Offshore exchanges beat CME, ICE to AI compute derivatives
Unregulated exchanges are trading crypto-style perpetual futures on GPU capacity, foreshadowing a regulated derivatives market that CME and ICE will not enter until late 2026.
Crypto-style derivatives have reached the AI compute market, with an offshore exchange listing perpetual futures on GPU pricing years before CME Group and Intercontinental Exchange plan to launch regulated equivalents. Architect's AX exchange, registered in Bermuda and outside the jurisdiction of the Commodity Futures Trading Commission, is already offering the contracts. Bernstein analysts Gautam Chhugani and Madison Rezaei noted that CME and ICE are targeting late 2026 for their own cash-settled compute futures, pending CFTC review.
These perpetual contracts borrow their structure directly from digital asset markets. They carry no expiration date and instead track a spot index through a funding-rate mechanism, where longs pay shorts when the contract trades above the underlying index. For AI market participants, this offers a way to manage volatile pricing. Neocloud operators sitting on depreciating GPU fleets can hedge capacity sold on demand, while AI labs can lock in future compute costs.
Prediction markets are providing a parallel pricing mechanism. Kalshi, which operates under CFTC oversight, now lists event contracts on GPU rental prices, such as whether an NVIDIA B200 will clear $7 an hour by the end of 2026. Aggregated contract prices produce a forward curve. Since launching these curves for the B200, H200, and A100 chips on July 14, Kalshi's B200 forward has sat at $5.41, well below a historical peak of $7.39.
Bernstein compares compute more closely to electricity than to oil. An unused GPU hour cannot be stored, meaning forward-curve prices reflect real-time scarcity rather than the cost of carrying a physical commodity. This makes derivatives essential for transferring price risk between cloud operators and enterprise buyers.
However, the market currently lacks the depth required for institutional hedging. Analysts warned that liquidity remains nascent and is largely driven by speculative flows rather than commercial demand. The fundamental constraint is benchmark construction. Because most compute capacity is still traded through private negotiations, building a representative settlement index is difficult.
Two data providers are attempting to solve this. Silicon Data, partnered with CME, collects roughly 150,000 verified pricing records daily across 50 regions and up to 100 platforms. Ornn, which supplies the index for ICE and Kalshi, relies on negotiated transaction prices. Until these benchmarks gain widespread trust, the regulated futures planned for 2026 may struggle to attract the institutional liquidity needed to make AI compute a standard asset class.
