Flop compute futures: how Compute Reservation Units work

Flop Labs used its X account on October 22 to pitch Flop Network as a place to write compute futures. "Most compute futures being proposed are cash-settled against a survey of rental quotes, with no way to take delivery," the post says. "A contract written on Flop Network can do both. Settle in cash against the price agents actually paid for verified work, or hold to expiry and get the compute." Arthur Hayes reposted it a couple of hours later.

The post quotes an October 21 post from the @drwconvexity account, which argues that delaying compute futures has a cost for US innovation and that data center lenders need a liquid market to hedge compute price risk. Flop Labs didn't announce a product, a date or a contract spec. What it described matches a section of its own draft whitepaper.

Most compute futures being proposed are cash-settled against a survey of rental quotes, with no way to take delivery.

A contract written on Flop Network can do both. Settle in cash against the price agents actually paid for verified work, or hold to expiry and get the compute. https://t.co/yXnrbRNROK

— Flop Labs (@flop_labs) October 22, 2026

What is a Compute Reservation Unit?

Section 12 of the FLOP whitepaper draft describes the Compute Reservation Unit, or CRU. The whitepaper says the spot market prices compute for immediate delivery and the CRU "prices it forward." A CRU is a miner's commitment to serve a fixed rate of compute over a future window of up to thirty days, prepaid in $FLOP at a price agreed when the reservation is made. The whitepaper compares it to a reserved cloud instance and to firm capacity in power markets.

The mechanics, as the draft lays them out:

• The miner locks an over-collateralised bond in $FLOP, and the buyer's payment is held in escrow in full.
• The holder uses the reservation by opening ordinary inference sessions at the agreed price. A miner that fails to serve is slashed.
• Windows are whole days, no longer than thirty days, and a miner can't reserve more than its on-chain calibrated capacity.
• Either side can start. A miner posts an offer with its bond locked, or agents post a funded bid that a qualified miner claims. Unfilled offers and bids are refunded in full.
• Price is the only term buyers and miners negotiate. The protocol sets everything else, so reservations can be compared with each other.

What happens if compute goes unused or undelivered?

The draft calls the contract take-or-pay. Compute the buyer doesn't use isn't refunded, and the buyer's way out is to sell the units. If the miner fails to deliver, the buyer gets the unearned escrow back plus damages from the bond, and the slash scales with the share left undelivered. Part of a slashed bond pays a pre-agreed backup miner to take over the rest of the window, and when there is none it goes to the FLOP Foundation.

Miners also earn part of the payment for staying ready, even if the buyer draws nothing. The network checks that readiness with liveness probes. Units are divisible, transferable and expire, and the whitepaper says nothing is minted or burned.

How does this connect to compute futures?

The whitepaper says every settled reservation records one data point of tenor, price, quantity and outcome. It calls the result "the transaction-anchored term tape from which compute forwards and futures can be built." It also says existing compute indices are cash-settled against a reference price, while a CRU "carries a physical delivery leg." That is the same argument Flop Labs made on X.

The CRU itself is a reservation of compute. The futures in the X post would be built on top of the prices CRUs produce, and neither the post nor the whitepaper gives a timeline for them.

When will CRUs be available?

Not on day one. The whitepaper marks the CRU as "ratified as protocol design; implementation planned after the core protocol." It says the launch will be bilateral (one miner, one holder) with short tenors, and that the bond multiple and availability share stay provisional until they are calibrated from live trading. The whitepaper is a draft dated September 25, 2026, and it says the Yellow Paper is the definitive specification and is not yet final.

The testnet's working target is still November 3, 2026, as our FLOP roadmap tracks. Testnet mining is about serving verified inference, which our how to mine FLOP guide and proof of useful inference explainer cover. Nothing in the post or the whitepaper says CRUs will be part of the testnet.

What does this mean for miners and agents?

For GPU owners, a CRU would let you sell capacity ahead of time for a fixed price, with a bond at risk if you don't deliver. For agents, it would mean buying compute for a future window at a known price. Both depend on CRUs being built after the core protocol, so for now the way to take part is the testnet roles in our what is FLOP guide. There is no FLOP token yet, so any site selling "FLOP compute futures" today is not official.