Kalshi has started publishing a forward curve for computing power, letting traders see where the market expects GPU rental prices to sit weeks and months from now. Bloomberg reported the tool on July 14, citing executives at the prediction markets exchange. The move puts a dollar sign on a resource that, until recently, only cloud providers and their largest customers could price with any confidence.
The curve is built from Kalshi’s own weekly and monthly event contracts tied to compute costs, strung together into a single line that extends out about a year. Compute here is shorthand for the power, storage, memory and processing capacity that AI workloads consume, not just the chips themselves. The shape of that line is meant to signal whether renting a GPU will get cheaper or more expensive over the coming months.
That is the same logic bond traders use to read a Treasury yield curve for clues about where the Federal Reserve is headed, or that oil traders use to gauge future crude supply. Kalshi is applying it to a resource that has no equivalent regulated futures market today, despite becoming one of the most consequential inputs in the economy. Compute pricing currently lives in opaque, bilateral cloud contracts and spot-market GPU brokers, with no shared reference point.
A forward curve for compute would matter most to three groups. AI labs planning multi-year training runs could use it to decide whether to lock in capacity now or wait. Cloud providers renting out GPU clusters could hedge against a price collapse if a new chip generation floods supply. Investors evaluating AI infrastructure spending could get an independent read on whether the capacity buildout is outrunning demand, rather than relying on vendor guidance alone.
The obvious limitation is liquidity. A forward curve is only as reliable as the volume of trading behind it, and Kalshi has not disclosed how much money is actually flowing through its compute contracts. A curve built on thin trading can move sharply on a handful of trades, which would make it a noisy signal rather than a dependable one. Bloomberg’s report does not include volume figures or a breakdown of which contracts are most active, so the curve’s predictive value is untested in public.
Kalshi is not alone in trying to build markets around AI infrastructure costs. Other exchange and index operators are pursuing similar products, according to Bloomberg, though the report does not name them or describe how far along they are. Whichever platform draws real trading volume first will likely become the reference price the rest of the industry cites, the way Brent crude became the benchmark for oil.
For any team budgeting GPU spend into 2027, Kalshi’s curve is worth watching even before it is liquid enough to trust: it is the first public attempt to turn a private cost into a market price.
Bloomberg, reported by Katherine Doherty on July 14, 2026, first described Kalshi’s compute forward curve.
