Our founder and CTO, Oskar Boëthius Lissheim, just published a piece on how — and why — we moved Multiply's infrastructure off a US cloud partner that had become unreliable and expensive, onto Hetzner: a German company running its own datacenters since before "cloud" was a marketing word. We've adapted it below. Read the original on LinkedIn.
The bill for prod dropped from about $2,100 a month to $580. But the money was the least of it. We moved for reliability, a simpler architecture, and infrastructure we have more control over; the ~3.5× cost cut just came along. The partner stays unnamed. This isn't a takedown — the point was never who we left. It's what we found.
What "unreliable" meant
Our old provider didn't own its datacenters. It ran its own hardware in leased space across third-party facilities around the world. Too often an operator ran maintenance that was badly communicated — to the provider, to us, or both — and we lost service with no insight into why. Support was responsive but often as blind as we were: one more party waiting on an operator they didn't control. Every location had its own networking and hardware, so outages and performance were nearly impossible to predict.
The sharp end was the network between our own services. Dropped connections between apps — catastrophic for a ZooKeeper cluster trying to hold consensus, which our Rama database depends on. Unexplained partitions between the database and the web tier, with no recourse. They'd heal on their own after minutes, hours, sometimes days. In between, we burned time and nerves chasing failures that were never ours to fix.
On Hetzner, when something breaks, it breaks somewhere we can see.
What Hetzner is
It owns its datacenters — real buildings, real metal — at least across Europe, where we run: Falkenstein, Nuremberg, Helsinki. That sounds like a footnote until you've spent years debugging infrastructure you can't see. When the company you pay owns the floor the servers sit on, the parties between an incident and someone who can fix it drop from at least two to one. Reliability stops being a status page you refresh and becomes something you can reason about.
The rest is almost boring:
- Latency we can feel. Production sits in Helsinki, a low-teens-millisecond hop from Stockholm, with German datacenters a bit further south. Close to our users, close to us.
- An admin UI that's simple and pleasant — a rare sentence to write about infrastructure.
- Support that knows the systems. They own the stack end to end, so the person you reach has real insight — not a tier-one script. No paid "support tier" on the invoice, either.
- And the bonus: the bill. Our whole production environment — web tier, Rama cluster, ZooKeeper, ops box, load balancer — runs about €500/month (~$580), roughly 3.5× less than before.
Hetzner reads like a stripped-down AWS: the parts a modern SaaS needs, none of the legacy cruft. No 200-service console you navigate with a map. No bill that needs a FinOps hire to decode.
How we run it
Everything is Terraform, in layers. Modules, environment bundles, dev/staging/prod classes. A new environment is a config file, not a weekend. Networks, subnets, firewalls, load balancers, the servers themselves: all code against a clean Hetzner API. On the old provider, infrastructure-as-code was a constant manual burden; here it's the default. Deploys go out through Kamal — health-checked, one server at a time, no drama. A Tailscale mesh ties the private network together. We still lean on a few managed AWS pieces like S3 — but the compute, the part that has to be stable at 3am, lives on hardware we understand.
We ran dev, staging, and QA on Hetzner for many months before cutting production over recently. It's already the quietest stretch of infrastructure I've had in years. We changed provider and simplified the topology in one move, so I can't cleanly credit one — but the unexplained partitions and spurious timeouts stopped, and things stay up.
The part that reads like geopolitics
Moving compute from an American provider to a European one that owns its metal isn't, in 2026, a purely technical decision. "Where does our infrastructure physically live, and who controls it?" is a board-level question now. For a European company serving European clients, "Helsinki and Germany, on a vendor we can call who answers" is worth more than it was two years ago.
We won't romanticize it. AWS is an extraordinary machine, and there are workloads we'd still put nowhere else. But for a lean, AI-native company, the hyperscaler's gravity — the sprawl, the lock-in, the bill — is a tax you pay for capabilities you mostly don't use. Nobody picks Multiply for where the servers sit. They stay because it's up, fast, and doesn't lose their work — and that's built one boring, well-understood decision at a time.
This is part of an ongoing series on how we build Multiply. Read Oskar's full essay on LinkedIn.