Somewhere between a futures exchange and a betting parlor, Kalshi has quietly built something that the AI industry has been missing: a way to lock in the future cost of raw compute power.
On July 14, the CFTC-regulated prediction market exchange launched what it calls compute forward curves, a tool that lets traders and businesses see implied future prices for renting Nvidia GPU capacity across a window stretching up to a year ahead.
The cost of renting a high-end GPU is, increasingly, the cost of doing business in AI. Data centers, model trainers, and inference providers all live and die by what it costs to access compute. Until now, there was no standardized market mechanism to hedge that exposure, which meant companies were flying blind on one of their largest variable costs.
What Kalshi actually built
The forward curves are derived directly from trading activity on Kalshi’s event contracts, covering three of Nvidia’s most consequential chips: the B200, H200, and A100.
