Arthur Hayes Plans to Build Derivatives Around AI Compute

Arthur Hayes published a new essay on October 27 outlining futures and perpetual swaps tied to Flop Network’s proposed compute market, where AI agents would pay for computing power using FLOP.

The plan aims to establish a reference price for compute that traders could use to build financial contracts around the resources AI systems consume.

Hayes Details How FRCR Futures Would Work

The proposed Flop Reference Compute Rate (FRCR) would be calculated hourly using a volume-weighted FLOP price per GigaFLOP of computing capacity. Quarterly futures would settle on the last Friday of March, June, September, and December, with contracts using US dollars for margin and profits.

Hayes explained that the dollar-based structure would require a quanto derivative, a contract whose payout uses a different currency from its underlying asset. Because the compute market would be priced in FLOP while traders could prefer dollar-denominated returns, pricing the contract would require an adjustment for the relationship between the FRCR and the FLOP-to-dollar exchange rate.

The proposed perpetual swap would use an eight-hour time-weighted average of the difference between its trading price and the FRCR to calculate funding payments. As with crypto perpetuals, traders on the expensive side of the contract would pay those on the cheaper side, depending on the premium.

Market makers could recreate futures exposure by buying or supplying Flop Compute Units (FCUs), which represent reserved inference capacity for a specified AI model and delivery time. Hayes’s design also requires suppliers to post FLOP bonds, which they risk losing if they fail to deliver the promised computing work.

Flop’s Spot Market Is the Foundation

The derivatives proposal depends on Flop Network first establishing a functioning spot market. Hayes argues that existing indexes tracking chip rental prices and individual AI token prices provide incomplete information because both are derivatives of the underlying cost of computation.

Under his design, network blocks would record inference tasks, the computing work completed, and the fees paid to miners. Those receipts would help calculate the exchange rate between FLOP and computing capacity, giving buyers and suppliers a common reference.

Hayes introduced Flop Labs in August 2026, describing its proposed native token FLOP as “food for your AI agent.” In an August 27 explanation, the project argued that autonomous agents need money directly exchangeable for computing resources rather than payment systems designed primarily around people and businesses.