How We Saved $82,920/Year by Leaving the Cloud

The $88,440 Problem

Eulerpool is a financial data platform. Think Bloomberg Terminal, but accessible. We serve millions of API requests, run dozens of background workers, and store terabytes of financial data.

Our infrastructure was "modern" — Fly.io for compute, Vercel for the frontend, managed Postgres, managed Redis. All the things you're supposed to use in 2024.

The monthly bill: $7,370.

That's $88,440 per year. For what amounts to about 32 vCPUs and 64 GB of RAM.

The Breaking Point

We weren't growing into this cost — we were already there. And every new feature, every new worker process, every bump in traffic made it worse. The cloud pricing model is designed to scale your costs faster than your revenue.

The breaking point was a $1,200 egress charge in a single month. For bandwidth. For our own data leaving our own servers.

The Migration

We rented dedicated servers from Hetzner. The specs:

• 3x AMD EPYC servers (32 cores, 128 GB RAM each)
• NVMe storage (2 TB per server)
• 20 TB included bandwidth per server
• Located in Falkenstein, Germany

Total cost: $460/month.

Same workload. Same uptime. Better latency (single-digit ms between servers). Zero egress fees.

The Savings

Why We Built RAW

The migration worked. But the process was painful. Setting up servers manually, configuring firewalls, managing SSH keys, dealing with cloud-init — it's all stuff that should be automated.

So we built RAW. One command to deploy a bare metal server:

We open-sourced the CLI, built an API, and added a free tier so anyone can try it. Because nobody should pay $7,370/month for compute that costs $460.

Try It