TLDR
- Arthur Hayes proposed giving roughly 20% of the FLOP token supply to testnet participants over 10 years.
- Flop Network would price AI computing work using floating-point operations instead of model-specific tokens.
- Miners would earn block rewards plus fees for processing AI inference requests, a system called Proof of Useful Inference.
- Hayes said he self-funded the development team, so there was no presale for early investors.
- Key details remain unpublished, including total token supply, blockchain choice, and how validators would check miner output.
Arthur Hayes has laid out new details about Flop Network, a computing project designed to serve artificial intelligence agents. The BitMEX co-founder shared the plan in an Aug. 19 Substack post.
He said testnet participants could receive about 20% of the total FLOP token supply. That allocation would be spread out over a 10-year period.
Hayes did not say how many tokens would exist in total. He also did not explain how fast the testnet rewards would be released.
The network would connect AI agents that need computing power with miners who run internet-connected hardware. FLOP would be the token used for payment and rewards on the network.
How Flop Network Would Price Computing Work
Flop Network plans to charge for AI work based on floating-point operations, known as FLOPs. This differs from how most AI companies bill customers today.
Hayes said current AI providers each define their own input and output tokens. That makes it hard for customers to compare prices across different services.
He argued that a FLOP-based pricing system would create one common measure. Buyers could compare compute costs no matter which model, hardware, or location handled their request.
Anyone with an internet-connected computer could become a compute provider under this design. Users would submit jobs with details on the work needed, the time available, and the model to use.
Mining Rewards and Unanswered Questions
Flop Network would use a system Hayes called Proof of Useful Inference. Miners would earn two types of income under this model.
They would receive FLOP block rewards simply for supporting the network. They would also collect fees for completing specific inference requests from users.
The post did not explain how the network would verify a miner’s work. It is unclear how validators would confirm the correct model was used or check for false results.
No white paper, security audit, or token contract had been released. The blockchain that will support FLOP has also not been named.
Hayes said he personally funded the development team, so there was no presale. He said this avoids leaving retail buyers with too many tokens once trading begins.
The plan calls for a FLOP airdrop in the fourth quarter of 2026. The Flop Network genesis block is scheduled for the first quarter of 2027.
Flop Network would compete in a market where stablecoins already handle agent payments. A May 2026 Keyrock report found AI agents settled $73 million in stablecoin transactions over 12 months, with USDC making up 98.6% of that volume.
Coinbase began letting business customers accept agent payments in USDC through its x402 standard in July. Hayes said FLOP would work differently by tying its value to actual computing supply rather than the dollar.
Hayes said his next article will explain why he believes the AI agent economy needs a spot market priced by floating-point operations.
The post Arthur Hayes Proposes FLOP Token Plan for AI Computing Network appeared first on Blockonomi.
