Arthur Hayes Called AI a Bubble, Now He’s Launching an AI Project

Arthur Hayes has said that he is coming out of retirement to lead Flop Labs, a new project centered on a token called $FLOP that aims to work as a kind of currency for AI agents.

The announcement doubled as a defense of the move: Hayes, who has spent months warning that AI investment is a bubble, argued the excess sits in the debt piling up to build data centers, not in the underlying technology his new venture is built on.

A Fair Launch and a Compute-Backed Token

Hayes described $FLOP as “food for your AI agent” and said the token would launch without a presale or venture capital funding. “100% fair launch,” he wrote, adding that he expects a “massive airdrop in Q4” followed by a genesis block in the first quarter of 2027.

Flop Labs’ announcement describes the network as a proof-of-useful-inference protocol. Its stated goal is to give AI agents a native currency for buying computing power and storing memories. The project uses floating-point operations, or FLOPs, as the basis for its economic model. Miners would provide computing power and receive $FLOP through block rewards and inference payments.

Validators would verify that miners delivered the computing work requested by agents. They would also store agent memories and receive $FLOP through block rewards and inference payments.

On their part, AI agents would spend the token on computing and memory services, while community partners could receive $FLOP based on network activity.

Flop Labs stresses that the network has not launched and remains under development. Its current design can change, and the project makes no guarantees about receiving tokens or making money.

The AI Bubble Argument

Before announcing Flop Labs, Hayes had compared the AI buildout to the 2008 housing crisis, arguing that lenders, private credit funds, and governments are financing data centers on the assumption that demand keeps climbing without limit.

He expects AI capital spending to slow in 2027 before contracting, which he said would force governments into bailouts larger than those that followed the 2008 financial crisis, sending new money into crypto markets and potentially pushing Bitcoin toward $1 million.

When asked why he would build an AI project while calling AI a bubble, Hayes said the excess lies in debt used to fund data centers and in the shares of hyperscalers and frontier labs that are not yet profitable, not in agentic technology itself.

“Price is what you pay, value is what you get,” he wrote, adding that the compute overcapacity built on borrowed money strengthens his case for Flop Labs.

Hayes’s bet on an agent-native currency also arrives against thin real-world usage elsewhere. Analyst Jamie Coutts reported on August 12 that daily settlement volume on x402, the Coinbase-built payment protocol for AI agents, is down 93% year-to-date, with the seven-day average around $41,800, well below the $800,000 to $1 million peaks seen in late 2025.

Coutts called the drop a “reality check” on claims that the agentic economy has arrived, though he expects volume to climb again in the fourth quarter.