ChainCatcher reported that X user gpugene, writing about compute trading, said some B300 transactions have priced above $24 per GPU hour. For contracts with terms shorter than one year, prices have stayed above $7 per GPU hour and are still rising.
gpugene said compute derivatives may help hedge against higher GPU-hour prices while still leaving room to adjust rental duration. He added that when an inference cloud sells services at a fixed price while its own GPU bill is variable, it is effectively carrying compute price risk.
A five-year B300 leasing example
The post gave an example in which a supplier offers 2,048 B300 GPUs for a five-year lease starting immediately at $4.15 per GPU hour. The total value of that arrangement would be $372,264,960.
If the customer goes live after 12 months and uses that scale from months 13 through 15, total usage would amount to 4,485,120 GPU hours.
Option pricing comparison
According to gpugene's calculations, a call option with a $4.50 strike currently carries a premium of about $4.83 million. Under a compound option structure, the upfront payment would be about $2.70 million, followed by another roughly $4.49 million when exercising in month 12.
Results across 4,000 hypothetical paths
gpugene also compared 4,000 hypothetical five-year rent and demand paths. Under annual renewals, the average cost came to $4.72 per hour, while the average cost in the worst 5% of outcomes reached $10.91 per hour.
After adding a call option, those figures changed to $4.88 per hour and $7.20 per hour, with the premium already included.
gpugene said the emerging market for compute derivatives could change how neocloud companies and related participants expand and protect themselves.