Sizing Guide
Sizing for AI Loads
Translate GPU/server wattage into battery capacity and daily solar generation requirements.
## Powering the Sovereignty Stack
Running Local LLMs (Large Language Models) or training models requires consistent, high-wattage power. Here is how to size your setup based on real-world consumption.
### The Power Consumption Hierarchy
| Device Class | Typical Load (W) | Peak Load (W) |
| :--- | :--- | :--- |
| **Mac Studio M2 Ultra** | 30W (Idle) | 120W (Full Inference) |
| **Single RTX 4090 Rig** | 100W (Idle) | 600W (Training) |
| **Dual A100 Server** | 400W (Idle) | 1200W (Peak) |
| **H100 8-GPU Cluster** | 800W (Idle) | 5000W+ (Burst) |
### Calculating Your Autonomy
Autonomy is the number of hours your lab can run without sun. For high-availability AI nodes, we recommend a 2:1 solar-to-battery ratio.
To run a **500W load** (Standard GPU Rig) for **12 hours** of total battery runtime:
- **Usable Energy Required:** 6,000 Watt-hours (6kWh).
- **Physical Capacity (LiFePO4):** ~6.5kWh (accounting for 95% DoD).
- **Recommended Configuration:** 48V 135Ah or 12.8V 500Ah bank.
### Solar Recharge Velocity
To fully recharge that 6.5kWh bank within 5 hours of peak sun while also powering the 500W load:
- **Total Panel Wattage:** (6500Wh / 5h) + 500W = **1800W of Solar Panels**.
Need specific hardware numbers?
Use the Sovereign Lab Sizing Calculator to get exact battery and solar requirements for your setup.