GPU Infrastructure

Blue Note Logic's GPU infrastructure is the foundation everything else runs on. Every CaveauAI query, every KaaP Exchange lookup, every API call — all of it executes on bare-metal servers we specified, procured, and deployed in EU data centres.

The Infrastructure Thesis

Cloud GPU rental is a trap. You pay per hour, you share hardware with unknown tenants, your data traverses infrastructure you don't control, and the moment you stop paying, your models stop running. Blue Note Logic took the opposite approach: own the hardware, choose the data centres, control the encryption, and build the networking layer from scratch.

Hardware Stack

• GPUs: NVIDIA RTX PRO 6000 Blackwell — the latest professional-grade AI accelerators
• Locations: Hetzner Helsinki (Finland) and Hetzner Nuremberg (Germany)
• Networking: WireGuard mesh connecting all nodes with encrypted tunnels
• Storage: NVMe arrays with redundant backups across sites

Why Bare-Metal Matters

Bare-metal means no hypervisor overhead, no noisy neighbours, no cloud vendor lock-in. When you run 72B parameter inference, every CUDA core is dedicated to your workload. There is no contention, no throttling, no surprise performance degradation at 2 AM because another tenant started a training run.