free” in euros and costs hundreds of hours.
I’d rather be honest about the trade-off than sell past it. On a spreadsheet this isn’t the cheapest option. It asks you to pay for something the surveillance economy gives away, and what you get back is sovereignty without the operational burden. For some people that’s obviously worth it. For others it never will be, and that’s a fair call.
Where this stands
I’m building this in stages, on purpose, slowly enough to break things in private. Right now my own service cluster manages a handful of customer clusters for friends, family and a few early adopters. Next comes documenting the architecture and open-sourcing the management tooling. After that the LLM integration, once I trust it on my own infrastructure. And eventually, something anyone can sign up for and own.
If this resonates, whether you want to be an early adopter, help with the tooling or just watch, I’ll keep writing about it here, including the parts that go wrong.
I said I’d never hand my parents a Raspberry Pi. A box that’s theirs, data that’s theirs, and someone competent keeping it running without ever looking inside, that I would hand them. That’s what I’m building.