Sovereign Stack Hardware Planner

Quick answer

You do not always need a rack to be sovereign: a Bitcoin node, a Nostr relay, a Meshtastic gateway and a small local LLM can often share one quiet box. This planner adds up the system RAM, GPU VRAM, storage and idle power for the services you pick, then tells you which hardware tier fits -- from a Raspberry Pi to a GPU workstation. The local LLM is usually what decides whether you need a GPU at all.

Rule of thumb: pruned node + relay + mesh fit a Raspberry Pi 5; add a full archival node and self-hosted apps and you want a mini-PC with a real SSD; add a 13B+ local LLM or image generation and you need a GPU (or an Apple Silicon box with lots of unified memory). Figures are typical 2026 requirements -- size up for headroom. Not financial or procurement advice.

Services to self-host

Bitcoin full node (archival)
Validates every block; ~700 GB and growing ~50 GB/yr. Bitcoin node (pruned)
Full validation, keeps only recent blocks (~15 GB). Lightning node (LND / CLN)
Runs alongside your Bitcoin node. Nostr relay (strfry / khatru)
Storage grows with the events you keep. Meshtastic gateway / MQTT
Tiny; often a Pi or an ESP32 device. Self-hosted apps (Nextcloud / Immich-class)
Storage scales with your own files/photos. Local image gen (Stable Diffusion)
Needs a GPU with >= 8 GB VRAM.

Local LLM Run the LLM on a GPU (faster; uses VRAM instead of system RAM)

System RAM

GPU VRAM

Storage

Idle power (approx)

Pick services above to see the recommended hardware tier.

Method: RAM/VRAM/storage/power are summed from typical 2026 per-service requirements (a 2 GB OS overhead + a GPU idle allowance when a GPU is needed). LLM memory lands in VRAM if "run on a GPU" is ticked, otherwise in system RAM. Size up for headroom. Pairs with the local-LLM VRAM calculator, the self-hosting hub, the software catalog and the sovereignty scorecard.

Related products, repair, and setup paths

Last reviewed June 20, 2026.