Arthur Hayes said on Aug. 18 that he is “ending retirement” to become CEO of Flop Labs, a new crypto project centered on the AI agent economy. The former BitMEX chief, who has spent recent years speaking more as an investor and macro commentator than an operator, is now returning with a plan to launch $FLOP, a token Flop describes as fuel for AI agents.
Earlier that day, Hayes had posted a selfie on X showing himself using a nasal oxygen tube and wrote that he was “physically preparing for the bull market.” A little more than 10 hours later, he unveiled Flop Labs. The announcement quickly spread across crypto social media, with users rushing to apply for participation.
Flop says it is building for AI-native settlement
According to the project’s public materials, Flop Network is a protocol based on what it calls useful proof of inference and is designed for the AI agent economy. Its premise is that AI agents will eventually execute tasks, transact, and coordinate at scale, creating an economy that could exceed human economic activity in size.
For that to happen, the project argues, agents need a native medium of settlement that can pay for compute, memory storage and retrieval, and transactions with other agents. FLOP is presented as the token for that role.
The project’s messaging includes “no presale,” “no VC,” “100% fair launch,” and a “Q4 airdrop.” Combined with Hayes’s direct involvement, that was enough to turn Flop into a fresh talking point in the market.
It has outlined four types of participants
Flop Network says it wants to move away from mining models built around hash computation alone. Instead, miners would supply real AI inference power, run useful inference tasks, and receive token rewards for doing so. In Flop’s framing, network resources are meant to produce compute that can directly serve the AI economy rather than simply burn power in a race for block production.
Based on the information released so far, the network has four participant groups:
- Miners: They provide compute resources, run AI inference tasks, and receive FLOP block rewards plus task payments. Applicants are asked to submit personal information, hardware preferences, and intended operating scale.
- Validators: They verify miners’ inference results, handle decentralized storage and management of AI agent memory, and can earn FLOP through validation work, inference payments, and memory storage and retrieval services. Applicants are asked for personal information, validator configuration preferences, and the number of nodes they plan to run.
- AI agents: They act as demand-side users of the network, using FLOP to buy compute, store and retrieve memory, and conduct agent-to-agent economic activity.
- KOLs and creators: They are positioned as promoters and community guides. Applicants can submit details on their audience, network reach, distribution channels, and prior community project experience, and may earn FLOP based on the activity they bring into the community.
Even so, Flop Labs has disclosed only a limited amount. At this stage, the public material consists mainly of a landing page, three application forms, and one project overview image. There is still no white paper, no tokenomics, no contract address, and no clear statement on the underlying chain type or the project’s technical implementation.
That leaves several open questions around the protocol architecture, the token model, and whether the project can ultimately be deployed as described.
The roadmap puts the airdrop before the genesis block
Under the roadmap released by Flop Network, a large-scale airdrop is scheduled for the fourth quarter of 2026, while the genesis block is planned for the first quarter of 2027.
That timing means the token distribution would take place roughly one quarter before the network’s first block is produced. Participants would be receiving an expectation of token entitlement at that stage, rather than an asset already circulating on a live network.
Hayes’s past AI bubble warnings are now part of the story
Hayes’s decision to personally lead the project has generated as much attention as Flop itself, and also skepticism. One reason is that his previous public stance on AI was not especially bullish.
The report said Hayes repeatedly warned in 2026 that the rapid expansion of AI infrastructure was, in his view, a debt-driven credit bubble. He argued that the setup looked less like the internet valuation bubble of 2000 and more like the housing bubble of 2008.
In Hayes’s telling, data center construction has been financed largely with heavy borrowing. GPU loans often run for five to six years, while the effective life of the chips may be only two to three years. If demand growth slows, or if cheaper Chinese models compress inference prices, the cash-flow assumptions behind that debt could break down.
He had also estimated that AI-related debt grew to about $1.5 trillion from late 2022 to mid-2026, absorbing nearly all of the increase in U.S. M2 over the same period. In that view, the process drained liquidity that might otherwise have flowed into crypto and also planted the seeds for future overcapacity in compute.
Hayes expected capital expenditure growth could start slowing in the second half of 2027, with credit stress becoming more visible around 2028. He said governments and central banks might then be forced to print money on a large scale to stabilize the system, which he believed would ultimately benefit Bitcoin.
He says his criticism was about AI infrastructure finance, not agents
Hayes’s response to the apparent shift is that he was not criticizing AI itself. His target, he said, was the financial bubble built around AI infrastructure.
He argued that the excess sits in the debt issued to build data centers and in the richly valued shares of some large technology companies and frontier AI labs that have yet to achieve profitability, rather than in the agent economy itself. Hayes said he believes in the agent economy “100%,” adding that overbuilt compute capacity financed with borrowed money actually supports the case he is making for Flop Labs.
Past criticism of his token calls resurfaced as well
Another source of skepticism comes from Hayes’s own trading commentary. In June this year, on-chain investigator ZachXBT publicly questioned Hayes’s trading behavior on social media. ZachXBT reviewed Hayes’s earlier bullish comments on tokens including NEAR, HYPE, ZEC, and WLD, then said Hayes sold those positions a short time later, raising the question of whether followers could have become exit liquidity.
Hayes replied that he had simply made the correct trading decisions based on predetermined targets, that he had never instructed anyone to buy or sell a particular crypto asset, and that he does not provide investment advisory services. ZachXBT rejected that defense and said the pattern of publicly praising tokens and then quickly exiting the positions was hard to reconcile.
Still, Hayes remains one of the better-known figures in crypto, and that visibility is now one of Flop Labs’ biggest assets. Whether the attention he brings can turn into real network demand will depend on what the project reveals next and whether it can move beyond early marketing into execution.