The Sovereign Cloud
Can europe build a sovereign cloud on open source primitives without turning operations into the real product?
Is this a real business if Europe can get sovereignty from hour one for about €50 a month — or does the moment you leave AWS, the hidden cost become your own time and on-call burden?
The trap is that cheap hardware can make the idea look easy, while support, reliability, and bursty workloads decide whether it scales beyond a niche.
Sees the product as freedom from lock-in, with a strong demo and a narrow wedge before any grand infrastructure ambition.
Focuses on customer segmentation, support burden, and whether the economics work once real people need help at 3 a.m.
Treats the stack as feasible but warns the hard part is matching hyperscaler elasticity and automating operations.
The math that has to work
SARAH'S BACK-OF-NAPKINThe pitch gets real when a team can start on a Hetzner box in Germany for around €50 a month and be sovereign from hour one, outside the US CLOUD Act.
That only matters if the company can keep the experience boring, because sovereignty without reliability just moves the pain closer to home.
This week on Tri Diligence: the sovereign cloud — the idea that you don't need to rebuild Amazon to break free of it. About 90% of apps need only three primitives — object storage (S3), compute (Lambda), and a database (DynamoDB) — and mature open-source equivalents now exist for all of them (Garage / MinIO, Knative, ScyllaDB), plus containers for portability.
Our three hosts pull it apart:
- Jake (the marketer) — the genuinely liberating fact: about €50 a month on a Hetzner box in Germany and you're sovereign from hour one, outside the US CLOUD Act. And the bigger story — Europe has no hyperscaler of its own and builds its future on rented ground.
- Sarah (the backer) — the asterisk on "free and scalable": hardware is cheap, your time is not. Which of three plays is a real business — personal/SMB independence, a "sovereign PaaS" (a Heroku-for-Europe layer), or a full IaaS provider — and where the money actually is.
- Ryan (the technologist) — the real stack and its one genuinely hard problem (matching hyperscaler burst elasticity and scale-to-zero), plus the AI angle where AI compute and sovereignty collide.
Proof it's real: 37signals (Basecamp and HEY, led by DHH) left the cloud — about $600k on hardware, saving roughly $2M a year — and even they chose colocation, not their own server room. The memorable frame: the server room isn't the starting line, it's the finish line you grow into — and the first step costs less than dinner.
Each host ends with a verdict — invest, wait, or pass — and one concrete first step.
Transcript
Welcome to Try Diligence, the show where three people lovingly interrogate one business idea until it either stands up or asks for a lawyer. Today, a European sovereign cloud built from the essentials. Storage, compute, database, containers, on European soil, for less than a fancy dinner to start.
I like the political urgency. I don't yet like the word cloud, because that word usually means capex wearing a hoodie.
Builder reaction, this isn't insane. Garage or MinIO for object storage, Knative for scale-to-zero compute, ScyllaDB or PostgreSQL for state, Docker and O-C-I containers for portability. The parts exist.
The hook is gorgeous. Europe doesn't need to rebuild all two hundred A-W-S services. Most apps need three primitives and a way to move without being trapped.
Customer segments first. Not everyone in Europe wakes up whispering sovereignty into their pillow. The real early customers are regulated small businesses, agencies serving public clients, and software teams with steady workloads.
Also technical founders who currently have five services on Amazon and a monthly bill that makes them open a spreadsheet with fear.
Exactly. The first lover isn't the average bakery. It's the team that says, I want Heroku simplicity, but I want my data in Germany or Sweden, and I want an exit door.
That's a narrower market, which is good. Narrow means we can price it. Broad patriotic cloud talk is a conference panel, not a business.
Value proposition, then. Sovereignty plus portability plus pleasantness. Not just, here are some open-source bricks, enjoy your weekend in YAML prison.
Pleasantness is the product. An installer, sane defaults, automatic certificates, backups, observability, deploy from GitHub, rollback, and a dashboard that doesn't look like it was assembled during a power outage.
But the value proposition must include responsibility. If you move off Amazon, you didn't delete operations. You moved them into your kitchen.
True, but the narrative is empowerment versus responsibility. The server room is the finish line, not the starting line. Start on Hetzner for around fifty euros a month and get real sovereignty today.
Sovereignty from hour one is credible if the provider is European and the operator isn't subject to the CLOUD Act. Schrems Two and NIS two make that more than marketing.
Channels. I wouldn't buy billboards saying, escape the empire. Fun, but expensive. I'd sell through developer content, compliance webinars, GitHub, and partnerships with European hosting providers.
I want a killer demo. Deploy a real app on a rented German server in fifteen minutes. Then move it to a Swedish provider without changing code. That's the shareable moment.
The channel could include marketplace images for Hetzner, O-V-Hcloud, Scaleway, IONOS, UpCloud, Elastx, Cleura, and Safespring. Make the first deploy stupidly boring.
Customer relationship is tricky. Open-source community gets adoption, but paying customers want someone awake at three a.m. That means tiers. Community, professional support, and managed sovereign platform.
Community is still the moat starter. Templates, migration stories, public roadmaps, office hours, maybe a sovereignty scorecard for apps.
Fine, but support is where margins go to nap. If the five hundred euro a month customer consumes four engineer hours, you have built a charity with nicer stickers.
This is where AI helps for real. An operations copilot can read logs, suggest fixes, generate runbooks, flag bad configs, forecast capacity, and answer support questions from product docs.
And that becomes customer experience. Not hello, welcome to self-hosting, please suffer. More like, your cluster is healthy, your backup passed, and here's the one scary thing.
Revenue. I see three lines. A paid control plane, managed hosting on European bare metal, and enterprise support with compliance documentation.
Could price the starter at ninety-nine euros a month, professional at four hundred ninety-nine, and enterprise from two thousand. Migration projects are separate, because migrations are where optimism goes to learn humility.
For managed workloads, charge per node plus platform fee. But don't pretend it's serverless economics. You'll have idle capacity unless you overbook carefully.
Assume C-A-C through content and founder sales is eight hundred euros for professional accounts. If gross margin after support is sixty percent, payback can work in under six months. Paid ads would likely wreck it.
Key resources are boring and essential. A small elite infrastructure team, security credibility, upgrade automation, test clusters, incident playbooks, and a brand people trust with production data.
Brand trust is huge. This can't feel like a weekend script with a logo. It needs the calm confidence of a bank and the speed of a good developer tool.
Key activities are where I get nervous. Patching, backups, data durability, compliance, migrations, on-call, billing, and documentation. The company must become excellent at reducing operational surprises.
Data durability needs default multi-region backup across European providers. If one rented box dies and the customer loses everything, sovereignty becomes a very poetic lawsuit.
There's the tagline. Sovereignty, now with fewer poetic lawsuits. Put it on the mug, but maybe not the homepage.
Partnerships matter. Hetzner gets you cheap entry. Colocation partners get you rung two. Auditors and compliance consultants help sell into public-sector-adjacent buyers.
And upstream open-source communities are partners, not vending machines. Garage, MinIO, Knative, ScyllaDB, PostgreSQL, Kubernetes, Nomad. Contribute fixes or you become a fragile wrapper.
Cost structure is mostly people. Servers are cheap at the start. Good infrastructure engineers aren't. Neither are security reviews, support coverage, insurance, test environments, and incident response.
The 37signals proof point helps, though. Six hundred thousand dollars of hardware, roughly two million dollars saved per year, and they still chose colocation. That's a reality check, not fantasy fuel.
It shows steady workloads can beat hyperscaler economics. But spiky workloads are different. Open-source scale-to-zero exists, but matching Amazon burst elasticity without waste is hard.
Which means the business shouldn't claim cheaper for everyone. It should claim better control for teams with predictable workloads, regulatory pressure, or strategic fear of lock-in.
Positioning choice, then. I wouldn't start as full I-A-A-S. That's a capital bonfire. Start as sovereign P-A-A-S on European bare metal.
Agree. Buy the hardware layer from existing providers. Build the deploy experience, portability layer, security defaults, and operations automation. Concrete isn't the moat.
Risk round. What has to be true? Customers must care enough about sovereignty to switch, the stack must be boringly reliable, support must scale, and the open-source pieces mustn't behave like five products in a trench coat.
Also the story has to be sharper than Europe good, America bad. It's freedom to move, simpler operations, and compliance confidence.
AI against us is serious. A well-funded competitor could use AI to auto-configure the same open-source stack, auto-migrate apps, auto-answer support, and commoditize the wrapper fast.
Or a hyperscaler launches sovereign regions with European operators and a giant sales team. Even if the legal purity is debatable, many buyers will accept close enough.
Then our defensibility is community, trust, portability, and speed. Be the opinionated standard before the big players finish naming their task force.
My verdict is build, but narrow. First next step, ship a working prototype on Hetzner with storage, compute, database, backups, deploy from Git, and a migration demo between two European providers.
My verdict is wait to invest until ten design partners pay. I want three regulated customers, churn risk understood, support hours tracked, and proof that professional accounts can gross margin above sixty percent.
My verdict is invest a small seed, not a moon mission. First next step, find the customer who hates cloud lock-in the most and make their before-and-after story impossible to ignore.
And please don't raise money to build a data center. That's how pitch decks become expensive furniture.
Build the ladder. Rent first, colocate later, own the server room only when the business has earned that migraine.
That's Try Diligence. Three primitives, one sovereignty bet, and one very important reminder that cheap servers don't come with free sleep.
The episode lands on a narrow bet: sovereign PaaS can work for predictable, regulated teams, but not as a full-cloud replacement for everyone.