A calculation estimates inference gross margins at 87.76% to 96.94%

A user estimates inference gross margins of 87.76%–93.88% at off-peak prices and 93.88%–96.94% at peak prices. The calculation assumes 97% cached input, 2% uncached input and 1% output tokens, with GPU rental costs of $3–$6 per hour. It estimates GPU system costs of $0.73–$1.46 per billion tokens, against $11.91 in off-peak revenue or $23.82 at peak prices.