Perpetual and event contracts trade GPU prices before CME, ICE

Offshore perpetual futures and prediction-market contracts tied to GPU-hour pricing are trading now, months before planned cash-settled futures from CME Group and Intercontinental Exchange that are under Commodity Futures Trading Commission review. Perpetual-style compute futures and event contracts are establishing early price signals as regulators and index providers work on formal benchmarks.

Exchange Architect’s AX, registered in Bermuda, lists perpetual compute futures. Those venues operate outside U.S. derivatives oversight, allowing perpetuals to launch earlier than traditional futures. Analysts at Bernstein attributed the timing to contract structure: perpetuals have no expiry and use a funding-rate mechanism to link contract prices to a spot index.

Under the funding-rate model, when the perpetual trades above the spot index, long holders make periodic payments to short holders; a persistent positive funding rate signals sustained bullishness on the spot price. Bernstein’s analysts described the design as originating in crypto markets. They also noted that perpetuals can be used as hedges by cloud operators that sell compute capacity on demand or by buyers who need exposure without fixing a delivery date.

Prediction-market contracts provide another form of price discovery. Kalshi, a CFTC-regulated exchange, lists event contracts tied to GPU rental rates. One contract pays $1 if an NVIDIA B200 clears $7 an hour by the end of 2026 and pays nothing otherwise. A trading price of $0.40 on that contract implies a 40% probability. Kalshi rolled out forward curves for B200, H200 and A100 chips on July 14; Bernstein’s reading showed the B200 forward at $5.41 versus a $7.39 historical peak.

Market participants include neocloud providers and large cloud operators that run fleets of depreciating GPUs, and AI labs and enterprises that need visibility on future compute costs. Because unused GPU-hours cannot be stored, forward prices reflect expected scarcity and usage rather than a cost of carry; hedging is therefore the primary way to transfer multi-period price risk.

Analysts noted limitations in liquidity and benchmark quality. Early liquidity for perpetual and prediction products is limited and has been driven largely by speculative flows, Bernstein reported. Many compute transactions still occur through private, negotiated deals, complicating the creation of a representative, tradable index for cash settlement.

Data firms and index builders are developing reference prices. Silicon Data, which is working with CME, aggregates roughly 150,000 verified pricing records daily across about 50 regions and between 50 and 100 platforms to normalize an H100 hour into a tradable asset. Ornn, which supplies the settlement index used by Kalshi and is working with ICE, builds its index from negotiated transaction prices.

Bernstein’s note was prepared by analysts including Gautam Chhugani and Madison Rezaei. The firm disclosed that those analysts hold long positions in various cryptocurrencies. Regulators’ review of cash-settled futures and the establishment of transparent spot indexes will affect whether current offshore and prediction-style contracts develop into deeper, institutionally accepted markets for AI compute.