News
Arthur Hayes Unveils AI Compute Futures Plan Ahead of Q1 2027 Flop Genesis
Arthur Hayes proposes AI compute futures and perpetual swaps benchmarked to the FRCR index on his unlaunched Flop Network, targeting genesis in Q1 2027.
AI SummaryAI
- Arthur Hayes proposed quarterly futures and perpetual swaps on AI compute in an essay on October 26.
- The Flop Reference Compute Rate (FRCR) index would be built hourly from Flop Network inference jobs.
- Quarterly futures would take dollar margin through a quanto structure despite the FLOP-denominated index.
- Perpetual swap funding would follow an eight-hour premium index; sellers forfeit FLOP bonds after 24 hours.
A Benchmark From an Unlaunched Chain
Arthur Hayes has laid out a plan for a derivatives market built on artificial intelligence compute, and the design leans on infrastructure that does not exist yet. The BitMEX co-founder set out the proposal in an essay titled “The Trifecta,” published on X on October 26, framing quarterly futures and perpetual swaps, the staples of contract trading, as the first real hedging tools for AI compute costs. His starting point is that no global price for compute exists today: pricing is fragmented between chip rental rates and per-model token rates, so risk managers have no single reference to trade against. The fix he proposes is the Flop Reference Compute Rate (FRCR), an hourly index assembled from inference jobs on the Flop Network, a blockchain that developer Flop Labs has not launched. On that network, AI agents pay miners running blockchain nodes in FLOP, the project's token, to execute model queries, and the completed jobs feed the index. The quarterly futures would sit on dollar collateral, a margin trading structure, even though FLOP denominates the index. That quanto design pays traders in one currency on an asset valued in another. A perpetual swap, meaning a futures contract with no expiry date, would follow, with funding tied to an eight-hour premium index. Market makers could replicate positions by rolling strips of 30-day forward compute units. Sellers would post FLOP bonds sized above their fees, and a missed delivery within 24 hours forfeits the bond.
Update (28 Oct 01:33 UTC): Further reporting on the essay adds new detail on the Flop Network's forward market and timeline.
Further detail from Hayes' essay clarifies the benchmark's mechanics. The FRCR would be computed hourly as a volume-weighted FLOP price per GigaFLOP of computing capacity, and the quarterly futures would settle on the last Friday of March, June, September and December. Because the compute market is priced in FLOP while contracts take US dollars for margin and profits, the quanto structure requires an adjustment for the relationship between the FRCR and the FLOP-to-dollar exchange rate. The perpetual swap would calculate funding as an eight-hour time-weighted average of the gap between its price and the FRCR, with traders on the expensive side paying those on the cheaper side.
Further reporting on the essay adds new detail on the Flop Network's forward market and timeline. Hayes argues that existing chip and per-model token pricing tracks only partial data, and that the market should price compute itself, derived from inference processing time, operation volume and fees recorded on the network. The network also includes a unit called FCU, which lets users reserve inference compute up to 30 days ahead and trade those reservations, with futures positions replicable through combinations of FCUs. The project's technical specification and verification mechanics remain at draft stage, with no dates finalized: Flop Labs targets a testnet and airdrop in the fourth quarter of 2026, followed by mainnet launch in the first quarter of 2027.
@CryptoHayes · X post
X on October 26.
View on X
The USDe Precedent
Hayes pitches the whole package as his bid for a third structural hit, after the perpetual swap he introduced at BitMEX in 2016 and the synthetic dollar design behind Ethena's USDe. USDe is a stablecoin whose backing is hedged with short perpetual swap positions, placing it in the family of algorithmic stablecoins rather than fiat-reserve tokens. Its record is mixed: supply peaked near $14.8 billion in October 2025 and has since fallen below $5 billion, on-chain data from DefiLlama shows. A design that attracted nearly $15 billion gave up roughly two thirds of its float within a year, evidence that appetite for synthetic dollar structures can contract as fast as it grows. The delivery side of the compute plan faces its own timing problem. Flop Labs' page listed no launch date for the FLOP testnet as of mid-October, and the team targets genesis, the network's first block, for the first quarter of 2027. Every layer of the blueprint therefore sits upstream of the thing it depends on: the index needs a live spot market of agents paying miners, the perpetual swap needs the index, and any hedged position needs both. Hayes has also predicted a government bailout after the AI data center boom, and he argues a compute hedging market could help price that risk. That argument holds only if traders accept a benchmark tied to one token network, the very feature compute buyers and market makers say they would hedge against. The essay does not explain how liquidity forms before genesis, since FLOP has no trading market yet.
The essay arrives against a defensive backdrop: Hayes disclosed on Saturday that he “dumped it all,” closing the directional altcoin book at Maelstrom on “dicey” macroeconomic and geopolitical conditions. On-chain records later flagged by the tracking account Lookonchain put roughly $22 million of tokens sold by Hayes-linked wallets, including $15 million in ETH, $2.7 million in ENA and $2.2 million in UNI, with smaller trims in ETHFI, AERO, PENDLE and ZAMA. In the same Monday Substack essay, Hayes argued that if Flop becomes the payments network for AI agents, the FLOP/compute exchange rate would become the global benchmark for spot compute. He also confirmed he serves as Flop Labs' CEO and reiterated a planned “massive airdrop” in the fourth quarter of 2026, ahead of the targeted Q1 2027 genesis.
The commercial backdrop to the plan is quantified in a July report from Boston Consulting Group, which projects the AI compute market growing from $360 billion in 2025 to roughly $2.3 trillion by 2030, more than a fivefold expansion over five years. BCG's framing anticipates compute moving from hourly cloud purchases toward a market where it can be priced, hedged and resold, the exact seat Hayes is trying to claim. The design still depends on FLOP becoming a genuine payments channel for AI agents, and the report's figures say nothing about whether Flop Labs captures that flow.
Adoption Before Genesis
Read as one package, the essay is a bet that compute becomes a traded commodity before the chain behind it exists. The quanto margin design handles the currency mismatch sensibly, but it does not resolve the deeper issue: an index built from one token's jobs is only as trustworthy as that token's network, and FLOP cannot trade until genesis arrives in Q1 2027. The USDe arc, from $14.8 billion down to under $5 billion, shows users will not stay loyal to a synthetic structure once its mechanics disappoint. Genuine hedging demand exists, and Hayes is right that the sector lacks a reference price for compute. The open question is whether the market adopts a benchmark authored by the issuer of the very token it references.
Primary sources
AI-generated, AI-reviewed, under COINOTAG editorial oversight.

