BitMex co-founder Arthur Hayes says AI compute has no global price benchmark, proposing his Flop Network and its FLOP token to become the standard that derivatives markets build on top of.
Why Hayes Says Compute Needs Its Own Price Benchmark
Hayes wrote in a Substack essay Tuesday that current indices tracking chip prices and AI token costs only capture imperfect, indirect measures of compute value.
Chip prices and AI model token prices are both just derivatives of compute itself, he argued, so the real fix is creating a direct, tradable price for compute, measured in floating point operations per second, or FLOPS.
Hayes pointed to his own track record as justification for betting on this idea. He co-invented the perpetual swap at BitMEX in 2016, which became crypto’s most-traded derivative product and now powers platforms like Hyperliquid as measured by Hyperliquid Strategies (NASDAQ:PURR).
He also outlined the concept behind Ethena’s (CRYPTO: ENA) synthetic dollar in a 2023 essay, which grew into the stablecoin USDe, reaching more than $15 billion in circulation within 18 months.
How the Flop Network Would Actually Work
Miners on the Flop Network earn FLOP tokens by completing AI inference tasks for agents. Each transaction tracks three things: how long it took, how much compute work got done, and how much the miner got paid.
Dividing those numbers together produces a live price for compute itself, one Hayes says becomes tradable by anyone holding a chip or FLOP tokens, since the network stays fully open-source.
Buyers can also lock in compute up to 30 days ahead through standardized units Hayes calls Flop Compute Units, which work similarly to a forward contract.
Suppliers must post a FLOP bond well above what they’d earn from the deal, and if they fail to deliver within 24 hours of expiry, they lose that bond to a backup miner who finishes the job.
Why Hayes Wants a Futures Market Built on Top
Hayes laid out a cash-settled futures contract called the FRCR, short for Flop Reference Compute Rate, designed to trade on both crypto and traditional exchanges.
The contract uses a quanto structure:
- Margin posted: in dollars
- Profits collected: in dollars
- Underlying price: trades in FLOP
Hayes said that setup matches what most market participants actually want, letting them trade compute exposure without ever touching the token itself.
He also showed how market makers could replicate long or short exposure by continuously rolling 30-day compute reservations, building synthetic futures and perpetual swaps entirely from the network’s spot and forward markets.
Since each AI model runs its own separate compute market, Hayes said the composite rate will publish per-model weightings, letting traders track exactly how the benchmark shifts as model usage changes.
Image: Shutterstock
Recent Comments