Serious first-principles model
Aerial's Economics of Hybrid Orbital-Terrestrial Data Farming
A parameterized model for a system where GPU drones run autonomy and inference, fishpond hubs supply liquid-cooled ground compute, and sun-synchronous LEO nodes exploit radiative heat rejection and high sunlight fraction.
Thesis in one line
We win if GPUs stay utilized, cooling stays stable, the link budget is honest, and orbital mass is justified by compute value produced each month.
What this model is (and is not)
This is a sensitivity tool to test whether the hybrid farm-to-orbit story is coherent under realistic constraints: power, cooling, comms, mass to orbit, and utilization. Trust the relationships, not the absolute numbers. Site-specific engineering still rules.
Why hybrid at all
Bandwidth is limited, eclipses happen, radiators cost mass, downlinks are weather sensitive, and autonomy on farms still needs low latency. The hybrid split keeps drones on autonomy, hubs on bulk inference and caching, and LEO on compute windows where radiative rejection and sunlight fraction pay off.
Core KPI definition
The page tracks one master KPI: effective cost per useful MW-month of compute. Useful means utilization-adjusted, not peak. Idle GPUs or data shuffling do not count as value.
How to read this
Adjust sliders, watch the tornado chart for dominant drivers, sanity check link budget headroom, and read the assumptions like an auditor. If launch cost or power price dominates, focus there before storytelling.
Inputs
System parameters
Model assumptions
Outputs
Economics and physics
Sensitivity tornado
Each bar shows how the master KPI moves when a single parameter shifts plus or minus ten percent.
LEO ⇄ drone ⇄ pond link budget
Simple RF budget: FSPL plus gains minus losses. Throughput uses Shannon capacity as a ceiling.
Cost model breakdown
Interpretation
The hybrid claim only holds if four constraints stay true together: utilization stays high, cooling stays controlled, the link has margin, and mass to orbit is justified by the monthly compute value. Use the tornado to see which driver dominates right now.
If launch cost or radiator mass dominates, reduce mass per useful compute. If power cost dominates, prioritize solar plus efficiency. If link margin is tight, lean on compression, caching, and edge aggregation to avoid over-promising throughput.