5. Comparing CAPEX and OPEX
CAPEX Comparison
Hardware Investment
Approximately $6 billion for a 100k–chip cluster
Minimal nodes are owned by end users
Data Center Infrastructure
Requires extensive investment (cooling, racks, etc.)
Software–based; no centralized data center needed
Overall CAPEX
Extremely high
Negligible (only software and coordination CAPEX, e.g., $2M over 8 years)
OPEX Comparison
Compute Payment (Hourly)
$1.75 per chip × 100,000 chips ≈ $175,000 per hour
$0.005 per node × ~6.7 million nodes ≈ $33,500 per hour
Annual Compute Cost
~$1.53 billion per year
~$293 million per year (paid directly to node operators)
Electricity & Cooling
High estimated ~$61 million per year for electricity alone
Borne by individual node operators (minimal impact on GNUS)
Network & Storage Costs
Additional expenses (~$100 million/year)
Minimal—local connectivity and distributed storage
Total Annual OPEX
Approximately $1.7 billion per year
On the order of $2–$10 million per year (true OPEX for coordination)
Note: For GNUS.ai, the $293 million figure represents the total “compute payment” distributed in tokens, which is a pass–through payment to node operators and not an overhead cost for the GNUS network itself.
Last updated