The federally regulated exchange is turning AI compute into a tradable commodity, letting investors bet on the future cost of NVIDIA chips by the hour.
GPU computing power now has its own futures market. Kalshi, the first federally regulated prediction market exchange in the US, has launched contracts that let traders speculate on the per-hour cost of running NVIDIA’s most sought-after chips.
The contracts cover NVIDIA’s H100, H200, B200, and RTX 5090 hardware, resolving based on real-time pricing data from Ornn, an index provider that tracks average compute-per-hour costs across the GPU rental market.
How GPU compute became a commodity #
Ornn defines GPU compute as a commodity class comparable to energy and metals. Their index provides the settlement mechanism for Kalshi’s contracts, giving traders a transparent benchmark rather than relying on anecdotal marketplace pricing.
In early coverage of the index, Ornn reported H100 compute prices averaging around $1.70 per hour.
Kalshi’s GPU compute markets first went live in March 2026, with initial contracts covering RTX 5090 and H200 per-hour rates set to resolve by March 31, 2026. The platform lets traders see visible odds and historical trends, creating a price discovery mechanism for a resource that had previously been negotiated in private deals and cloud provider dashboards.
The B200 tells an interesting story #
NVIDIA’s B200 saw its per-hour compute rate peak at $6.11 on May 30, 2026. By June 21, it had dropped to $4.22. That’s a roughly 31% decline in less than a month.
Traders on the platform have been increasingly betting against sustained high rates for B200 compute.
What this means for investors #
The immediate utility is hedging. AI companies that need to lock in compute costs for multi-month training runs can now use prediction markets to manage their exposure to price swings.
For traders and investors, the contracts offer a way to express a view on AI infrastructure demand without buying NVIDIA stock or taking positions in cloud provider equities. Compute pricing captures real-time supply and demand dynamics at the infrastructure layer, stripped of the multiple expansion and narrative trading that drive equity prices. The declining B200 rates visible in Kalshi’s data carry a warning for anyone positioned heavily in the “GPU scarcity” trade. Conversely, falling compute costs are bullish for AI application companies that consume GPU hours as a primary input.
Disclosure: This article was edited by Editorial Team. For more information on how we create and review content, see our