Author: Arthur Hayes
Translated by: TechFlow
Original link: https://www.techflowpost.com/article/34885
Disclaimer: This article is republished content. Readers can find more information via the original link. If the author has any objections to the way it has been republished, please contact us and we will make changes as requested. This repost is for informational purposes only, does not constitute investment advice, and does not represent the views or positions of Wu Says.
Throughout the evolution of crypto finance, every shift in asset paradigms has been accompanied by the creation of entirely new foundational derivatives. In 2016, Arthur Hayes and his BitMEX team invented the perpetual swap, reshaping the global crypto derivatives trading landscape. In 2023, in (Dust on Crust), he proposed the idea of a synthetic dollar created by hedging inverse contracts with spot longs, directly giving rise to Ethena (USDe), a synthetic dollar protocol whose current scale exceeds $10 billion. Now, Arthur Hayes is back with his third grand vision: Flop Network.
In response to the geometric explosion in compute demand in the age of AI—and the pain points of the traditional cloud compute market, which is fragmented, opaque, and lacks a unified spot benchmark—Hayes proposes anchoring compute to physical floating-point processing power (FLOPS). A decentralized inference verification network would establish a global standardized spot exchange rate for compute, with standardized deliverable compute forward contracts (FCUs) and cash-settled Quanto futures and perpetual contracts built on this underlying asset.
This article rigorously develops the complete mathematical framework for the financialization and tokenization of compute assets, along with the logic behind market-maker hedging. It presents the ultimate blueprint for how crypto-native finance can reshape macro liquidity and control over compute pricing in the age of AI agents.
I want to achieve a “three-peat” (Three for Three). I’m not talking about basketball, but about my ability to envision new financial derivatives or use cases and watch them evolve into industry standards.
My track record so far:
Perpetual Swaps: A huge success (in Borat’s voice)
In May 2016, my team at BitMEX and I invented the perpetual swap, commonly known as “perps.” Perpetual swaps quickly became the most traded financial derivative in the history of crypto. Thanks to their enormous liquidity and popularity, innovative platforms like Hyperliquid now use this mechanism to power 24/7 trading in RWA (real-world asset) derivatives.[1]
Synthetic US Dollar Stablecoin: A huge success (in Borat’s voice)
In March 2023, I wrote an in-depth blog post titled “Dust on Crust,” explaining how to create a synthetic US dollar through a basis hedge consisting of a long spot Bitcoin position and a short perpetual contract. This synthetic dollar can be packaged as a stablecoin and used widely for payments and settlement throughout the crypto capital markets. The Ethena team adopted my idea and executed it brilliantly. Ethena’s synthetic dollar, USDe, became the fastest-growing stablecoin in history: in less than 18 months, its circulating supply soared from $0 to over $15 billion.
Flop Network Spot Compute Market: The Verdict Is Still Out (TBD)
This summer, I first shared my vision for Flop Network. $FLOP is the “food” for your AI agents. Agents will use $FLOP for commercial interactions because it can be directly exchanged for physical compute, measured in floating-point operations per second (FLOPS). If Flop becomes the native payment and settlement network for AI agents, it follows naturally that the exchange rate between $FLOP and compute will become the global benchmark for spot compute prices.
A truly globally liquid spot compute market will inevitably give rise to a derivatives market built on that spot benchmark. May this article serve as the definitive blueprint for ambitious entrepreneurs seeking to build a compute derivatives market fit for the new age of AI agents.
The Core Problem
There is currently no unified benchmark for compute pricing anywhere in the world. Although various indices attempt to track the sales and rental prices of specific CPU, GPU, and TPU chip models, their data samples are severely incomplete. Other indices attempt to track the price of AI tokens for each model. But chip prices and model token prices are both secondary derivatives of compute. If I can accurately determine the benchmark price of compute (FLOPS), I can derive the intrinsic value of any chip and the fair cost of any model token from the top down. We should not waste our efforts collecting fragmented and imperfect chip and token price data. Instead, we must focus on establishing a unified price discovery mechanism for physical, underlying compute.
The Solution: Flop Network’s Compute Market
Flop Network produces one block per second. Each block contains verifiable proof-of-receipt confirming that a miner successfully completed an inference task for an agent. For every inference task, we can obtain three key parameters to precisely calculate the exchange rate between $FLOP and physical compute.
Block data objects include:
Latency (in seconds): The actual time taken to complete the inference request
Work completed: The total number of floating-point operations actually performed (FLOPS)
Transaction fee: The amount of $FLOP earned by the miner for completing the task
$FLOP / (Compute / second) = [3] / [2] / [1]
A spot price is a true and valid market price only when everyone is free to trade. On the supply side, if you own compute chips—whether GPUs, CPUs, or TPUs—you can mine $FLOP by completing inference requests for agents. On the demand side, if you hold $FLOP in your wallet, you can submit inference tasks at any time. Flop Network is an open-source network accessible to all. Therefore, the spot compute price discovered here will become the globally recognized benchmark for compute.
Flop Compute Unit (FCU)
Before discussing cash-settled financial derivatives, let’s continue exploring the path of the “physically delivered” compute market. One of Flop Network’s key features is that it allows buyers to reserve compute from miners for delivery up to 30 days in the future. To improve liquidity and tradability, the network standardizes compute into “Flop Compute Units” (FCUs).
Imagine you know that you have a major compute job to run in 7 days and want to lock in your compute costs today. You can place a bid to reserve a specified amount of compute from a miner for delivery in 7 days. Once the 7 days have elapsed, the miner has a 24-hour window to deliver the actual compute. When a miner accepts your bid, the reservation creates a standardized 7-day FCU object, which is essentially a physically deliverable forward contract. If the spot price of compute rises sharply before delivery and you no longer need the compute, you can simply resell the FCU directly on Flop Network to another participant who needs it.
Here is a practical example of how FCUs work. Please stay tuned for updates to the official Flop documentation, where all the technical details will be provided.
Specifications for a single FCU underlying:
Latency: 1 second
Compute capacity: 1 PetaFLOP (one quadrillion floating-point operations)
AI model: Specified by the buyer
Delivery time: 12:00 UTC, 7 days after creation
Please note: FCUs cover inference scenarios only and are not suitable for model training. The overall value of a compute cluster for model training is far greater than the simple sum of its individual components. Inference tasks, by contrast, are inherently highly divisible, which is the fundamental premise that makes a single benchmark unit like the FCU viable.
For each AI model, the market can have up to 30 FCU expiries available for trading at any given time, with one expiry date per day. If you need to reserve 1 ExaFLOP of compute (one quintillion operations), you only need to buy 1,000 FCUs. As a buyer, to attract miners to fill your order, you must post the $FLOP price you are willing to pay for the FCU and pre-fund the full amount for network verification. As a seller (compute provider), you post the $FLOP price you wish to receive, and must also stake an equity-weighted amount of $FLOP as collateral to ensure that, if you default, other miners can take over and complete delivery profitably in the spot market. The network dynamically determines the stake amount based on the total inference volume processed across the network recently. It includes the baseline $FLOP cost as well as a penalty multiplier to deter malicious fraud and delivery defaults. This mechanism creates a fully funded, two-sided order book that completely eliminates junk orders: funded buy and sell quotes are finally settled on-chain, while the high-frequency, continuously updated two-sided order book and matching engine run off-chain, just like the spot market.
Forward markets inherently carry a risk of failure to deliver. As noted above, to mitigate this risk, FCU compute providers must post a substantial amount of $FLOP collateral—far more than the inference fees they receive. If a miner fails to deliver the agreed compute within 24 hours of expiry, their entire $FLOP stake is forfeited. Backup miners then bid to take over and complete the order. They are willing to do so because the fee per PetaFLOP they earn is far higher than the prevailing spot price at the time. If the spot price surges above the collateral amount and no miners are willing to fulfill the order, the system deems a default to have occurred. The buyer then receives a full refund, plus compensation equal to all the forfeited collateral.
FRCR Futures Contract
FRCR: Flop Reference Compute Rate
Hourly volume-weighted settlement rate for $FLOP / GigaFLOPS
I’ve explained how real-world buyers of compute use $FLOP to procure physical compute in the spot and forward markets. Now let’s turn to how to build cash-settled financial derivatives that can be listed on traditional finance (TradFi) and cryptocurrency exchanges. I’ll lay out a complete contract design and walk through how market makers can hedge long or short futures exposure.
Contract specifications:
Contract name: FRCR Quanto Quarterly Futures Contract; expiry: 12:00 UTC on the last Friday of March, June, September, and December; underlying index: FRCR.
Contract multiplier: $1 per point
Example: Assume the FRCR index is 100 points
Contract value: Multiplier × FRCR = $1 × 100 = $100
Margin currency: US dollars (USD)
If you go long 1 contract at an entry price of 100 $FLOP and settle at 200 $FLOP, your profit and loss (PNL) is:
Multiplier × Number of contracts × (Settlement price - Entry price) = $1 × 1 × (200 - 100) = $100
If you go short 1 contract at an entry price of 200 $FLOP and settle at 100 $FLOP, your PNL is:
Multiplier × Number of contracts × (Settlement price - Entry price) = $1 × (-1) × (100 - 200) = $100
A Quanto derivative is an exotic derivative. A Quanto structure is necessary because I assume that most market participants want to post margin in US dollars and withdraw profits in US dollars, while the underlying spot market is quoted in $FLOP. We therefore need a Quanto structure to deliver the payoff profile traders want. To price Quanto derivatives accurately, market makers must dynamically adjust the risk premium based on the covariance between the FRCR index and the $FLOP / USD exchange rate.
If the market prefers to keep $FLOP as the margin currency, the contract design can be simplified as follows:
Contract multiplier = 1 GigaFLOP
Contract value = Multiplier × FRCR = 1 × 100 $FLOP = 100 $FLOP
The success of any financial derivatives contract hinges on how easily market makers can use the underlying assets to replicate long and short futures positions.
Replicating a long futures position:
To replicate long exposure, a market maker needs to acquire $FLOP and buy a series of FCUs with consecutive expiries. Here are the specific steps for synthesizing a 60-day FRCR long futures position:
Spend $FLOP to buy a 30-day FCU; when it expires after 30 days, roll the position into another 30-day FCU.
Assume that on November 1, a 30-day FCU is priced at 100 $FLOP. On November 30, you sell and settle the original 30-day FCU for 100 $FLOP, while buying a new 30-day FCU for 200 $FLOP that expires on December 30.
At this point, you have successfully synthesized a 60-day long futures contract with an entry price locked in at 200 $FLOP.
Replicating a short futures position:
To replicate short exposure, a market maker needs to acquire $FLOP and offer a series of FCU compute supply contracts with consecutive expiries. This requires more $FLOP capital because the collateral stake is far greater than the compute fees received. Market makers do not need to actually procure chips, since they can roll over and close out their positions before expiry. Here is the process for synthesizing a 60-day FRCR short futures position:
Spend $FLOP on collateral and place an offer to supply a 30-day FCU; before it expires, roll the position into another 30-day FCU supply contract.
Assume that on November 1, a 30-day FCU is priced at 100 $FLOP. On November 30, you buy back the original 30-day FCU for 100 $FLOP to close the position, and place an offer at 200 $FLOP to supply a new 30-day FCU expiring on December 30. As long as you roll over the position before the FCU expires, or designate a backup miner to fulfill the order within the 24-hour delivery window, you will never default and your collateral will remain safe.
This perfectly replicates a 60-day short futures contract, with an entry price locked in at 200 $FLOP.
FRCR Perpetual Contract
Once you thoroughly understand how futures contracts work, extending them into perpetual contracts is a natural next step. Remember, a perpetual contract is essentially an indefinitely rolling chain of continuous futures positions.
Perpetual contract specifications:
Contract name: FRCR Quanto Perpetual Contract; underlying premium index: the 8-hour time-weighted average price (TWAP) of [Perpetual Contract Price - FRCR Index].
The perpetual contract price will use an hourly volume-weighted average price to strictly match the FRCR calculation method, with eight samples collected over an 8-hour period and averaged arithmetically.
Contract multiplier: $1 per point
Example: Assume the FRCR index is 100 points
Contract value: Multiplier × FRCR = $1 × 100 = $100
Margin currency: US dollars (USD)
The funding rate mechanism is identical to that of traditional perpetual contracts. If the premium index is positive, the funding rate is positive and longs pay funding to shorts. If the premium index is negative, the funding rate is negative and shorts pay funding to longs. Since this is a Quanto derivative, all funding payments are settled in US dollars.
To synthesize a long perpetual position, traders simply go long FCUs and roll them over continuously. For example, to establish synthetic long perpetual exposure for 120 days, a trader only needs to buy one 30-day FCU and roll it over four times. Conversely, to synthesize a short perpetual position, traders simply place offers to supply FCUs and roll them over continuously. For example, to establish 120 days of synthetic short exposure, a trader only needs to offer one 30-day FCU for supply and roll it over four times.
Choosing the Right AI Models
One subtle challenge in the compute derivatives market is that the benchmark compute price underlying financial derivatives is model-agnostic, while in practice each AI model has its own FCU trading market. Therefore, to synthesize futures and perpetual contracts accurately, traders must buy or supply a basket of FCUs for individual models, weighted to match the FRCR composite index.
Alongside publishing the composite index, FRCR provides real-time breakdowns for each model asset pool, making the weighting of the basket fully transparent. The composite index shifts as the mix of model usage changes. For hedgers, the key risk to manage therefore becomes the risk of model iteration and retirement, rather than the physical depreciation of underlying hardware.
Frictionless Trading
To foster and grow the derivatives market based on spot and FCU compute prices from Flop Network, FCU trades incur no additional platform trading fees. The parties to a trade only need to pay validators the network’s basic $FLOP fee to package the transaction. This fee is no different from the fee for a regular $FLOP transfer between wallets.
Key Advantages of Flop’s Spot and Forward Markets
Anyone can participate in compute procurement or supply on Flop Network without any barriers to entry.
All settlement transactions are executed in a fully transparent, open market.
All data calls generated by Flop Network are completely free.
It’s that simple and clear. If you aspire to build any kind of compute derivative, your exploration must begin with a deep understanding of on-chain activity and price discovery in Flop Network’s spot and forward markets.
Hail to the king!
[1] RWA: Real-world asset (e.g., crude oil or gold).