Arthur Hayes On The Early Days Of BitMEX, Where Bitcoin Is Going, And Why We Are Over-Indexing On AI

Arthur Hayes says he found Bitcoin after leaving Citibank, studied its white paper and identified arbitrage opportunities. He began trading and investing in crypto in 2013, then founded BitMEX in Hong Kong in 2014 with technologists Ben Delo and Samuel Reed. He chose derivatives because spot trading was crowded and derivatives appeared less developed. Early markets had few safeguards: after Bitcoin rose from about $250 to $1,300 in four months in 2013, a collapse in the futures basis exposed an exchange’s margin and liquidation failures. Hayes says ICBit also failed to margin positions properly, but Cumberland bailed it out and traders recovered their money.

BitMEX initially offered quarterly inverse BTC-margined BTC/USD futures and struggled to attract liquidity. Hayes says Chinese competitors offered roughly 20x leverage, while BitMEX adapted socialized-loss margining concepts and used its technology to offer 100x leverage. The exchange became cash-flow positive in late 2015, as Bitcoin rose from roughly $130 to $600. In May 2016, BitMEX introduced the perpetual swap to remove expiry-date confusion and concentrate liquidity in one contract. Its first funding design failed during a rally, leaving the swap at a large premium to spot; a self-referencing funding mechanism, premium index and eight-hour lookback helped discourage trend-following positions and attract market makers. Hayes says other exchanges later copied the features. BitMEX built an asynchronous, tick-by-tick engine for the product using KDB, a language Ben Delo knew from bank trading systems. Hayes says the aim was easier trading and fewer support requests, not anticipating how widely perpetuals would spread.

Hayes says he is relieved to have left exchange operations, where mistakes involving customer assets could cost billions. BitMEX had about 250 employees at its peak and left exchange operations without losing customer funds. He describes trading as an intellectual exercise and a game whose measurable objective is profit, as well as a way to study how people allocate time and resources. More than 90% of his net worth is in crypto, alongside real estate, gold and stocks; he says he could manage if crypto went to zero. His stated position is long Bitcoin.

Hayes sees AI as potentially useful but argues that AI companies are poor investments at current prices because customers are not paying the true cost of the service. Citing an Anthropic statement from 2025, he says it cost $3 to generate $1 of revenue, and questions whether users would pay enough to support profits at the scale investors expect. He says compute is getting cheaper while data centers are increasingly costly to build because of labor and material expenses and competition for energy. In his estimate, Anthropic and OpenAI need roughly $50–100 billion in debt sales each year, and he puts the probability at about 75% that AI companies cannot afford their compute commitments without government bailouts. He compares the investment cycle with the dot-com era: the technology may prove useful, but investors may be paying the wrong price now.

Hayes expects governments to treat AI companies as too big to fail, which he thinks could benefit crypto, though the interviewer raises the possibility that a bailout could provoke a sharp political shift and draw crypto into a backlash. Hayes argues that governments across political systems tend to avoid assigning economic losses, preferring money printing or preserving the status quo to a sharp downturn. He says China recognizes distortions from property and credit expansion but has not agreed on who should bear the losses, and has continued printing and exporting; he describes its 2009–2018 credit expansion as probably the century’s largest and expects the deferred losses to come due eventually, without knowing when. He also points to pandemic stimulus backed by both parties, characterizing Trump’s spending as probably the largest money-printing exercise since FDR, with Biden continuing it. The interviewer sees the prospect of further US and Chinese money printing as supportive of Bitcoin. Hayes argues that $1,000 payments would not solve social disengagement and could raise prices for scarce goods.

Hayes came out of retirement to develop Flop Network, which aims to let AI agents pay for inference using miners’ GPUs; miners would earn Flop through block rewards and transaction fees, tying the currency to compute. He says billions of agents already exist, though their number is unknown, compared with a few hundred million people supporting Bitcoin. Agents are not yet doing much commerce, but their activity is growing quickly, and Hayes believes a network serving them could become extremely valuable and potentially create a cryptocurrency larger than Bitcoin. He expects more data-center capacity to lower compute costs, enabling more agents and more complex tasks; he says his team uses agents for useful work when people with relevant experience direct them responsibly. He also says he likes Zcash’s privacy narrative and hopes its blockchain issues are fixed, with formal verification expected to enter the protocol in Q4 of that year.